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"
26 #include "hw/pci/msi.h"
27 #include "hw/pci/msix.h"
28 #include "hw/s390x/adapter.h"
29 #include "exec/gdbstub.h"
30 #include "sysemu/kvm_int.h"
31 #include "sysemu/runstate.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/sysemu.h"
34 #include "qemu/bswap.h"
35 #include "exec/memory.h"
36 #include "exec/ram_addr.h"
37 #include "exec/address-spaces.h"
38 #include "qemu/event_notifier.h"
39 #include "qemu/main-loop.h"
42 #include "sysemu/sev.h"
43 #include "sysemu/balloon.h"
44 #include "qapi/visitor.h"
45 #include "qapi/qapi-types-common.h"
46 #include "qapi/qapi-visit-common.h"
48 #include "hw/boards.h"
50 /* This check must be after config-host.h is included */
52 #include <sys/eventfd.h>
55 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
56 * need to use the real host PAGE_SIZE, as that's what KVM will use.
58 #define PAGE_SIZE qemu_real_host_page_size
63 #define DPRINTF(fmt, ...) \
64 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
66 #define DPRINTF(fmt, ...) \
70 #define KVM_MSI_HASHTAB_SIZE 256
72 struct KVMParkedVcpu
{
73 unsigned long vcpu_id
;
75 QLIST_ENTRY(KVMParkedVcpu
) node
;
80 AccelState parent_obj
;
87 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
88 bool coalesced_flush_in_progress
;
90 int robust_singlestep
;
92 #ifdef KVM_CAP_SET_GUEST_DEBUG
93 QTAILQ_HEAD(, kvm_sw_breakpoint
) kvm_sw_breakpoints
;
95 int max_nested_state_len
;
99 bool kernel_irqchip_allowed
;
100 bool kernel_irqchip_required
;
101 OnOffAuto kernel_irqchip_split
;
103 bool manual_dirty_log_protect
;
104 /* The man page (and posix) say ioctl numbers are signed int, but
105 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
106 * unsigned, and treating them as signed here can break things */
107 unsigned irq_set_ioctl
;
108 unsigned int sigmask_len
;
110 #ifdef KVM_CAP_IRQ_ROUTING
111 struct kvm_irq_routing
*irq_routes
;
112 int nr_allocated_irq_routes
;
113 unsigned long *used_gsi_bitmap
;
114 unsigned int gsi_count
;
115 QTAILQ_HEAD(, KVMMSIRoute
) msi_hashtab
[KVM_MSI_HASHTAB_SIZE
];
117 KVMMemoryListener memory_listener
;
118 QLIST_HEAD(, KVMParkedVcpu
) kvm_parked_vcpus
;
120 /* memory encryption */
121 void *memcrypt_handle
;
122 int (*memcrypt_encrypt_data
)(void *handle
, uint8_t *ptr
, uint64_t len
);
124 /* For "info mtree -f" to tell if an MR is registered in KVM */
127 KVMMemoryListener
*ml
;
133 bool kvm_kernel_irqchip
;
134 bool kvm_split_irqchip
;
135 bool kvm_async_interrupts_allowed
;
136 bool kvm_halt_in_kernel_allowed
;
137 bool kvm_eventfds_allowed
;
138 bool kvm_irqfds_allowed
;
139 bool kvm_resamplefds_allowed
;
140 bool kvm_msi_via_irqfd_allowed
;
141 bool kvm_gsi_routing_allowed
;
142 bool kvm_gsi_direct_mapping
;
144 bool kvm_readonly_mem_allowed
;
145 bool kvm_vm_attributes_allowed
;
146 bool kvm_direct_msi_allowed
;
147 bool kvm_ioeventfd_any_length_allowed
;
148 bool kvm_msi_use_devid
;
149 static bool kvm_immediate_exit
;
150 static hwaddr kvm_max_slot_size
= ~0;
152 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
153 KVM_CAP_INFO(USER_MEMORY
),
154 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
155 KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS
),
159 static NotifierList kvm_irqchip_change_notifiers
=
160 NOTIFIER_LIST_INITIALIZER(kvm_irqchip_change_notifiers
);
162 #define kvm_slots_lock(kml) qemu_mutex_lock(&(kml)->slots_lock)
163 #define kvm_slots_unlock(kml) qemu_mutex_unlock(&(kml)->slots_lock)
165 int kvm_get_max_memslots(void)
167 KVMState
*s
= KVM_STATE(current_accel());
172 bool kvm_memcrypt_enabled(void)
174 if (kvm_state
&& kvm_state
->memcrypt_handle
) {
181 int kvm_memcrypt_encrypt_data(uint8_t *ptr
, uint64_t len
)
183 if (kvm_state
->memcrypt_handle
&&
184 kvm_state
->memcrypt_encrypt_data
) {
185 return kvm_state
->memcrypt_encrypt_data(kvm_state
->memcrypt_handle
,
192 /* Called with KVMMemoryListener.slots_lock held */
193 static KVMSlot
*kvm_get_free_slot(KVMMemoryListener
*kml
)
195 KVMState
*s
= kvm_state
;
198 for (i
= 0; i
< s
->nr_slots
; i
++) {
199 if (kml
->slots
[i
].memory_size
== 0) {
200 return &kml
->slots
[i
];
207 bool kvm_has_free_slot(MachineState
*ms
)
209 KVMState
*s
= KVM_STATE(ms
->accelerator
);
211 KVMMemoryListener
*kml
= &s
->memory_listener
;
214 result
= !!kvm_get_free_slot(kml
);
215 kvm_slots_unlock(kml
);
220 /* Called with KVMMemoryListener.slots_lock held */
221 static KVMSlot
*kvm_alloc_slot(KVMMemoryListener
*kml
)
223 KVMSlot
*slot
= kvm_get_free_slot(kml
);
229 fprintf(stderr
, "%s: no free slot available\n", __func__
);
233 static KVMSlot
*kvm_lookup_matching_slot(KVMMemoryListener
*kml
,
237 KVMState
*s
= kvm_state
;
240 for (i
= 0; i
< s
->nr_slots
; i
++) {
241 KVMSlot
*mem
= &kml
->slots
[i
];
243 if (start_addr
== mem
->start_addr
&& size
== mem
->memory_size
) {
252 * Calculate and align the start address and the size of the section.
253 * Return the size. If the size is 0, the aligned section is empty.
255 static hwaddr
kvm_align_section(MemoryRegionSection
*section
,
258 hwaddr size
= int128_get64(section
->size
);
259 hwaddr delta
, aligned
;
261 /* kvm works in page size chunks, but the function may be called
262 with sub-page size and unaligned start address. Pad the start
263 address to next and truncate size to previous page boundary. */
264 aligned
= ROUND_UP(section
->offset_within_address_space
,
265 qemu_real_host_page_size
);
266 delta
= aligned
- section
->offset_within_address_space
;
272 return (size
- delta
) & qemu_real_host_page_mask
;
275 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
278 KVMMemoryListener
*kml
= &s
->memory_listener
;
282 for (i
= 0; i
< s
->nr_slots
; i
++) {
283 KVMSlot
*mem
= &kml
->slots
[i
];
285 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
286 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
291 kvm_slots_unlock(kml
);
296 static int kvm_set_user_memory_region(KVMMemoryListener
*kml
, KVMSlot
*slot
, bool new)
298 KVMState
*s
= kvm_state
;
299 struct kvm_userspace_memory_region mem
;
302 mem
.slot
= slot
->slot
| (kml
->as_id
<< 16);
303 mem
.guest_phys_addr
= slot
->start_addr
;
304 mem
.userspace_addr
= (unsigned long)slot
->ram
;
305 mem
.flags
= slot
->flags
;
307 if (slot
->memory_size
&& !new && (mem
.flags
^ slot
->old_flags
) & KVM_MEM_READONLY
) {
308 /* Set the slot size to 0 before setting the slot to the desired
309 * value. This is needed based on KVM commit 75d61fbc. */
311 ret
= kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
316 mem
.memory_size
= slot
->memory_size
;
317 ret
= kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
318 slot
->old_flags
= mem
.flags
;
320 trace_kvm_set_user_memory(mem
.slot
, mem
.flags
, mem
.guest_phys_addr
,
321 mem
.memory_size
, mem
.userspace_addr
, ret
);
323 error_report("%s: KVM_SET_USER_MEMORY_REGION failed, slot=%d,"
324 " start=0x%" PRIx64
", size=0x%" PRIx64
": %s",
325 __func__
, mem
.slot
, slot
->start_addr
,
326 (uint64_t)mem
.memory_size
, strerror(errno
));
331 int kvm_destroy_vcpu(CPUState
*cpu
)
333 KVMState
*s
= kvm_state
;
335 struct KVMParkedVcpu
*vcpu
= NULL
;
338 DPRINTF("kvm_destroy_vcpu\n");
340 ret
= kvm_arch_destroy_vcpu(cpu
);
345 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
348 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
352 ret
= munmap(cpu
->kvm_run
, mmap_size
);
357 vcpu
= g_malloc0(sizeof(*vcpu
));
358 vcpu
->vcpu_id
= kvm_arch_vcpu_id(cpu
);
359 vcpu
->kvm_fd
= cpu
->kvm_fd
;
360 QLIST_INSERT_HEAD(&kvm_state
->kvm_parked_vcpus
, vcpu
, node
);
365 static int kvm_get_vcpu(KVMState
*s
, unsigned long vcpu_id
)
367 struct KVMParkedVcpu
*cpu
;
369 QLIST_FOREACH(cpu
, &s
->kvm_parked_vcpus
, node
) {
370 if (cpu
->vcpu_id
== vcpu_id
) {
373 QLIST_REMOVE(cpu
, node
);
374 kvm_fd
= cpu
->kvm_fd
;
380 return kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, (void *)vcpu_id
);
383 int kvm_init_vcpu(CPUState
*cpu
)
385 KVMState
*s
= kvm_state
;
389 DPRINTF("kvm_init_vcpu\n");
391 ret
= kvm_get_vcpu(s
, kvm_arch_vcpu_id(cpu
));
393 DPRINTF("kvm_create_vcpu failed\n");
399 cpu
->vcpu_dirty
= true;
401 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
404 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
408 cpu
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
410 if (cpu
->kvm_run
== MAP_FAILED
) {
