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"
45 #include "hw/boards.h"
47 /* This check must be after config-host.h is included */
49 #include <sys/eventfd.h>
52 /* KVM uses PAGE_SIZE in its definition of KVM_COALESCED_MMIO_MAX. We
53 * need to use the real host PAGE_SIZE, as that's what KVM will use.
55 #define PAGE_SIZE qemu_real_host_page_size
60 #define DPRINTF(fmt, ...) \
61 do { fprintf(stderr, fmt, ## __VA_ARGS__); } while (0)
63 #define DPRINTF(fmt, ...) \
67 #define KVM_MSI_HASHTAB_SIZE 256
69 struct KVMParkedVcpu
{
70 unsigned long vcpu_id
;
72 QLIST_ENTRY(KVMParkedVcpu
) node
;
77 AccelState parent_obj
;
84 struct kvm_coalesced_mmio_ring
*coalesced_mmio_ring
;
85 bool coalesced_flush_in_progress
;
87 int robust_singlestep
;
89 #ifdef KVM_CAP_SET_GUEST_DEBUG
90 QTAILQ_HEAD(, kvm_sw_breakpoint
) kvm_sw_breakpoints
;
92 int max_nested_state_len
;
96 bool manual_dirty_log_protect
;
97 /* The man page (and posix) say ioctl numbers are signed int, but
98 * they're not. Linux, glibc and *BSD all treat ioctl numbers as
99 * unsigned, and treating them as signed here can break things */
100 unsigned irq_set_ioctl
;
101 unsigned int sigmask_len
;
103 #ifdef KVM_CAP_IRQ_ROUTING
104 struct kvm_irq_routing
*irq_routes
;
105 int nr_allocated_irq_routes
;
106 unsigned long *used_gsi_bitmap
;
107 unsigned int gsi_count
;
108 QTAILQ_HEAD(, KVMMSIRoute
) msi_hashtab
[KVM_MSI_HASHTAB_SIZE
];
110 KVMMemoryListener memory_listener
;
111 QLIST_HEAD(, KVMParkedVcpu
) kvm_parked_vcpus
;
113 /* memory encryption */
114 void *memcrypt_handle
;
115 int (*memcrypt_encrypt_data
)(void *handle
, uint8_t *ptr
, uint64_t len
);
117 /* For "info mtree -f" to tell if an MR is registered in KVM */
120 KVMMemoryListener
*ml
;
126 bool kvm_kernel_irqchip
;
127 bool kvm_split_irqchip
;
128 bool kvm_async_interrupts_allowed
;
129 bool kvm_halt_in_kernel_allowed
;
130 bool kvm_eventfds_allowed
;
131 bool kvm_irqfds_allowed
;
132 bool kvm_resamplefds_allowed
;
133 bool kvm_msi_via_irqfd_allowed
;
134 bool kvm_gsi_routing_allowed
;
135 bool kvm_gsi_direct_mapping
;
137 bool kvm_readonly_mem_allowed
;
138 bool kvm_vm_attributes_allowed
;
139 bool kvm_direct_msi_allowed
;
140 bool kvm_ioeventfd_any_length_allowed
;
141 bool kvm_msi_use_devid
;
142 static bool kvm_immediate_exit
;
143 static hwaddr kvm_max_slot_size
= ~0;
145 static const KVMCapabilityInfo kvm_required_capabilites
[] = {
146 KVM_CAP_INFO(USER_MEMORY
),
147 KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS
),
148 KVM_CAP_INFO(JOIN_MEMORY_REGIONS_WORKS
),
152 static NotifierList kvm_irqchip_change_notifiers
=
153 NOTIFIER_LIST_INITIALIZER(kvm_irqchip_change_notifiers
);
155 #define kvm_slots_lock(kml) qemu_mutex_lock(&(kml)->slots_lock)
156 #define kvm_slots_unlock(kml) qemu_mutex_unlock(&(kml)->slots_lock)
158 int kvm_get_max_memslots(void)
160 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
165 bool kvm_memcrypt_enabled(void)
167 if (kvm_state
&& kvm_state
->memcrypt_handle
) {
174 int kvm_memcrypt_encrypt_data(uint8_t *ptr
, uint64_t len
)
176 if (kvm_state
->memcrypt_handle
&&
177 kvm_state
->memcrypt_encrypt_data
) {
178 return kvm_state
->memcrypt_encrypt_data(kvm_state
->memcrypt_handle
,
185 /* Called with KVMMemoryListener.slots_lock held */
186 static KVMSlot
*kvm_get_free_slot(KVMMemoryListener
*kml
)
188 KVMState
*s
= kvm_state
;
191 for (i
= 0; i
< s
->nr_slots
; i
++) {
192 if (kml
->slots
[i
].memory_size
== 0) {
193 return &kml
->slots
[i
];
200 bool kvm_has_free_slot(MachineState
*ms
)
202 KVMState
*s
= KVM_STATE(ms
->accelerator
);
204 KVMMemoryListener
*kml
= &s
->memory_listener
;
207 result
= !!kvm_get_free_slot(kml
);
208 kvm_slots_unlock(kml
);
213 /* Called with KVMMemoryListener.slots_lock held */
214 static KVMSlot
*kvm_alloc_slot(KVMMemoryListener
*kml
)
216 KVMSlot
*slot
= kvm_get_free_slot(kml
);
222 fprintf(stderr
, "%s: no free slot available\n", __func__
);
226 static KVMSlot
*kvm_lookup_matching_slot(KVMMemoryListener
*kml
,
230 KVMState
*s
= kvm_state
;
233 for (i
= 0; i
< s
->nr_slots
; i
++) {
234 KVMSlot
*mem
= &kml
->slots
[i
];
236 if (start_addr
== mem
->start_addr
&& size
== mem
->memory_size
) {
245 * Calculate and align the start address and the size of the section.
246 * Return the size. If the size is 0, the aligned section is empty.
248 static hwaddr
kvm_align_section(MemoryRegionSection
*section
,
251 hwaddr size
= int128_get64(section
->size
);
252 hwaddr delta
, aligned
;
254 /* kvm works in page size chunks, but the function may be called
255 with sub-page size and unaligned start address. Pad the start
256 address to next and truncate size to previous page boundary. */
257 aligned
= ROUND_UP(section
->offset_within_address_space
,
258 qemu_real_host_page_size
);
259 delta
= aligned
- section
->offset_within_address_space
;
265 return (size
- delta
) & qemu_real_host_page_mask
;
268 int kvm_physical_memory_addr_from_host(KVMState
*s
, void *ram
,
271 KVMMemoryListener
*kml
= &s
->memory_listener
;
275 for (i
= 0; i
< s
->nr_slots
; i
++) {
276 KVMSlot
*mem
= &kml
->slots
[i
];
278 if (ram
>= mem
->ram
&& ram
< mem
->ram
+ mem
->memory_size
) {
279 *phys_addr
= mem
->start_addr
+ (ram
- mem
->ram
);
284 kvm_slots_unlock(kml
);
289 static int kvm_set_user_memory_region(KVMMemoryListener
*kml
, KVMSlot
*slot
, bool new)
291 KVMState
*s
= kvm_state
;
292 struct kvm_userspace_memory_region mem
;
295 mem
.slot
= slot
->slot
| (kml
->as_id
<< 16);
296 mem
.guest_phys_addr
= slot
->start_addr
;
297 mem
.userspace_addr
= (unsigned long)slot
->ram
;
298 mem
.flags
= slot
->flags
;
300 if (slot
->memory_size
&& !new && (mem
.flags
^ slot
->old_flags
) & KVM_MEM_READONLY
) {
301 /* Set the slot size to 0 before setting the slot to the desired
302 * value. This is needed based on KVM commit 75d61fbc. */
304 kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
306 mem
.memory_size
= slot
->memory_size
;
307 ret
= kvm_vm_ioctl(s
, KVM_SET_USER_MEMORY_REGION
, &mem
);
308 slot
->old_flags
= mem
.flags
;
309 trace_kvm_set_user_memory(mem
.slot
, mem
.flags
, mem
.guest_phys_addr
,
310 mem
.memory_size
, mem
.userspace_addr
, ret
);
314 int kvm_destroy_vcpu(CPUState
*cpu
)
316 KVMState
*s
= kvm_state
;
318 struct KVMParkedVcpu
*vcpu
= NULL
;
321 DPRINTF("kvm_destroy_vcpu\n");
323 ret
= kvm_arch_destroy_vcpu(cpu
);
328 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
331 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
335 ret
= munmap(cpu
->kvm_run
, mmap_size
);
340 vcpu
= g_malloc0(sizeof(*vcpu
));
341 vcpu
->vcpu_id
= kvm_arch_vcpu_id(cpu
);
342 vcpu
->kvm_fd
= cpu
->kvm_fd
;
343 QLIST_INSERT_HEAD(&kvm_state
->kvm_parked_vcpus
, vcpu
, node
);
348 static int kvm_get_vcpu(KVMState
*s
, unsigned long vcpu_id
)
350 struct KVMParkedVcpu
*cpu
;
352 QLIST_FOREACH(cpu
, &s
->kvm_parked_vcpus
, node
) {
353 if (cpu
->vcpu_id
== vcpu_id
) {
356 QLIST_REMOVE(cpu
, node
);