412 DPRINTF("mmap'ing vcpu state failed\n");
416 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
417 s
->coalesced_mmio_ring
=
418 (void *)cpu
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
421 ret
= kvm_arch_init_vcpu(cpu
);
427 * dirty pages logging control
430 static int kvm_mem_flags(MemoryRegion
*mr
)
432 bool readonly
= mr
->readonly
|| memory_region_is_romd(mr
);
435 if (memory_region_get_dirty_log_mask(mr
) != 0) {
436 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
438 if (readonly
&& kvm_readonly_mem_allowed
) {
439 flags
|= KVM_MEM_READONLY
;
444 /* Called with KVMMemoryListener.slots_lock held */
445 static int kvm_slot_update_flags(KVMMemoryListener
*kml
, KVMSlot
*mem
,
448 mem
->flags
= kvm_mem_flags(mr
);
450 /* If nothing changed effectively, no need to issue ioctl */
451 if (mem
->flags
== mem
->old_flags
) {
455 return kvm_set_user_memory_region(kml
, mem
, false);
458 static int kvm_section_update_flags(KVMMemoryListener
*kml
,
459 MemoryRegionSection
*section
)
461 hwaddr start_addr
, size
, slot_size
;
465 size
= kvm_align_section(section
, &start_addr
);
472 while (size
&& !ret
) {
473 slot_size
= MIN(kvm_max_slot_size
, size
);
474 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
476 /* We don't have a slot if we want to trap every access. */
480 ret
= kvm_slot_update_flags(kml
, mem
, section
->mr
);
481 start_addr
+= slot_size
;
486 kvm_slots_unlock(kml
);
490 static void kvm_log_start(MemoryListener
*listener
,
491 MemoryRegionSection
*section
,
494 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
501 r
= kvm_section_update_flags(kml
, section
);
507 static void kvm_log_stop(MemoryListener
*listener
,
508 MemoryRegionSection
*section
,
511 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
518 r
= kvm_section_update_flags(kml
, section
);
524 /* get kvm's dirty pages bitmap and update qemu's */
525 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
526 unsigned long *bitmap
)
528 ram_addr_t start
= section
->offset_within_region
+
529 memory_region_get_ram_addr(section
->mr
);
530 ram_addr_t pages
= int128_get64(section
->size
) / qemu_real_host_page_size
;
532 cpu_physical_memory_set_dirty_lebitmap(bitmap
, start
, pages
);
536 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
538 /* Allocate the dirty bitmap for a slot */
539 static void kvm_memslot_init_dirty_bitmap(KVMSlot
*mem
)
542 * XXX bad kernel interface alert
543 * For dirty bitmap, kernel allocates array of size aligned to
544 * bits-per-long. But for case when the kernel is 64bits and
545 * the userspace is 32bits, userspace can't align to the same
546 * bits-per-long, since sizeof(long) is different between kernel
547 * and user space. This way, userspace will provide buffer which
548 * may be 4 bytes less than the kernel will use, resulting in
549 * userspace memory corruption (which is not detectable by valgrind
550 * too, in most cases).
551 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
552 * a hope that sizeof(long) won't become >8 any time soon.
554 hwaddr bitmap_size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
555 /*HOST_LONG_BITS*/ 64) / 8;
556 mem
->dirty_bmap
= g_malloc0(bitmap_size
);
560 * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space
562 * This function will first try to fetch dirty bitmap from the kernel,
563 * and then updates qemu's dirty bitmap.
565 * NOTE: caller must be with kml->slots_lock held.
567 * @kml: the KVM memory listener object
568 * @section: the memory section to sync the dirty bitmap with
570 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener
*kml
,
571 MemoryRegionSection
*section
)
573 KVMState
*s
= kvm_state
;
574 struct kvm_dirty_log d
= {};
576 hwaddr start_addr
, size
;
577 hwaddr slot_size
, slot_offset
= 0;
580 size
= kvm_align_section(section
, &start_addr
);
582 MemoryRegionSection subsection
= *section
;
584 slot_size
= MIN(kvm_max_slot_size
, size
);
585 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
587 /* We don't have a slot if we want to trap every access. */
591 if (!mem
->dirty_bmap
) {
592 /* Allocate on the first log_sync, once and for all */
593 kvm_memslot_init_dirty_bitmap(mem
);
596 d
.dirty_bitmap
= mem
->dirty_bmap
;
597 d
.slot
= mem
->slot
| (kml
->as_id
<< 16);
598 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
599 DPRINTF("ioctl failed %d\n", errno
);
604 subsection
.offset_within_region
+= slot_offset
;
605 subsection
.size
= int128_make64(slot_size
);
606 kvm_get_dirty_pages_log_range(&subsection
, d
.dirty_bitmap
);
608 slot_offset
+= slot_size
;
609 start_addr
+= slot_size
;
616 /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */
617 #define KVM_CLEAR_LOG_SHIFT 6
618 #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size << KVM_CLEAR_LOG_SHIFT)
619 #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN)
621 static int kvm_log_clear_one_slot(KVMSlot
*mem
, int as_id
, uint64_t start
,
624 KVMState
*s
= kvm_state
;
625 uint64_t end
, bmap_start
, start_delta
, bmap_npages
;
626 struct kvm_clear_dirty_log d
;
627 unsigned long *bmap_clear
= NULL
, psize
= qemu_real_host_page_size
;
631 * We need to extend either the start or the size or both to
632 * satisfy the KVM interface requirement. Firstly, do the start
633 * page alignment on 64 host pages
635 bmap_start
= start
& KVM_CLEAR_LOG_MASK
;
636 start_delta
= start
- bmap_start
;
640 * The kernel interface has restriction on the size too, that either:
642 * (1) the size is 64 host pages aligned (just like the start), or
643 * (2) the size fills up until the end of the KVM memslot.
645 bmap_npages
= DIV_ROUND_UP(size
+ start_delta
, KVM_CLEAR_LOG_ALIGN
)
646 << KVM_CLEAR_LOG_SHIFT
;
647 end
= mem
->memory_size
/ psize
;
648 if (bmap_npages
> end
- bmap_start
) {
649 bmap_npages
= end
- bmap_start
;
651 start_delta
/= psize
;
654 * Prepare the bitmap to clear dirty bits. Here we must guarantee
655 * that we won't clear any unknown dirty bits otherwise we might
656 * accidentally clear some set bits which are not yet synced from
657 * the kernel into QEMU's bitmap, then we'll lose track of the
658 * guest modifications upon those pages (which can directly lead
659 * to guest data loss or panic after migration).
661 * Layout of the KVMSlot.dirty_bmap:
663 * |<-------- bmap_npages -----------..>|
666 * |----------------|-------------|------------------|------------|
669 * start bmap_start (start) end
670 * of memslot of memslot
672 * [1] bmap_npages can be aligned to either 64 pages or the end of slot
675 assert(bmap_start
% BITS_PER_LONG
== 0);
676 /* We should never do log_clear before log_sync */
677 assert(mem
->dirty_bmap
);
679 /* Slow path - we need to manipulate a temp bitmap */
680 bmap_clear
= bitmap_new(bmap_npages
);
681 bitmap_copy_with_src_offset(bmap_clear
, mem
->dirty_bmap
,
682 bmap_start
, start_delta
+ size
/ psize
);
684 * We need to fill the holes at start because that was not
685 * specified by the caller and we extended the bitmap only for
688 bitmap_clear(bmap_clear
, 0, start_delta
);
689 d
.dirty_bitmap
= bmap_clear
;
691 /* Fast path - start address aligns well with BITS_PER_LONG */
692 d
.dirty_bitmap
= mem
->dirty_bmap
+ BIT_WORD(bmap_start
);
695 d
.first_page
= bmap_start
;
696 /* It should never overflow. If it happens, say something */
697 assert(bmap_npages
<= UINT32_MAX
);
698 d
.num_pages
= bmap_npages
;
699 d
.slot
= mem
->slot
| (as_id
<< 16);
701 if (kvm_vm_ioctl(s
, KVM_CLEAR_DIRTY_LOG
, &d
) == -1) {
703 error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, "
704 "start=0x%"PRIx64
", size=0x%"PRIx32
", errno=%d",
705 __func__
, d
.slot
, (uint64_t)d
.first_page
,
706 (uint32_t)d
.num_pages
, ret
);
709 trace_kvm_clear_dirty_log(d
.slot
, d
.first_page
, d
.num_pages
);
713 * After we have updated the remote dirty bitmap, we update the
714 * cached bitmap as well for the memslot, then if another user
715 * clears the same region we know we shouldn't clear it again on
716 * the remote otherwise it's data loss as well.
718 bitmap_clear(mem
->dirty_bmap
, bmap_start
+ start_delta
,
720 /* This handles the NULL case well */
727 * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range
729 * NOTE: this will be a no-op if we haven't enabled manual dirty log
730 * protection in the host kernel because in that case this operation
731 * will be done within log_sync().