357 kvm_fd
= cpu
->kvm_fd
;
363 return kvm_vm_ioctl(s
, KVM_CREATE_VCPU
, (void *)vcpu_id
);
366 int kvm_init_vcpu(CPUState
*cpu
)
368 KVMState
*s
= kvm_state
;
372 DPRINTF("kvm_init_vcpu\n");
374 ret
= kvm_get_vcpu(s
, kvm_arch_vcpu_id(cpu
));
376 DPRINTF("kvm_create_vcpu failed\n");
382 cpu
->vcpu_dirty
= true;
384 mmap_size
= kvm_ioctl(s
, KVM_GET_VCPU_MMAP_SIZE
, 0);
387 DPRINTF("KVM_GET_VCPU_MMAP_SIZE failed\n");
391 cpu
->kvm_run
= mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
393 if (cpu
->kvm_run
== MAP_FAILED
) {
395 DPRINTF("mmap'ing vcpu state failed\n");
399 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
) {
400 s
->coalesced_mmio_ring
=
401 (void *)cpu
->kvm_run
+ s
->coalesced_mmio
* PAGE_SIZE
;
404 ret
= kvm_arch_init_vcpu(cpu
);
410 * dirty pages logging control
413 static int kvm_mem_flags(MemoryRegion
*mr
)
415 bool readonly
= mr
->readonly
|| memory_region_is_romd(mr
);
418 if (memory_region_get_dirty_log_mask(mr
) != 0) {
419 flags
|= KVM_MEM_LOG_DIRTY_PAGES
;
421 if (readonly
&& kvm_readonly_mem_allowed
) {
422 flags
|= KVM_MEM_READONLY
;
427 /* Called with KVMMemoryListener.slots_lock held */
428 static int kvm_slot_update_flags(KVMMemoryListener
*kml
, KVMSlot
*mem
,
431 mem
->flags
= kvm_mem_flags(mr
);
433 /* If nothing changed effectively, no need to issue ioctl */
434 if (mem
->flags
== mem
->old_flags
) {
438 return kvm_set_user_memory_region(kml
, mem
, false);
441 static int kvm_section_update_flags(KVMMemoryListener
*kml
,
442 MemoryRegionSection
*section
)
444 hwaddr start_addr
, size
, slot_size
;
448 size
= kvm_align_section(section
, &start_addr
);
455 while (size
&& !ret
) {
456 slot_size
= MIN(kvm_max_slot_size
, size
);
457 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
459 /* We don't have a slot if we want to trap every access. */
463 ret
= kvm_slot_update_flags(kml
, mem
, section
->mr
);
464 start_addr
+= slot_size
;
469 kvm_slots_unlock(kml
);
473 static void kvm_log_start(MemoryListener
*listener
,
474 MemoryRegionSection
*section
,
477 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
484 r
= kvm_section_update_flags(kml
, section
);
490 static void kvm_log_stop(MemoryListener
*listener
,
491 MemoryRegionSection
*section
,
494 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
501 r
= kvm_section_update_flags(kml
, section
);
507 /* get kvm's dirty pages bitmap and update qemu's */
508 static int kvm_get_dirty_pages_log_range(MemoryRegionSection
*section
,
509 unsigned long *bitmap
)
511 ram_addr_t start
= section
->offset_within_region
+
512 memory_region_get_ram_addr(section
->mr
);
513 ram_addr_t pages
= int128_get64(section
->size
) / qemu_real_host_page_size
;
515 cpu_physical_memory_set_dirty_lebitmap(bitmap
, start
, pages
);
519 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
521 /* Allocate the dirty bitmap for a slot */
522 static void kvm_memslot_init_dirty_bitmap(KVMSlot
*mem
)
525 * XXX bad kernel interface alert
526 * For dirty bitmap, kernel allocates array of size aligned to
527 * bits-per-long. But for case when the kernel is 64bits and
528 * the userspace is 32bits, userspace can't align to the same
529 * bits-per-long, since sizeof(long) is different between kernel
530 * and user space. This way, userspace will provide buffer which
531 * may be 4 bytes less than the kernel will use, resulting in
532 * userspace memory corruption (which is not detectable by valgrind
533 * too, in most cases).