733 * @kml: the kvm memory listener
734 * @section: the memory range to clear dirty bitmap
736 static int kvm_physical_log_clear(KVMMemoryListener
*kml
,
737 MemoryRegionSection
*section
)
739 KVMState
*s
= kvm_state
;
740 uint64_t start
, size
, offset
, count
;
744 if (!s
->manual_dirty_log_protect
) {
745 /* No need to do explicit clear */
749 start
= section
->offset_within_address_space
;
750 size
= int128_get64(section
->size
);
753 /* Nothing more we can do... */
759 for (i
= 0; i
< s
->nr_slots
; i
++) {
760 mem
= &kml
->slots
[i
];
761 /* Discard slots that are empty or do not overlap the section */
762 if (!mem
->memory_size
||
763 mem
->start_addr
> start
+ size
- 1 ||
764 start
> mem
->start_addr
+ mem
->memory_size
- 1) {
768 if (start
>= mem
->start_addr
) {
769 /* The slot starts before section or is aligned to it. */
770 offset
= start
- mem
->start_addr
;
771 count
= MIN(mem
->memory_size
- offset
, size
);
773 /* The slot starts after section. */
775 count
= MIN(mem
->memory_size
, size
- (mem
->start_addr
- start
));
777 ret
= kvm_log_clear_one_slot(mem
, kml
->as_id
, offset
, count
);
783 kvm_slots_unlock(kml
);
788 static void kvm_coalesce_mmio_region(MemoryListener
*listener
,
789 MemoryRegionSection
*secion
,
790 hwaddr start
, hwaddr size
)
792 KVMState
*s
= kvm_state
;
794 if (s
->coalesced_mmio
) {
795 struct kvm_coalesced_mmio_zone zone
;
801 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
805 static void kvm_uncoalesce_mmio_region(MemoryListener
*listener
,
806 MemoryRegionSection
*secion
,
807 hwaddr start
, hwaddr size
)
809 KVMState
*s
= kvm_state
;
811 if (s
->coalesced_mmio
) {
812 struct kvm_coalesced_mmio_zone zone
;
818 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
822 static void kvm_coalesce_pio_add(MemoryListener
*listener
,
823 MemoryRegionSection
*section
,
824 hwaddr start
, hwaddr size
)
826 KVMState
*s
= kvm_state
;
828 if (s
->coalesced_pio
) {
829 struct kvm_coalesced_mmio_zone zone
;
835 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
839 static void kvm_coalesce_pio_del(MemoryListener
*listener
,
840 MemoryRegionSection
*section
,
841 hwaddr start
, hwaddr size
)
843 KVMState
*s
= kvm_state
;
845 if (s
->coalesced_pio
) {
846 struct kvm_coalesced_mmio_zone zone
;
852 (void)kvm_vm_ioctl(s
, KVM_UNREGISTER_COALESCED_MMIO
, &zone
);
856 static MemoryListener kvm_coalesced_pio_listener
= {
857 .coalesced_io_add
= kvm_coalesce_pio_add
,
858 .coalesced_io_del
= kvm_coalesce_pio_del
,
861 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
865 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
873 int kvm_vm_check_extension(KVMState
*s
, unsigned int extension
)
877 ret
= kvm_vm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
879 /* VM wide version not implemented, use global one instead */
880 ret
= kvm_check_extension(s
, extension
);
886 static uint32_t adjust_ioeventfd_endianness(uint32_t val
, uint32_t size
)
888 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
889 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
890 * endianness, but the memory core hands them in target endianness.
891 * For example, PPC is always treated as big-endian even if running
892 * on KVM and on PPC64LE. Correct here.
906 static int kvm_set_ioeventfd_mmio(int fd
, hwaddr addr
, uint32_t val
,
907 bool assign
, uint32_t size
, bool datamatch
)
910 struct kvm_ioeventfd iofd
= {
911 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
918 trace_kvm_set_ioeventfd_mmio(fd
, (uint64_t)addr
, val
, assign
, size
,
920 if (!kvm_enabled()) {
925 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
928 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
931 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
940 static int kvm_set_ioeventfd_pio(int fd
, uint16_t addr
, uint16_t val
,
941 bool assign
, uint32_t size
, bool datamatch
)
943 struct kvm_ioeventfd kick
= {
944 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
946 .flags
= KVM_IOEVENTFD_FLAG_PIO
,
951 trace_kvm_set_ioeventfd_pio(fd
, addr
, val
, assign
, size
, datamatch
);
952 if (!kvm_enabled()) {
956 kick
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
959 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
961 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
969 static int kvm_check_many_ioeventfds(void)
971 /* Userspace can use ioeventfd for io notification. This requires a host
972 * that supports eventfd(2) and an I/O thread; since eventfd does not
973 * support SIGIO it cannot interrupt the vcpu.
975 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
976 * can avoid creating too many ioeventfds.
978 #if defined(CONFIG_EVENTFD)
981 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
982 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
983 if (ioeventfds
[i
] < 0) {
986 ret
= kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, true, 2, true);
988 close(ioeventfds
[i
]);
993 /* Decide whether many devices are supported or not */
994 ret
= i
== ARRAY_SIZE(ioeventfds
);
997 kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, false, 2, true);
998 close(ioeventfds
[i
]);
1006 static const KVMCapabilityInfo
*
1007 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
1009 while (list
->name
) {
1010 if (!kvm_check_extension(s
, list
->value
)) {
1018 void kvm_set_max_memslot_size(hwaddr max_slot_size
)
1021 ROUND_UP(max_slot_size
, qemu_real_host_page_size
) == max_slot_size
1023 kvm_max_slot_size
= max_slot_size
;
1026 static void kvm_set_phys_mem(KVMMemoryListener
*kml
,
1027 MemoryRegionSection
*section
, bool add
)
1031 MemoryRegion
*mr
= section
->mr
;
1032 bool writeable
= !mr
->readonly
&& !mr
->rom_device
;
1033 hwaddr start_addr
, size
, slot_size
;
1036 if (!memory_region_is_ram(mr
)) {
1037 if (writeable
|| !kvm_readonly_mem_allowed
) {
1039 } else if (!mr
->romd_mode
) {
1040 /* If the memory device is not in romd_mode, then we actually want
1041 * to remove the kvm memory slot so all accesses will trap. */
1046 size
= kvm_align_section(section
, &start_addr
);
1051 /* use aligned delta to align the ram address */
1052 ram
= memory_region_get_ram_ptr(mr
) + section
->offset_within_region
+
1053 (start_addr
- section
->offset_within_address_space
);
1055 kvm_slots_lock(kml
);
1059 slot_size
= MIN(kvm_max_slot_size
, size
);
1060 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
1064 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
1065 kvm_physical_sync_dirty_bitmap(kml
, section
);
1068 /* unregister the slot */
1069 g_free(mem
->dirty_bmap
);
1070 mem
->dirty_bmap
= NULL
;
1071 mem
->memory_size
= 0;
1073 err
= kvm_set_user_memory_region(kml
, mem
, false);
1075 fprintf(stderr
, "%s: error unregistering slot: %s\n",
1076 __func__
, strerror(-err
));
1079 start_addr
+= slot_size
;
1085 /* register the new slot */
1087 slot_size
= MIN(kvm_max_slot_size
, size
);
1088 mem
= kvm_alloc_slot(kml
);
1089 mem
->memory_size
= slot_size
;
1090 mem
->start_addr
= start_addr
;
1092 mem
->flags
= kvm_mem_flags(mr
);
1094 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
1096 * Reallocate the bmap; it means it doesn't disappear in
1097 * middle of a migrate.