534 * So for now, let's align to 64 instead of HOST_LONG_BITS here, in
535 * a hope that sizeof(long) won't become >8 any time soon.
537 hwaddr bitmap_size
= ALIGN(((mem
->memory_size
) >> TARGET_PAGE_BITS
),
538 /*HOST_LONG_BITS*/ 64) / 8;
539 mem
->dirty_bmap
= g_malloc0(bitmap_size
);
543 * kvm_physical_sync_dirty_bitmap - Sync dirty bitmap from kernel space
545 * This function will first try to fetch dirty bitmap from the kernel,
546 * and then updates qemu's dirty bitmap.
548 * NOTE: caller must be with kml->slots_lock held.
550 * @kml: the KVM memory listener object
551 * @section: the memory section to sync the dirty bitmap with
553 static int kvm_physical_sync_dirty_bitmap(KVMMemoryListener
*kml
,
554 MemoryRegionSection
*section
)
556 KVMState
*s
= kvm_state
;
557 struct kvm_dirty_log d
= {};
559 hwaddr start_addr
, size
;
560 hwaddr slot_size
, slot_offset
= 0;
563 size
= kvm_align_section(section
, &start_addr
);
565 MemoryRegionSection subsection
= *section
;
567 slot_size
= MIN(kvm_max_slot_size
, size
);
568 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
570 /* We don't have a slot if we want to trap every access. */
574 if (!mem
->dirty_bmap
) {
575 /* Allocate on the first log_sync, once and for all */
576 kvm_memslot_init_dirty_bitmap(mem
);
579 d
.dirty_bitmap
= mem
->dirty_bmap
;
580 d
.slot
= mem
->slot
| (kml
->as_id
<< 16);
581 if (kvm_vm_ioctl(s
, KVM_GET_DIRTY_LOG
, &d
) == -1) {
582 DPRINTF("ioctl failed %d\n", errno
);
587 subsection
.offset_within_region
+= slot_offset
;
588 subsection
.size
= int128_make64(slot_size
);
589 kvm_get_dirty_pages_log_range(&subsection
, d
.dirty_bitmap
);
591 slot_offset
+= slot_size
;
592 start_addr
+= slot_size
;
599 /* Alignment requirement for KVM_CLEAR_DIRTY_LOG - 64 pages */
600 #define KVM_CLEAR_LOG_SHIFT 6
601 #define KVM_CLEAR_LOG_ALIGN (qemu_real_host_page_size << KVM_CLEAR_LOG_SHIFT)
602 #define KVM_CLEAR_LOG_MASK (-KVM_CLEAR_LOG_ALIGN)
604 static int kvm_log_clear_one_slot(KVMSlot
*mem
, int as_id
, uint64_t start
,
607 KVMState
*s
= kvm_state
;
608 uint64_t end
, bmap_start
, start_delta
, bmap_npages
;
609 struct kvm_clear_dirty_log d
;
610 unsigned long *bmap_clear
= NULL
, psize
= qemu_real_host_page_size
;
614 * We need to extend either the start or the size or both to
615 * satisfy the KVM interface requirement. Firstly, do the start
616 * page alignment on 64 host pages
618 bmap_start
= start
& KVM_CLEAR_LOG_MASK
;
619 start_delta
= start
- bmap_start
;
623 * The kernel interface has restriction on the size too, that either:
625 * (1) the size is 64 host pages aligned (just like the start), or
626 * (2) the size fills up until the end of the KVM memslot.
628 bmap_npages
= DIV_ROUND_UP(size
+ start_delta
, KVM_CLEAR_LOG_ALIGN
)
629 << KVM_CLEAR_LOG_SHIFT
;
630 end
= mem
->memory_size
/ psize
;
631 if (bmap_npages
> end
- bmap_start
) {
632 bmap_npages
= end
- bmap_start
;
634 start_delta
/= psize
;
637 * Prepare the bitmap to clear dirty bits. Here we must guarantee
638 * that we won't clear any unknown dirty bits otherwise we might
639 * accidentally clear some set bits which are not yet synced from
640 * the kernel into QEMU's bitmap, then we'll lose track of the
641 * guest modifications upon those pages (which can directly lead
642 * to guest data loss or panic after migration).
644 * Layout of the KVMSlot.dirty_bmap:
646 * |<-------- bmap_npages -----------..>|
649 * |----------------|-------------|------------------|------------|
652 * start bmap_start (start) end
653 * of memslot of memslot
655 * [1] bmap_npages can be aligned to either 64 pages or the end of slot
658 assert(bmap_start
% BITS_PER_LONG
== 0);
659 /* We should never do log_clear before log_sync */
660 assert(mem
->dirty_bmap
);
662 /* Slow path - we need to manipulate a temp bitmap */
663 bmap_clear
= bitmap_new(bmap_npages
);
664 bitmap_copy_with_src_offset(bmap_clear
, mem
->dirty_bmap
,
665 bmap_start
, start_delta
+ size
/ psize
);
667 * We need to fill the holes at start because that was not
668 * specified by the caller and we extended the bitmap only for
671 bitmap_clear(bmap_clear
, 0, start_delta
);
672 d
.dirty_bitmap
= bmap_clear
;
674 /* Fast path - start address aligns well with BITS_PER_LONG */
675 d
.dirty_bitmap
= mem
->dirty_bmap
+ BIT_WORD(bmap_start
);
678 d
.first_page
= bmap_start
;
679 /* It should never overflow. If it happens, say something */
680 assert(bmap_npages
<= UINT32_MAX
);
681 d
.num_pages
= bmap_npages
;
682 d
.slot
= mem
->slot
| (as_id
<< 16);
684 if (kvm_vm_ioctl(s
, KVM_CLEAR_DIRTY_LOG
, &d
) == -1) {
686 error_report("%s: KVM_CLEAR_DIRTY_LOG failed, slot=%d, "
687 "start=0x%"PRIx64
", size=0x%"PRIx32
", errno=%d",
688 __func__
, d
.slot
, (uint64_t)d
.first_page
,
689 (uint32_t)d
.num_pages
, ret
);
692 trace_kvm_clear_dirty_log(d
.slot
, d
.first_page
, d
.num_pages
);
696 * After we have updated the remote dirty bitmap, we update the
697 * cached bitmap as well for the memslot, then if another user
698 * clears the same region we know we shouldn't clear it again on
699 * the remote otherwise it's data loss as well.
701 bitmap_clear(mem
->dirty_bmap
, bmap_start
+ start_delta
,
703 /* This handles the NULL case well */
710 * kvm_physical_log_clear - Clear the kernel's dirty bitmap for range
712 * NOTE: this will be a no-op if we haven't enabled manual dirty log
713 * protection in the host kernel because in that case this operation
714 * will be done within log_sync().