1099 kvm_memslot_init_dirty_bitmap(mem
);
1101 err
= kvm_set_user_memory_region(kml
, mem
, true);
1103 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
1107 start_addr
+= slot_size
;
1113 kvm_slots_unlock(kml
);
1116 static void kvm_region_add(MemoryListener
*listener
,
1117 MemoryRegionSection
*section
)
1119 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1121 memory_region_ref(section
->mr
);
1122 kvm_set_phys_mem(kml
, section
, true);
1125 static void kvm_region_del(MemoryListener
*listener
,
1126 MemoryRegionSection
*section
)
1128 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1130 kvm_set_phys_mem(kml
, section
, false);
1131 memory_region_unref(section
->mr
);
1134 static void kvm_log_sync(MemoryListener
*listener
,
1135 MemoryRegionSection
*section
)
1137 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1140 kvm_slots_lock(kml
);
1141 r
= kvm_physical_sync_dirty_bitmap(kml
, section
);
1142 kvm_slots_unlock(kml
);
1148 static void kvm_log_clear(MemoryListener
*listener
,
1149 MemoryRegionSection
*section
)
1151 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1154 r
= kvm_physical_log_clear(kml
, section
);
1156 error_report_once("%s: kvm log clear failed: mr=%s "
1157 "offset=%"HWADDR_PRIx
" size=%"PRIx64
, __func__
,
1158 section
->mr
->name
, section
->offset_within_region
,
1159 int128_get64(section
->size
));
1164 static void kvm_mem_ioeventfd_add(MemoryListener
*listener
,
1165 MemoryRegionSection
*section
,
1166 bool match_data
, uint64_t data
,
1169 int fd
= event_notifier_get_fd(e
);
1172 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1173 data
, true, int128_get64(section
->size
),
1176 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1177 __func__
, strerror(-r
), -r
);
1182 static void kvm_mem_ioeventfd_del(MemoryListener
*listener
,
1183 MemoryRegionSection
*section
,
1184 bool match_data
, uint64_t data
,
1187 int fd
= event_notifier_get_fd(e
);
1190 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1191 data
, false, int128_get64(section
->size
),
1194 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1195 __func__
, strerror(-r
), -r
);
1200 static void kvm_io_ioeventfd_add(MemoryListener
*listener
,
1201 MemoryRegionSection
*section
,
1202 bool match_data
, uint64_t data
,
1205 int fd
= event_notifier_get_fd(e
);
1208 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1209 data
, true, int128_get64(section
->size
),
1212 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1213 __func__
, strerror(-r
), -r
);
1218 static void kvm_io_ioeventfd_del(MemoryListener
*listener
,
1219 MemoryRegionSection
*section
,
1220 bool match_data
, uint64_t data
,
1224 int fd
= event_notifier_get_fd(e
);
1227 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1228 data
, false, int128_get64(section
->size
),
1231 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1232 __func__
, strerror(-r
), -r
);
1237 void kvm_memory_listener_register(KVMState
*s
, KVMMemoryListener
*kml
,
1238 AddressSpace
*as
, int as_id
)
1242 qemu_mutex_init(&kml
->slots_lock
);
1243 kml
->slots
= g_malloc0(s
->nr_slots
* sizeof(KVMSlot
));
1246 for (i
= 0; i
< s
->nr_slots
; i
++) {
1247 kml
->slots
[i
].slot
= i
;
1250 kml
->listener
.region_add
= kvm_region_add
;
1251 kml
->listener
.region_del
= kvm_region_del
;
1252 kml
->listener
.log_start
= kvm_log_start
;
1253 kml
->listener
.log_stop
= kvm_log_stop
;
1254 kml
->listener
.log_sync
= kvm_log_sync
;
1255 kml
->listener
.log_clear
= kvm_log_clear
;
1256 kml
->listener
.priority
= 10;
1258 memory_listener_register(&kml
->listener
, as
);
1260 for (i
= 0; i
< s
->nr_as
; ++i
) {
1269 static MemoryListener kvm_io_listener
= {
1270 .eventfd_add
= kvm_io_ioeventfd_add
,
1271 .eventfd_del
= kvm_io_ioeventfd_del
,
1275 int kvm_set_irq(KVMState
*s
, int irq
, int level
)
1277 struct kvm_irq_level event
;
1280 assert(kvm_async_interrupts_enabled());
1282 event
.level
= level
;
1284 ret
= kvm_vm_ioctl(s
, s
->irq_set_ioctl
, &event
);
1286 perror("kvm_set_irq");
1290 return (s
->irq_set_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
1293 #ifdef KVM_CAP_IRQ_ROUTING
1294 typedef struct KVMMSIRoute
{
1295 struct kvm_irq_routing_entry kroute
;
1296 QTAILQ_ENTRY(KVMMSIRoute
) entry
;
1299 static void set_gsi(KVMState
*s
, unsigned int gsi
)
1301 set_bit(gsi
, s
->used_gsi_bitmap
);
1304 static void clear_gsi(KVMState
*s
, unsigned int gsi
)
1306 clear_bit(gsi
, s
->used_gsi_bitmap
);
1309 void kvm_init_irq_routing(KVMState
*s
)
1313 gsi_count
= kvm_check_extension(s
, KVM_CAP_IRQ_ROUTING
) - 1;
1314 if (gsi_count
> 0) {
1315 /* Round up so we can search ints using ffs */
1316 s
->used_gsi_bitmap
= bitmap_new(gsi_count
);
1317 s
->gsi_count
= gsi_count
;
1320 s
->irq_routes
= g_malloc0(sizeof(*s
->irq_routes
));
1321 s
->nr_allocated_irq_routes
= 0;
1323 if (!kvm_direct_msi_allowed
) {
1324 for (i
= 0; i
< KVM_MSI_HASHTAB_SIZE
; i
++) {
1325 QTAILQ_INIT(&s
->msi_hashtab
[i
]);
1329 kvm_arch_init_irq_routing(s
);
1332 void kvm_irqchip_commit_routes(KVMState
*s
)
1336 if (kvm_gsi_direct_mapping()) {
1340 if (!kvm_gsi_routing_enabled()) {
1344 s
->irq_routes
->flags
= 0;
1345 trace_kvm_irqchip_commit_routes();
1346 ret
= kvm_vm_ioctl(s
, KVM_SET_GSI_ROUTING
, s
->irq_routes
);
1350 static void kvm_add_routing_entry(KVMState
*s
,
1351 struct kvm_irq_routing_entry
*entry
)
1353 struct kvm_irq_routing_entry
*new;
1356 if (s
->irq_routes
->nr
== s
->nr_allocated_irq_routes
) {
1357 n
= s
->nr_allocated_irq_routes
* 2;
1361 size
= sizeof(struct kvm_irq_routing
);
1362 size
+= n
* sizeof(*new);
1363 s
->irq_routes
= g_realloc(s
->irq_routes
, size
);
1364 s
->nr_allocated_irq_routes
= n
;
1366 n
= s
->irq_routes
->nr
++;
1367 new = &s
->irq_routes
->entries
[n
];
1371 set_gsi(s
, entry
->gsi
);
1374 static int kvm_update_routing_entry(KVMState
*s
,
1375 struct kvm_irq_routing_entry
*new_entry
)
1377 struct kvm_irq_routing_entry
*entry
;
1380 for (n
= 0; n
< s
->irq_routes
->nr
; n
++) {
1381 entry
= &s
->irq_routes
->entries
[n
];
1382 if (entry
->gsi
!= new_entry
->gsi
) {
1386 if(!memcmp(entry
, new_entry
, sizeof *entry
)) {
1390 *entry
= *new_entry
;
1398 void kvm_irqchip_add_irq_route(KVMState
*s
, int irq
, int irqchip
, int pin
)
1400 struct kvm_irq_routing_entry e
= {};
1402 assert(pin
< s
->gsi_count
);
1405 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
1407 e
.u
.irqchip
.irqchip
= irqchip
;
1408 e
.u
.irqchip
.pin
= pin
;
1409 kvm_add_routing_entry(s
, &e
);
1412 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1414 struct kvm_irq_routing_entry
*e
;
1417 if (kvm_gsi_direct_mapping()) {
1421 for (i
= 0; i
< s
->irq_routes
->nr
; i
++) {
1422 e
= &s
->irq_routes
->entries
[i
];
1423 if (e
->gsi
== virq
) {
1424 s
->irq_routes
->nr
--;
1425 *e
= s
->irq_routes
->entries
[s
->irq_routes
->nr
];
1429 kvm_arch_release_virq_post(virq
);
1430 trace_kvm_irqchip_release_virq(virq
);
1433 void kvm_irqchip_add_change_notifier(Notifier
*n
)
1435 notifier_list_add(&kvm_irqchip_change_notifiers
, n
);
1438 void kvm_irqchip_remove_change_notifier(Notifier
*n
)
1443 void kvm_irqchip_change_notify(void)
1445 notifier_list_notify(&kvm_irqchip_change_notifiers
, NULL
);
1448 static unsigned int kvm_hash_msi(uint32_t data
)
1450 /* This is optimized for IA32 MSI layout. However, no other arch shall
1451 * repeat the mistake of not providing a direct MSI injection API. */
1455 static void kvm_flush_dynamic_msi_routes(KVMState
*s
)
1457 KVMMSIRoute
*route
, *next
;
1460 for (hash
= 0; hash
< KVM_MSI_HASHTAB_SIZE
; hash
++) {
1461 QTAILQ_FOREACH_SAFE(route
, &s
->msi_hashtab
[hash
], entry
, next
) {
1462 kvm_irqchip_release_virq(s
, route
->kroute
.gsi
);
1463 QTAILQ_REMOVE(&s
->msi_hashtab
[hash
], route
, entry
);
1469 static int kvm_irqchip_get_virq(KVMState
*s
)
1474 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1475 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1476 * number can succeed even though a new route entry cannot be added.
1477 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1479 if (!kvm_direct_msi_allowed
&& s
->irq_routes
->nr
== s
->gsi_count
) {
1480 kvm_flush_dynamic_msi_routes(s
);
1483 /* Return the lowest unused GSI in the bitmap */
1484 next_virq
= find_first_zero_bit(s
->used_gsi_bitmap
, s
->gsi_count
);
1485 if (next_virq
>= s
->gsi_count
) {
1492 static KVMMSIRoute
*kvm_lookup_msi_route(KVMState
*s
, MSIMessage msg
)
1494 unsigned int hash
= kvm_hash_msi(msg
.data
);
1497 QTAILQ_FOREACH(route
, &s
->msi_hashtab
[hash
], entry
) {
1498 if (route
->kroute
.u
.msi
.address_lo
== (uint32_t)msg
.address
&&
1499 route
->kroute
.u
.msi
.address_hi
== (msg
.address
>> 32) &&
1500 route
->kroute
.u
.msi
.data
== le32_to_cpu(msg
.data
)) {
1507 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1512 if (kvm_direct_msi_allowed
) {
1513 msi
.address_lo
= (uint32_t)msg
.address
;
1514 msi
.address_hi
= msg
.address
>> 32;
1515 msi
.data
= le32_to_cpu(msg
.data
);
1517 memset(msi
.pad
, 0, sizeof(msi
.pad
));
1519 return kvm_vm_ioctl(s
, KVM_SIGNAL_MSI
, &msi
);
1522 route
= kvm_lookup_msi_route(s
, msg
);
1526 virq
= kvm_irqchip_get_virq(s
);
1531 route
= g_malloc0(sizeof(KVMMSIRoute
));
1532 route
->kroute
.gsi
= virq
;
1533 route
->kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1534 route
->kroute
.flags
= 0;
1535 route
->kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1536 route
->kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1537 route
->kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1539 kvm_add_routing_entry(s
, &route
->kroute
);
1540 kvm_irqchip_commit_routes(s
);
1542 QTAILQ_INSERT_TAIL(&s
->msi_hashtab
[kvm_hash_msi(msg
.data
)], route
,
1546 assert(route
->kroute
.type
== KVM_IRQ_ROUTING_MSI
);
1548 return kvm_set_irq(s
, route
->kroute
.gsi
, 1);
1551 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1553 struct kvm_irq_routing_entry kroute
= {};
1555 MSIMessage msg
= {0, 0};
1557 if (pci_available
&& dev
) {
1558 msg
= pci_get_msi_message(dev
, vector
);
1561 if (kvm_gsi_direct_mapping()) {
1562 return kvm_arch_msi_data_to_gsi(msg
.data
);
1565 if (!kvm_gsi_routing_enabled()) {
1569 virq
= kvm_irqchip_get_virq(s
);
1575 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1577 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1578 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1579 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1580 if (pci_available
&& kvm_msi_devid_required()) {
1581 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1582 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1584 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1585 kvm_irqchip_release_virq(s
, virq
);
1589 trace_kvm_irqchip_add_msi_route(dev
? dev
->name
: (char *)"N/A",
1592 kvm_add_routing_entry(s
, &kroute
);
1593 kvm_arch_add_msi_route_post(&kroute
, vector
, dev
);
1594 kvm_irqchip_commit_routes(s
);
1599 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
,
1602 struct kvm_irq_routing_entry kroute
= {};
1604 if (kvm_gsi_direct_mapping()) {
1608 if (!kvm_irqchip_in_kernel()) {
1613 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1615 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1616 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1617 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1618 if (pci_available
&& kvm_msi_devid_required()) {
1619 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1620 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1622 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1626 trace_kvm_irqchip_update_msi_route(virq
);
1628 return kvm_update_routing_entry(s
, &kroute
);
1631 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int rfd
, int virq
,
1634 struct kvm_irqfd irqfd
= {
1637 .flags
= assign
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1641 irqfd
.flags
|= KVM_IRQFD_FLAG_RESAMPLE
;
1642 irqfd
.resamplefd
= rfd
;
1645 if (!kvm_irqfds_enabled()) {
1649 return kvm_vm_ioctl(s
, KVM_IRQFD
, &irqfd
);
1652 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1654 struct kvm_irq_routing_entry kroute
= {};
1657 if (!kvm_gsi_routing_enabled()) {
1661 virq
= kvm_irqchip_get_virq(s
);
1667 kroute
.type
= KVM_IRQ_ROUTING_S390_ADAPTER
;
1669 kroute
.u
.adapter
.summary_addr
= adapter
->summary_addr
;
1670 kroute
.u
.adapter
.ind_addr
= adapter
->ind_addr
;
1671 kroute
.u
.adapter
.summary_offset
= adapter
->summary_offset
;
1672 kroute
.u
.adapter
.ind_offset
= adapter
->ind_offset
;
1673 kroute
.u
.adapter
.adapter_id
= adapter
->adapter_id
;
1675 kvm_add_routing_entry(s
, &kroute
);
1680 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1682 struct kvm_irq_routing_entry kroute
= {};
1685 if (!kvm_gsi_routing_enabled()) {
1688 if (!kvm_check_extension(s
, KVM_CAP_HYPERV_SYNIC
)) {
1691 virq
= kvm_irqchip_get_virq(s
);
1697 kroute
.type
= KVM_IRQ_ROUTING_HV_SINT
;
1699 kroute
.u
.hv_sint
.vcpu
= vcpu
;
1700 kroute
.u
.hv_sint
.sint
= sint
;
1702 kvm_add_routing_entry(s
, &kroute
);
1703 kvm_irqchip_commit_routes(s
);
1708 #else /* !KVM_CAP_IRQ_ROUTING */
1710 void kvm_init_irq_routing(KVMState
*s
)
1714 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1718 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1723 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1728 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1733 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1738 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1743 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
)
1747 #endif /* !KVM_CAP_IRQ_ROUTING */
1749 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1750 EventNotifier
*rn
, int virq
)
1752 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
),
1753 rn
? event_notifier_get_fd(rn
) : -1, virq
, true);
1756 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1759 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
), -1, virq
,
1763 int kvm_irqchip_add_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1764 EventNotifier
*rn
, qemu_irq irq
)
1767 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1772 return kvm_irqchip_add_irqfd_notifier_gsi(s
, n
, rn
, GPOINTER_TO_INT(gsi
));
1775 int kvm_irqchip_remove_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1779 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1784 return kvm_irqchip_remove_irqfd_notifier_gsi(s
, n
, GPOINTER_TO_INT(gsi
));
1787 void kvm_irqchip_set_qemuirq_gsi(KVMState
*s
, qemu_irq irq
, int gsi
)
1789 g_hash_table_insert(s
->gsimap
, irq
, GINT_TO_POINTER(gsi
));
1792 static void kvm_irqchip_create(KVMState
*s
)
1796 assert(s
->kernel_irqchip_split
!= ON_OFF_AUTO_AUTO
);
1797 if (kvm_check_extension(s
, KVM_CAP_IRQCHIP
)) {
1799 } else if (kvm_check_extension(s
, KVM_CAP_S390_IRQCHIP
)) {
1800 ret
= kvm_vm_enable_cap(s
, KVM_CAP_S390_IRQCHIP
, 0);
1802 fprintf(stderr
, "Enable kernel irqchip failed: %s\n", strerror(-ret
));
1809 /* First probe and see if there's a arch-specific hook to create the
1810 * in-kernel irqchip for us */
1811 ret
= kvm_arch_irqchip_create(s
);
1813 if (s
->kernel_irqchip_split
== ON_OFF_AUTO_ON
) {
1814 perror("Split IRQ chip mode not supported.");
1817 ret
= kvm_vm_ioctl(s
, KVM_CREATE_IRQCHIP
);
1821 fprintf(stderr
, "Create kernel irqchip failed: %s\n", strerror(-ret
));
1825 kvm_kernel_irqchip
= true;
1826 /* If we have an in-kernel IRQ chip then we must have asynchronous
1827 * interrupt delivery (though the reverse is not necessarily true)
1829 kvm_async_interrupts_allowed
= true;
1830 kvm_halt_in_kernel_allowed
= true;
1832 kvm_init_irq_routing(s
);
1834 s
->gsimap
= g_hash_table_new(g_direct_hash
, g_direct_equal
);
1837 /* Find number of supported CPUs using the recommended
1838 * procedure from the kernel API documentation to cope with
1839 * older kernels that may be missing capabilities.
1841 static int kvm_recommended_vcpus(KVMState
*s
)
1843 int ret
= kvm_vm_check_extension(s
, KVM_CAP_NR_VCPUS
);
1844 return (ret
) ? ret
: 4;
1847 static int kvm_max_vcpus(KVMState
*s
)
1849 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPUS
);
1850 return (ret
) ? ret
: kvm_recommended_vcpus(s
);
1853 static int kvm_max_vcpu_id(KVMState
*s
)
1855 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPU_ID
);
1856 return (ret
) ? ret
: kvm_max_vcpus(s
);
1859 bool kvm_vcpu_id_is_valid(int vcpu_id
)
1861 KVMState
*s
= KVM_STATE(current_accel());
1862 return vcpu_id
>= 0 && vcpu_id
< kvm_max_vcpu_id(s
);
1865 static int kvm_init(MachineState
*ms
)
1867 MachineClass
*mc
= MACHINE_GET_CLASS(ms
);
1868 static const char upgrade_note
[] =
1869 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1870 "(see http://sourceforge.net/projects/kvm).\n";
1875 { "SMP", ms
->smp
.cpus
},
1876 { "hotpluggable", ms
->smp
.max_cpus
},
1879 int soft_vcpus_limit
, hard_vcpus_limit
;
1881 const KVMCapabilityInfo
*missing_cap
;
1884 const char *kvm_type
;
1886 s
= KVM_STATE(ms
->accelerator
);
1889 * On systems where the kernel can support different base page
1890 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1891 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1892 * page size for the system though.