716 * @kml: the kvm memory listener
717 * @section: the memory range to clear dirty bitmap
719 static int kvm_physical_log_clear(KVMMemoryListener
*kml
,
720 MemoryRegionSection
*section
)
722 KVMState
*s
= kvm_state
;
723 uint64_t start
, size
, offset
, count
;
727 if (!s
->manual_dirty_log_protect
) {
728 /* No need to do explicit clear */
732 start
= section
->offset_within_address_space
;
733 size
= int128_get64(section
->size
);
736 /* Nothing more we can do... */
742 for (i
= 0; i
< s
->nr_slots
; i
++) {
743 mem
= &kml
->slots
[i
];
744 /* Discard slots that are empty or do not overlap the section */
745 if (!mem
->memory_size
||
746 mem
->start_addr
> start
+ size
- 1 ||
747 start
> mem
->start_addr
+ mem
->memory_size
- 1) {
751 if (start
>= mem
->start_addr
) {
752 /* The slot starts before section or is aligned to it. */
753 offset
= start
- mem
->start_addr
;
754 count
= MIN(mem
->memory_size
- offset
, size
);
756 /* The slot starts after section. */
758 count
= MIN(mem
->memory_size
, size
- (mem
->start_addr
- start
));
760 ret
= kvm_log_clear_one_slot(mem
, kml
->as_id
, offset
, count
);
766 kvm_slots_unlock(kml
);
771 static void kvm_coalesce_mmio_region(MemoryListener
*listener
,
772 MemoryRegionSection
*secion
,
773 hwaddr start
, hwaddr size
)
775 KVMState
*s
= kvm_state
;
777 if (s
->coalesced_mmio
) {
778 struct kvm_coalesced_mmio_zone zone
;
784 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
788 static void kvm_uncoalesce_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_UNREGISTER_COALESCED_MMIO
, &zone
);
805 static void kvm_coalesce_pio_add(MemoryListener
*listener
,
806 MemoryRegionSection
*section
,
807 hwaddr start
, hwaddr size
)
809 KVMState
*s
= kvm_state
;
811 if (s
->coalesced_pio
) {
812 struct kvm_coalesced_mmio_zone zone
;
818 (void)kvm_vm_ioctl(s
, KVM_REGISTER_COALESCED_MMIO
, &zone
);
822 static void kvm_coalesce_pio_del(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_UNREGISTER_COALESCED_MMIO
, &zone
);
839 static MemoryListener kvm_coalesced_pio_listener
= {
840 .coalesced_io_add
= kvm_coalesce_pio_add
,
841 .coalesced_io_del
= kvm_coalesce_pio_del
,
844 int kvm_check_extension(KVMState
*s
, unsigned int extension
)
848 ret
= kvm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
856 int kvm_vm_check_extension(KVMState
*s
, unsigned int extension
)
860 ret
= kvm_vm_ioctl(s
, KVM_CHECK_EXTENSION
, extension
);
862 /* VM wide version not implemented, use global one instead */
863 ret
= kvm_check_extension(s
, extension
);
869 static uint32_t adjust_ioeventfd_endianness(uint32_t val
, uint32_t size
)
871 #if defined(HOST_WORDS_BIGENDIAN) != defined(TARGET_WORDS_BIGENDIAN)
872 /* The kernel expects ioeventfd values in HOST_WORDS_BIGENDIAN
873 * endianness, but the memory core hands them in target endianness.
874 * For example, PPC is always treated as big-endian even if running
875 * on KVM and on PPC64LE. Correct here.
889 static int kvm_set_ioeventfd_mmio(int fd
, hwaddr addr
, uint32_t val
,
890 bool assign
, uint32_t size
, bool datamatch
)
893 struct kvm_ioeventfd iofd
= {
894 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
901 trace_kvm_set_ioeventfd_mmio(fd
, (uint64_t)addr
, val
, assign
, size
,
903 if (!kvm_enabled()) {
908 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
911 iofd
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
914 ret
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &iofd
);
923 static int kvm_set_ioeventfd_pio(int fd
, uint16_t addr
, uint16_t val
,
924 bool assign
, uint32_t size
, bool datamatch
)
926 struct kvm_ioeventfd kick
= {
927 .datamatch
= datamatch
? adjust_ioeventfd_endianness(val
, size
) : 0,
929 .flags
= KVM_IOEVENTFD_FLAG_PIO
,
934 trace_kvm_set_ioeventfd_pio(fd
, addr
, val
, assign
, size
, datamatch
);
935 if (!kvm_enabled()) {
939 kick
.flags
|= KVM_IOEVENTFD_FLAG_DATAMATCH
;
942 kick
.flags
|= KVM_IOEVENTFD_FLAG_DEASSIGN
;
944 r
= kvm_vm_ioctl(kvm_state
, KVM_IOEVENTFD
, &kick
);
952 static int kvm_check_many_ioeventfds(void)
954 /* Userspace can use ioeventfd for io notification. This requires a host
955 * that supports eventfd(2) and an I/O thread; since eventfd does not
956 * support SIGIO it cannot interrupt the vcpu.
958 * Older kernels have a 6 device limit on the KVM io bus. Find out so we
959 * can avoid creating too many ioeventfds.