1894 assert(TARGET_PAGE_SIZE
<= qemu_real_host_page_size
);
1898 #ifdef KVM_CAP_SET_GUEST_DEBUG
1899 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
1901 QLIST_INIT(&s
->kvm_parked_vcpus
);
1903 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
1905 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
1910 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
1911 if (ret
< KVM_API_VERSION
) {
1915 fprintf(stderr
, "kvm version too old\n");
1919 if (ret
> KVM_API_VERSION
) {
1921 fprintf(stderr
, "kvm version not supported\n");
1925 kvm_immediate_exit
= kvm_check_extension(s
, KVM_CAP_IMMEDIATE_EXIT
);
1926 s
->nr_slots
= kvm_check_extension(s
, KVM_CAP_NR_MEMSLOTS
);
1928 /* If unspecified, use the default value */
1933 s
->nr_as
= kvm_check_extension(s
, KVM_CAP_MULTI_ADDRESS_SPACE
);
1934 if (s
->nr_as
<= 1) {
1937 s
->as
= g_new0(struct KVMAs
, s
->nr_as
);
1939 kvm_type
= qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1941 type
= mc
->kvm_type(ms
, kvm_type
);
1942 } else if (kvm_type
) {
1944 fprintf(stderr
, "Invalid argument kvm-type=%s\n", kvm_type
);
1949 ret
= kvm_ioctl(s
, KVM_CREATE_VM
, type
);
1950 } while (ret
== -EINTR
);
1953 fprintf(stderr
, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret
,
1957 if (ret
== -EINVAL
) {
1959 "Host kernel setup problem detected. Please verify:\n");
1960 fprintf(stderr
, "- for kernels supporting the switch_amode or"
1961 " user_mode parameters, whether\n");
1963 " user space is running in primary address space\n");
1965 "- for kernels supporting the vm.allocate_pgste sysctl, "
1966 "whether it is enabled\n");
1974 /* check the vcpu limits */
1975 soft_vcpus_limit
= kvm_recommended_vcpus(s
);
1976 hard_vcpus_limit
= kvm_max_vcpus(s
);
1979 if (nc
->num
> soft_vcpus_limit
) {
1980 warn_report("Number of %s cpus requested (%d) exceeds "
1981 "the recommended cpus supported by KVM (%d)",
1982 nc
->name
, nc
->num
, soft_vcpus_limit
);
1984 if (nc
->num
> hard_vcpus_limit
) {
1985 fprintf(stderr
, "Number of %s cpus requested (%d) exceeds "
1986 "the maximum cpus supported by KVM (%d)\n",
1987 nc
->name
, nc
->num
, hard_vcpus_limit
);
1994 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
1997 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
2001 fprintf(stderr
, "kvm does not support %s\n%s",
2002 missing_cap
->name
, upgrade_note
);
2006 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
2007 s
->coalesced_pio
= s
->coalesced_mmio
&&
2008 kvm_check_extension(s
, KVM_CAP_COALESCED_PIO
);
2010 s
->manual_dirty_log_protect
=
2011 kvm_check_extension(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
);
2012 if (s
->manual_dirty_log_protect
) {
2013 ret
= kvm_vm_enable_cap(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
, 0, 1);
2015 warn_report("Trying to enable KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 "
2016 "but failed. Falling back to the legacy mode. ");
2017 s
->manual_dirty_log_protect
= false;
2021 #ifdef KVM_CAP_VCPU_EVENTS
2022 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
2025 s
->robust_singlestep
=
2026 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
2028 #ifdef KVM_CAP_DEBUGREGS
2029 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
2032 s
->max_nested_state_len
= kvm_check_extension(s
, KVM_CAP_NESTED_STATE
);
2034 #ifdef KVM_CAP_IRQ_ROUTING
2035 kvm_direct_msi_allowed
= (kvm_check_extension(s
, KVM_CAP_SIGNAL_MSI
) > 0);
2038 s
->intx_set_mask
= kvm_check_extension(s
, KVM_CAP_PCI_2_3
);
2040 s
->irq_set_ioctl
= KVM_IRQ_LINE
;
2041 if (kvm_check_extension(s
, KVM_CAP_IRQ_INJECT_STATUS
)) {
2042 s
->irq_set_ioctl
= KVM_IRQ_LINE_STATUS
;
2045 kvm_readonly_mem_allowed
=
2046 (kvm_check_extension(s
, KVM_CAP_READONLY_MEM
) > 0);
2048 kvm_eventfds_allowed
=
2049 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD
) > 0);
2051 kvm_irqfds_allowed
=
2052 (kvm_check_extension(s
, KVM_CAP_IRQFD
) > 0);
2054 kvm_resamplefds_allowed
=
2055 (kvm_check_extension(s
, KVM_CAP_IRQFD_RESAMPLE
) > 0);
2057 kvm_vm_attributes_allowed
=
2058 (kvm_check_extension(s
, KVM_CAP_VM_ATTRIBUTES
) > 0);
2060 kvm_ioeventfd_any_length_allowed
=
2061 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD_ANY_LENGTH
) > 0);
2066 * if memory encryption object is specified then initialize the memory
2067 * encryption context.
2069 if (ms
->memory_encryption
) {
2070 kvm_state
->memcrypt_handle
= sev_guest_init(ms
->memory_encryption
);
2071 if (!kvm_state
->memcrypt_handle
) {
2076 kvm_state
->memcrypt_encrypt_data
= sev_encrypt_data
;
2079 ret
= kvm_arch_init(ms
, s
);
2084 if (s
->kernel_irqchip_split
== ON_OFF_AUTO_AUTO
) {
2085 s
->kernel_irqchip_split
= mc
->default_kernel_irqchip_split
? ON_OFF_AUTO_ON
: ON_OFF_AUTO_OFF
;
2088 if (s
->kernel_irqchip_allowed
) {
2089 kvm_irqchip_create(s
);
2092 if (kvm_eventfds_allowed
) {
2093 s
->memory_listener
.listener
.eventfd_add
= kvm_mem_ioeventfd_add
;
2094 s
->memory_listener
.listener
.eventfd_del
= kvm_mem_ioeventfd_del
;
2096 s
->memory_listener
.listener
.coalesced_io_add
= kvm_coalesce_mmio_region
;
2097 s
->memory_listener
.listener
.coalesced_io_del
= kvm_uncoalesce_mmio_region
;
2099 kvm_memory_listener_register(s
, &s
->memory_listener
,
2100 &address_space_memory
, 0);
2101 memory_listener_register(&kvm_io_listener
,
2103 memory_listener_register(&kvm_coalesced_pio_listener
,
2106 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
2108 s
->sync_mmu
= !!kvm_vm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
2110 qemu_balloon_inhibit(true);
2123 g_free(s
->memory_listener
.slots
);
2128 void kvm_set_sigmask_len(KVMState
*s
, unsigned int sigmask_len
)
2130 s
->sigmask_len
= sigmask_len
;
2133 static void kvm_handle_io(uint16_t port
, MemTxAttrs attrs
, void *data
, int direction
,
2134 int size
, uint32_t count
)
2137 uint8_t *ptr
= data
;
2139 for (i
= 0; i
< count
; i
++) {
2140 address_space_rw(&address_space_io
, port
, attrs
,
2142 direction
== KVM_EXIT_IO_OUT
);
2147 static int kvm_handle_internal_error(CPUState
*cpu
, struct kvm_run
*run
)
2149 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
2150 run
->internal
.suberror
);
2152 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
2155 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
2156 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
2157 i
, (uint64_t)run
->internal
.data
[i
]);
2160 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
2161 fprintf(stderr
, "emulation failure\n");
2162 if (!kvm_arch_stop_on_emulation_error(cpu
)) {
2163 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2164 return EXCP_INTERRUPT
;
2167 /* FIXME: Should trigger a qmp message to let management know
2168 * something went wrong.
2173 void kvm_flush_coalesced_mmio_buffer(void)
2175 KVMState
*s
= kvm_state
;
2177 if (s
->coalesced_flush_in_progress
) {
2181 s
->coalesced_flush_in_progress
= true;
2183 if (s
->coalesced_mmio_ring
) {
2184 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
2185 while (ring
->first
!= ring
->last
) {
2186 struct kvm_coalesced_mmio
*ent
;
2188 ent
= &ring
->coalesced_mmio
[ring
->first
];
2190 if (ent
->pio
== 1) {
2191 address_space_write(&address_space_io
, ent
->phys_addr
,
2192 MEMTXATTRS_UNSPECIFIED
, ent
->data
,
2195 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
2198 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
2202 s
->coalesced_flush_in_progress
= false;
2205 static void do_kvm_cpu_synchronize_state(CPUState
*cpu
, run_on_cpu_data arg
)
2207 if (!cpu
->vcpu_dirty
) {
2208 kvm_arch_get_registers(cpu
);
2209 cpu
->vcpu_dirty
= true;
2213 void kvm_cpu_synchronize_state(CPUState
*cpu
)
2215 if (!cpu
->vcpu_dirty
) {
2216 run_on_cpu(cpu
, do_kvm_cpu_synchronize_state
, RUN_ON_CPU_NULL
);
2220 static void do_kvm_cpu_synchronize_post_reset(CPUState
*cpu
, run_on_cpu_data arg
)
2222 kvm_arch_put_registers(cpu
, KVM_PUT_RESET_STATE
);
2223 cpu
->vcpu_dirty
= false;
2226 void kvm_cpu_synchronize_post_reset(CPUState
*cpu
)
2228 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_reset
, RUN_ON_CPU_NULL
);
2231 static void do_kvm_cpu_synchronize_post_init(CPUState
*cpu
, run_on_cpu_data arg
)
2233 kvm_arch_put_registers(cpu
, KVM_PUT_FULL_STATE
);
2234 cpu
->vcpu_dirty
= false;
2237 void kvm_cpu_synchronize_post_init(CPUState
*cpu
)
2239 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_init
, RUN_ON_CPU_NULL
);
2242 static void do_kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
, run_on_cpu_data arg
)
2244 cpu
->vcpu_dirty
= true;
2247 void kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
)
2249 run_on_cpu(cpu
, do_kvm_cpu_synchronize_pre_loadvm
, RUN_ON_CPU_NULL
);
2252 #ifdef KVM_HAVE_MCE_INJECTION
2253 static __thread
void *pending_sigbus_addr
;
2254 static __thread
int pending_sigbus_code
;
2255 static __thread
bool have_sigbus_pending
;
2258 static void kvm_cpu_kick(CPUState
*cpu
)
2260 atomic_set(&cpu
->kvm_run
->immediate_exit
, 1);
2263 static void kvm_cpu_kick_self(void)
2265 if (kvm_immediate_exit
) {
2266 kvm_cpu_kick(current_cpu
);
2268 qemu_cpu_kick_self();
2272 static void kvm_eat_signals(CPUState
*cpu
)
2274 struct timespec ts
= { 0, 0 };
2280 if (kvm_immediate_exit
) {
2281 atomic_set(&cpu
->kvm_run
->immediate_exit
, 0);
2282 /* Write kvm_run->immediate_exit before the cpu->exit_request
2283 * write in kvm_cpu_exec.
2289 sigemptyset(&waitset
);
2290 sigaddset(&waitset
, SIG_IPI
);
2293 r
= sigtimedwait(&waitset
, &siginfo
, &ts
);
2294 if (r
== -1 && !(errno
== EAGAIN
|| errno
== EINTR
)) {
2295 perror("sigtimedwait");
2299 r
= sigpending(&chkset
);
2301 perror("sigpending");
2304 } while (sigismember(&chkset
, SIG_IPI
));
2307 int kvm_cpu_exec(CPUState
*cpu
)
2309 struct kvm_run
*run
= cpu
->kvm_run
;
2312 DPRINTF("kvm_cpu_exec()\n");
2314 if (kvm_arch_process_async_events(cpu
)) {
2315 atomic_set(&cpu
->exit_request
, 0);
2319 qemu_mutex_unlock_iothread();
2320 cpu_exec_start(cpu
);
2325 if (cpu
->vcpu_dirty
) {
2326 kvm_arch_put_registers(cpu
, KVM_PUT_RUNTIME_STATE
);
2327 cpu
->vcpu_dirty
= false;
2330 kvm_arch_pre_run(cpu
, run
);
2331 if (atomic_read(&cpu
->exit_request
)) {
2332 DPRINTF("interrupt exit requested\n");
2334 * KVM requires us to reenter the kernel after IO exits to complete
2335 * instruction emulation. This self-signal will ensure that we
2338 kvm_cpu_kick_self();
2341 /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
2342 * Matching barrier in kvm_eat_signals.
2346 run_ret
= kvm_vcpu_ioctl(cpu
, KVM_RUN
, 0);
2348 attrs
= kvm_arch_post_run(cpu
, run
);
2350 #ifdef KVM_HAVE_MCE_INJECTION
2351 if (unlikely(have_sigbus_pending
)) {
2352 qemu_mutex_lock_iothread();
2353 kvm_arch_on_sigbus_vcpu(cpu
, pending_sigbus_code
,
2354 pending_sigbus_addr
);
2355 have_sigbus_pending
= false;
2356 qemu_mutex_unlock_iothread();
2361 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
2362 DPRINTF("io window exit\n");
2363 kvm_eat_signals(cpu
);
2364 ret
= EXCP_INTERRUPT
;
2367 fprintf(stderr
, "error: kvm run failed %s\n",
2368 strerror(-run_ret
));
2370 if (run_ret
== -EBUSY
) {
2372 "This is probably because your SMT is enabled.\n"
2373 "VCPU can only run on primary threads with all "
2374 "secondary threads offline.\n");
2381 trace_kvm_run_exit(cpu
->cpu_index
, run
->exit_reason
);
2382 switch (run
->exit_reason
) {
2384 DPRINTF("handle_io\n");
2385 /* Called outside BQL */
2386 kvm_handle_io(run
->io
.port
, attrs
,
2387 (uint8_t *)run
+ run
->io
.data_offset
,
2394 DPRINTF("handle_mmio\n");
2395 /* Called outside BQL */
2396 address_space_rw(&address_space_memory
,
2397 run
->mmio
.phys_addr
, attrs
,
2400 run
->mmio
.is_write
);
2403 case KVM_EXIT_IRQ_WINDOW_OPEN
:
2404 DPRINTF("irq_window_open\n");
2405 ret
= EXCP_INTERRUPT
;
2407 case KVM_EXIT_SHUTDOWN
:
2408 DPRINTF("shutdown\n");
2409 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2410 ret
= EXCP_INTERRUPT
;
2412 case KVM_EXIT_UNKNOWN
:
2413 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
2414 (uint64_t)run
->hw
.hardware_exit_reason
);
2417 case KVM_EXIT_INTERNAL_ERROR
:
2418 ret
= kvm_handle_internal_error(cpu
, run
);
2420 case KVM_EXIT_SYSTEM_EVENT
:
2421 switch (run
->system_event
.type
) {
2422 case KVM_SYSTEM_EVENT_SHUTDOWN
:
2423 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN
);
2424 ret
= EXCP_INTERRUPT
;
2426 case KVM_SYSTEM_EVENT_RESET
:
2427 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2428 ret
= EXCP_INTERRUPT
;
2430 case KVM_SYSTEM_EVENT_CRASH
:
2431 kvm_cpu_synchronize_state(cpu
);
2432 qemu_mutex_lock_iothread();
2433 qemu_system_guest_panicked(cpu_get_crash_info(cpu
));
2434 qemu_mutex_unlock_iothread();
2438 DPRINTF("kvm_arch_handle_exit\n");
2439 ret
= kvm_arch_handle_exit(cpu
, run
);
2444 DPRINTF("kvm_arch_handle_exit\n");
2445 ret
= kvm_arch_handle_exit(cpu
, run
);
2451 qemu_mutex_lock_iothread();
2454 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2455 vm_stop(RUN_STATE_INTERNAL_ERROR
);
2458 atomic_set(&cpu
->exit_request
, 0);
2462 int kvm_ioctl(KVMState
*s
, int type
, ...)
2469 arg
= va_arg(ap
, void *);
2472 trace_kvm_ioctl(type
, arg
);
2473 ret
= ioctl(s
->fd
, type
, arg
);
2480 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
2487 arg
= va_arg(ap
, void *);
2490 trace_kvm_vm_ioctl(type
, arg
);
2491 ret
= ioctl(s
->vmfd
, type
, arg
);
2498 int kvm_vcpu_ioctl(CPUState
*cpu
, int type
, ...)
2505 arg
= va_arg(ap
, void *);
2508 trace_kvm_vcpu_ioctl(cpu
->cpu_index
, type
, arg
);
2509 ret
= ioctl(cpu
->kvm_fd
, type
, arg
);
2516 int kvm_device_ioctl(int fd
, int type
, ...)
2523 arg
= va_arg(ap
, void *);
2526 trace_kvm_device_ioctl(fd
, type
, arg
);
2527 ret
= ioctl(fd
, type
, arg
);
2534 int kvm_vm_check_attr(KVMState
*s
, uint32_t group
, uint64_t attr
)
2537 struct kvm_device_attr attribute
= {
2542 if (!kvm_vm_attributes_allowed
) {
2546 ret
= kvm_vm_ioctl(s
, KVM_HAS_DEVICE_ATTR
, &attribute
);
2547 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2551 int kvm_device_check_attr(int dev_fd
, uint32_t group
, uint64_t attr
)
2553 struct kvm_device_attr attribute
= {
2559 return kvm_device_ioctl(dev_fd
, KVM_HAS_DEVICE_ATTR
, &attribute
) ? 0 : 1;
2562 int kvm_device_access(int fd
, int group
, uint64_t attr
,
2563 void *val
, bool write
, Error
**errp
)
2565 struct kvm_device_attr kvmattr
;
2569 kvmattr
.group
= group
;
2570 kvmattr
.attr
= attr
;
2571 kvmattr
.addr
= (uintptr_t)val
;
2573 err
= kvm_device_ioctl(fd
,
2574 write
? KVM_SET_DEVICE_ATTR
: KVM_GET_DEVICE_ATTR
,
2577 error_setg_errno(errp
, -err
,
2578 "KVM_%s_DEVICE_ATTR failed: Group %d "
2579 "attr 0x%016" PRIx64
,
2580 write
? "SET" : "GET", group
, attr
);
2585 bool kvm_has_sync_mmu(void)
2587 return kvm_state
->sync_mmu
;
2590 int kvm_has_vcpu_events(void)
2592 return kvm_state
->vcpu_events
;
2595 int kvm_has_robust_singlestep(void)
2597 return kvm_state
->robust_singlestep
;
2600 int kvm_has_debugregs(void)
2602 return kvm_state
->debugregs
;
2605 int kvm_max_nested_state_length(void)
2607 return kvm_state
->max_nested_state_len
;
2610 int kvm_has_many_ioeventfds(void)
2612 if (!kvm_enabled()) {
2615 return kvm_state
->many_ioeventfds
;
2618 int kvm_has_gsi_routing(void)
2620 #ifdef KVM_CAP_IRQ_ROUTING
2621 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
2627 int kvm_has_intx_set_mask(void)
2629 return kvm_state
->intx_set_mask
;
2632 bool kvm_arm_supports_user_irq(void)
2634 return kvm_check_extension(kvm_state
, KVM_CAP_ARM_USER_IRQ
);
2637 #ifdef KVM_CAP_SET_GUEST_DEBUG
2638 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*cpu
,
2641 struct kvm_sw_breakpoint
*bp
;
2643 QTAILQ_FOREACH(bp
, &cpu
->kvm_state
->kvm_sw_breakpoints
, entry
) {
2651 int kvm_sw_breakpoints_active(CPUState
*cpu
)
2653 return !QTAILQ_EMPTY(&cpu
->kvm_state
->kvm_sw_breakpoints
);
2656 struct kvm_set_guest_debug_data
{
2657 struct kvm_guest_debug dbg
;
2661 static void kvm_invoke_set_guest_debug(CPUState
*cpu
, run_on_cpu_data data
)
2663 struct kvm_set_guest_debug_data
*dbg_data
=
2664 (struct kvm_set_guest_debug_data
*) data
.host_ptr
;
2666 dbg_data
->err
= kvm_vcpu_ioctl(cpu
, KVM_SET_GUEST_DEBUG
,
2670 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2672 struct kvm_set_guest_debug_data data
;
2674 data
.dbg
.control
= reinject_trap
;
2676 if (cpu
->singlestep_enabled
) {
2677 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
2679 kvm_arch_update_guest_debug(cpu
, &data
.dbg
);
2681 run_on_cpu(cpu
, kvm_invoke_set_guest_debug
,
2682 RUN_ON_CPU_HOST_PTR(&data
));
2686 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2687 target_ulong len
, int type
)
2689 struct kvm_sw_breakpoint
*bp
;
2692 if (type
== GDB_BREAKPOINT_SW
) {
2693 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2699 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
2702 err
= kvm_arch_insert_sw_breakpoint(cpu
, bp
);
2708 QTAILQ_INSERT_HEAD(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2710 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
2717 err
= kvm_update_guest_debug(cpu
, 0);
2725 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2726 target_ulong len
, int type