961 #if defined(CONFIG_EVENTFD)
964 for (i
= 0; i
< ARRAY_SIZE(ioeventfds
); i
++) {
965 ioeventfds
[i
] = eventfd(0, EFD_CLOEXEC
);
966 if (ioeventfds
[i
] < 0) {
969 ret
= kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, true, 2, true);
971 close(ioeventfds
[i
]);
976 /* Decide whether many devices are supported or not */
977 ret
= i
== ARRAY_SIZE(ioeventfds
);
980 kvm_set_ioeventfd_pio(ioeventfds
[i
], 0, i
, false, 2, true);
981 close(ioeventfds
[i
]);
989 static const KVMCapabilityInfo
*
990 kvm_check_extension_list(KVMState
*s
, const KVMCapabilityInfo
*list
)
993 if (!kvm_check_extension(s
, list
->value
)) {
1001 void kvm_set_max_memslot_size(hwaddr max_slot_size
)
1004 ROUND_UP(max_slot_size
, qemu_real_host_page_size
) == max_slot_size
1006 kvm_max_slot_size
= max_slot_size
;
1009 static void kvm_set_phys_mem(KVMMemoryListener
*kml
,
1010 MemoryRegionSection
*section
, bool add
)
1014 MemoryRegion
*mr
= section
->mr
;
1015 bool writeable
= !mr
->readonly
&& !mr
->rom_device
;
1016 hwaddr start_addr
, size
, slot_size
;
1019 if (!memory_region_is_ram(mr
)) {
1020 if (writeable
|| !kvm_readonly_mem_allowed
) {
1022 } else if (!mr
->romd_mode
) {
1023 /* If the memory device is not in romd_mode, then we actually want
1024 * to remove the kvm memory slot so all accesses will trap. */
1029 size
= kvm_align_section(section
, &start_addr
);
1034 /* use aligned delta to align the ram address */
1035 ram
= memory_region_get_ram_ptr(mr
) + section
->offset_within_region
+
1036 (start_addr
- section
->offset_within_address_space
);
1038 kvm_slots_lock(kml
);
1042 slot_size
= MIN(kvm_max_slot_size
, size
);
1043 mem
= kvm_lookup_matching_slot(kml
, start_addr
, slot_size
);
1047 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
1048 kvm_physical_sync_dirty_bitmap(kml
, section
);
1051 /* unregister the slot */
1052 g_free(mem
->dirty_bmap
);
1053 mem
->dirty_bmap
= NULL
;
1054 mem
->memory_size
= 0;
1056 err
= kvm_set_user_memory_region(kml
, mem
, false);
1058 fprintf(stderr
, "%s: error unregistering slot: %s\n",
1059 __func__
, strerror(-err
));
1062 start_addr
+= slot_size
;
1068 /* register the new slot */
1070 slot_size
= MIN(kvm_max_slot_size
, size
);
1071 mem
= kvm_alloc_slot(kml
);
1072 mem
->memory_size
= slot_size
;
1073 mem
->start_addr
= start_addr
;
1075 mem
->flags
= kvm_mem_flags(mr
);
1077 if (mem
->flags
& KVM_MEM_LOG_DIRTY_PAGES
) {
1079 * Reallocate the bmap; it means it doesn't disappear in
1080 * middle of a migrate.
1082 kvm_memslot_init_dirty_bitmap(mem
);
1084 err
= kvm_set_user_memory_region(kml
, mem
, true);
1086 fprintf(stderr
, "%s: error registering slot: %s\n", __func__
,
1090 start_addr
+= slot_size
;
1096 kvm_slots_unlock(kml
);
1099 static void kvm_region_add(MemoryListener
*listener
,
1100 MemoryRegionSection
*section
)
1102 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1104 memory_region_ref(section
->mr
);
1105 kvm_set_phys_mem(kml
, section
, true);
1108 static void kvm_region_del(MemoryListener
*listener
,
1109 MemoryRegionSection
*section
)
1111 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1113 kvm_set_phys_mem(kml
, section
, false);
1114 memory_region_unref(section
->mr
);
1117 static void kvm_log_sync(MemoryListener
*listener
,
1118 MemoryRegionSection
*section
)
1120 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1123 kvm_slots_lock(kml
);
1124 r
= kvm_physical_sync_dirty_bitmap(kml
, section
);
1125 kvm_slots_unlock(kml
);
1131 static void kvm_log_clear(MemoryListener