)
2728 struct kvm_sw_breakpoint
*bp
;
2731 if (type
== GDB_BREAKPOINT_SW
) {
2732 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2737 if (bp
->use_count
> 1) {
2742 err
= kvm_arch_remove_sw_breakpoint(cpu
, bp
);
2747 QTAILQ_REMOVE(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2750 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
2757 err
= kvm_update_guest_debug(cpu
, 0);
2765 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2767 struct kvm_sw_breakpoint
*bp
, *next
;
2768 KVMState
*s
= cpu
->kvm_state
;
2771 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
2772 if (kvm_arch_remove_sw_breakpoint(cpu
, bp
) != 0) {
2773 /* Try harder to find a CPU that currently sees the breakpoint. */
2774 CPU_FOREACH(tmpcpu
) {
2775 if (kvm_arch_remove_sw_breakpoint(tmpcpu
, bp
) == 0) {
2780 QTAILQ_REMOVE(&s
->kvm_sw_breakpoints
, bp
, entry
);
2783 kvm_arch_remove_all_hw_breakpoints();
2786 kvm_update_guest_debug(cpu
, 0);
2790 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2792 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2797 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2798 target_ulong len
, int type
)
2803 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2804 target_ulong len
, int type
)
2809 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2812 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2814 static int kvm_set_signal_mask(CPUState
*cpu
, const sigset_t
*sigset
)
2816 KVMState
*s
= kvm_state
;
2817 struct kvm_signal_mask
*sigmask
;
2820 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
2822 sigmask
->len
= s
->sigmask_len
;
2823 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
2824 r
= kvm_vcpu_ioctl(cpu
, KVM_SET_SIGNAL_MASK
, sigmask
);
2830 static void kvm_ipi_signal(int sig
)
2833 assert(kvm_immediate_exit
);
2834 kvm_cpu_kick(current_cpu
);
2838 void kvm_init_cpu_signals(CPUState
*cpu
)
2842 struct sigaction sigact
;
2844 memset(&sigact
, 0, sizeof(sigact
));
2845 sigact
.sa_handler
= kvm_ipi_signal
;
2846 sigaction(SIG_IPI
, &sigact
, NULL
);
2848 pthread_sigmask(SIG_BLOCK
, NULL
, &set
);
2849 #if defined KVM_HAVE_MCE_INJECTION
2850 sigdelset(&set
, SIGBUS
);
2851 pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2853 sigdelset(&set
, SIG_IPI
);
2854 if (kvm_immediate_exit
) {
2855 r
= pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2857 r
= kvm_set_signal_mask(cpu
, &set
);
2860 fprintf(stderr
, "kvm_set_signal_mask: %s\n", strerror(-r
));
2865 /* Called asynchronously in VCPU thread. */
2866 int kvm_on_sigbus_vcpu(CPUState
*cpu
, int code
, void *addr
)
2868 #ifdef KVM_HAVE_MCE_INJECTION
2869 if (have_sigbus_pending
) {
2872 have_sigbus_pending
= true;
2873 pending_sigbus_addr
= addr
;
2874 pending_sigbus_code
= code
;
2875 atomic_set(&cpu
->exit_request
, 1);
2882 /* Called synchronously (via signalfd) in main thread. */
2883 int kvm_on_sigbus(int code
, void *addr
)
2885 #ifdef KVM_HAVE_MCE_INJECTION
2886 /* Action required MCE kills the process if SIGBUS is blocked. Because
2887 * that's what happens in the I/O thread, where we handle MCE via signalfd,
2888 * we can only get action optional here.
2890 assert(code
!= BUS_MCEERR_AR
);
2891 kvm_arch_on_sigbus_vcpu(first_cpu
, code
, addr
);
2898 int kvm_create_device(KVMState
*s
, uint64_t type
, bool test
)
2901 struct kvm_create_device create_dev
;
2903 create_dev
.type
= type
;
2905 create_dev
.flags
= test
? KVM_CREATE_DEVICE_TEST
: 0;
2907 if (!kvm_check_extension(s
, KVM_CAP_DEVICE_CTRL
)) {
2911 ret
= kvm_vm_ioctl(s
, KVM_CREATE_DEVICE
, &create_dev
);
2916 return test
? 0 : create_dev
.fd
;
2919 bool kvm_device_supported(int vmfd
, uint64_t type
)
2921 struct kvm_create_device create_dev
= {
2924 .flags
= KVM_CREATE_DEVICE_TEST
,
2927 if (ioctl(vmfd
, KVM_CHECK_EXTENSION
, KVM_CAP_DEVICE_CTRL
) <= 0) {
2931 return (ioctl(vmfd
, KVM_CREATE_DEVICE
, &create_dev
) >= 0);
2934 int kvm_set_one_reg(CPUState
*cs
, uint64_t id
, void *source
)
2936 struct kvm_one_reg reg
;
2940 reg
.addr
= (uintptr_t) source
;
2941 r
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
2943 trace_kvm_failed_reg_set(id
, strerror(-r
));
2948 int kvm_get_one_reg(CPUState
*cs
, uint64_t id
, void *target
)
2950 struct kvm_one_reg reg
;
2954 reg
.addr
= (uintptr_t) target
;
2955 r
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
2957 trace_kvm_failed_reg_get(id
, strerror(-r
));
2962 static bool kvm_accel_has_memory(MachineState
*ms
, AddressSpace
*as
,
2963 hwaddr start_addr
, hwaddr size
)
2965 KVMState
*kvm
= KVM_STATE(ms
->accelerator
);
2968 for (i
= 0; i
< kvm
->nr_as
; ++i
) {
2969 if (kvm
->as
[i
].as
== as
&& kvm
->as
[i
].ml
) {
2970 size
= MIN(kvm_max_slot_size
, size
);
2971 return NULL
!= kvm_lookup_matching_slot(kvm
->as
[i
].ml
,
2979 static void kvm_get_kvm_shadow_mem(Object
*obj
, Visitor
*v
,
2980 const char *name
, void *opaque
,
2983 KVMState
*s
= KVM_STATE(obj
);
2984 int64_t value
= s
->kvm_shadow_mem
;
2986 visit_type_int(v
, name
, &value
, errp
);
2989 static void kvm_set_kvm_shadow_mem(Object
*obj
, Visitor
*v
,
2990 const char *name
, void *opaque
,
2993 KVMState
*s
= KVM_STATE(obj
);
2994 Error
*error
= NULL
;
2997 visit_type_int(v
, name
, &value
, &error
);
2999 error_propagate(errp
, error
);
3003 s
->kvm_shadow_mem
= value
;
3006 static void kvm_set_kernel_irqchip(Object
*obj
, Visitor
*v
,
3007 const char *name
, void *opaque
,
3011 KVMState
*s
= KVM_STATE(obj
);
3014 visit_type_OnOffSplit(v
, name
, &mode
, &err
);
3016 error_propagate(errp
, err
);
3020 case ON_OFF_SPLIT_ON
:
3021 s
->kernel_irqchip_allowed
= true;
3022 s
->kernel_irqchip_required
= true;
3023 s
->kernel_irqchip_split
= ON_OFF_AUTO_OFF
;
3025 case ON_OFF_SPLIT_OFF
:
3026 s
->kernel_irqchip_allowed
= false;
3027 s
->kernel_irqchip_required
= false;
3028 s
->kernel_irqchip_split
= ON_OFF_AUTO_OFF
;
3030 case ON_OFF_SPLIT_SPLIT
:
3031 s
->kernel_irqchip_allowed
= true;
3032 s
->kernel_irqchip_required
= true;
3033 s
->kernel_irqchip_split
= ON_OFF_AUTO_ON
;
3036 /* The value was checked in visit_type_OnOffSplit() above. If
3037 * we get here, then something is wrong in QEMU.
3044 bool kvm_kernel_irqchip_allowed(void)
3046 return kvm_state
->kernel_irqchip_allowed
;
3049 bool kvm_kernel_irqchip_required(void)
3051 return kvm_state
->kernel_irqchip_required
;
3054 bool kvm_kernel_irqchip_split(void)
3056 return kvm_state
->kernel_irqchip_split
== ON_OFF_AUTO_ON
;
3059 static void kvm_accel_instance_init(Object
*obj
)
3061 KVMState
*s
= KVM_STATE(obj
);
3063 s
->kvm_shadow_mem
= -1;
3064 s
->kernel_irqchip_allowed
= true;
3065 s
->kernel_irqchip_split
= ON_OFF_AUTO_AUTO
;
3068 static void kvm_accel_class_init(ObjectClass
*oc
, void *data
)
3070 AccelClass
*ac
= ACCEL_CLASS(oc
);
3072 ac
->init_machine
= kvm_init
;
3073 ac
->has_memory
= kvm_accel_has_memory
;
3074 ac
->allowed
= &kvm_allowed
;
3076 object_class_property_add(oc
, "kernel-irqchip", "on|off|split",
3077 NULL
, kvm_set_kernel_irqchip
,
3078 NULL
, NULL
, &error_abort
);
3079 object_class_property_set_description(oc
, "kernel-irqchip",
3080 "Configure KVM in-kernel irqchip", &error_abort
);
3082 object_class_property_add(oc
, "kvm-shadow-mem", "int",
3083 kvm_get_kvm_shadow_mem
, kvm_set_kvm_shadow_mem
,
3084 NULL
, NULL
, &error_abort
);
3085 object_class_property_set_description(oc
, "kvm-shadow-mem",
3086 "KVM shadow MMU size", &error_abort
);
3089 static const TypeInfo kvm_accel_type
= {
3090 .name
= TYPE_KVM_ACCEL
,
3091 .parent
= TYPE_ACCEL
,
3092 .instance_init
= kvm_accel_instance_init
,
3093 .class_init
= kvm_accel_class_init
,
3094 .instance_size
= sizeof(KVMState
),
3097 static void kvm_type_init(void)
3099 type_register_static(&kvm_accel_type
);
3102 type_init(kvm_type_init
);