*listener
,
1132 MemoryRegionSection
*section
)
1134 KVMMemoryListener
*kml
= container_of(listener
, KVMMemoryListener
, listener
);
1137 r
= kvm_physical_log_clear(kml
, section
);
1139 error_report_once("%s: kvm log clear failed: mr=%s "
1140 "offset=%"HWADDR_PRIx
" size=%"PRIx64
, __func__
,
1141 section
->mr
->name
, section
->offset_within_region
,
1142 int128_get64(section
->size
));
1147 static void kvm_mem_ioeventfd_add(MemoryListener
*listener
,
1148 MemoryRegionSection
*section
,
1149 bool match_data
, uint64_t data
,
1152 int fd
= event_notifier_get_fd(e
);
1155 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1156 data
, true, int128_get64(section
->size
),
1159 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1160 __func__
, strerror(-r
), -r
);
1165 static void kvm_mem_ioeventfd_del(MemoryListener
*listener
,
1166 MemoryRegionSection
*section
,
1167 bool match_data
, uint64_t data
,
1170 int fd
= event_notifier_get_fd(e
);
1173 r
= kvm_set_ioeventfd_mmio(fd
, section
->offset_within_address_space
,
1174 data
, false, int128_get64(section
->size
),
1177 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1178 __func__
, strerror(-r
), -r
);
1183 static void kvm_io_ioeventfd_add(MemoryListener
*listener
,
1184 MemoryRegionSection
*section
,
1185 bool match_data
, uint64_t data
,
1188 int fd
= event_notifier_get_fd(e
);
1191 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1192 data
, true, int128_get64(section
->size
),
1195 fprintf(stderr
, "%s: error adding ioeventfd: %s (%d)\n",
1196 __func__
, strerror(-r
), -r
);
1201 static void kvm_io_ioeventfd_del(MemoryListener
*listener
,
1202 MemoryRegionSection
*section
,
1203 bool match_data
, uint64_t data
,
1207 int fd
= event_notifier_get_fd(e
);
1210 r
= kvm_set_ioeventfd_pio(fd
, section
->offset_within_address_space
,
1211 data
, false, int128_get64(section
->size
),
1214 fprintf(stderr
, "%s: error deleting ioeventfd: %s (%d)\n",
1215 __func__
, strerror(-r
), -r
);
1220 void kvm_memory_listener_register(KVMState
*s
, KVMMemoryListener
*kml
,
1221 AddressSpace
*as
, int as_id
)
1225 qemu_mutex_init(&kml
->slots_lock
);
1226 kml
->slots
= g_malloc0(s
->nr_slots
* sizeof(KVMSlot
));
1229 for (i
= 0; i
< s
->nr_slots
; i
++) {
1230 kml
->slots
[i
].slot
= i
;
1233 kml
->listener
.region_add
= kvm_region_add
;
1234 kml
->listener
.region_del
= kvm_region_del
;
1235 kml
->listener
.log_start
= kvm_log_start
;
1236 kml
->listener
.log_stop
= kvm_log_stop
;
1237 kml
->listener
.log_sync
= kvm_log_sync
;
1238 kml
->listener
.log_clear
= kvm_log_clear
;
1239 kml
->listener
.priority
= 10;
1241 memory_listener_register(&kml
->listener
, as
);
1243 for (i
= 0; i
< s
->nr_as
; ++i
) {
1252 static MemoryListener kvm_io_listener
= {
1253 .eventfd_add
= kvm_io_ioeventfd_add
,
1254 .eventfd_del
= kvm_io_ioeventfd_del
,
1258 int kvm_set_irq(KVMState
*s
, int irq
, int level
)
1260 struct kvm_irq_level event
;
1263 assert(kvm_async_interrupts_enabled());
1265 event
.level
= level
;
1267 ret
= kvm_vm_ioctl(s
, s
->irq_set_ioctl
, &event
);
1269 perror("kvm_set_irq");
1273 return (s
->irq_set_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
1276 #ifdef KVM_CAP_IRQ_ROUTING
1277 typedef struct KVMMSIRoute
{
1278 struct kvm_irq_routing_entry kroute
;
1279 QTAILQ_ENTRY(KVMMSIRoute
) entry
;
1282 static void set_gsi(KVMState
*s
, unsigned int gsi
)
1284 set_bit(gsi
, s
->used_gsi_bitmap
);
1287 static void clear_gsi(KVMState
*s
, unsigned int gsi
)
1289 clear_bit(gsi
, s
->used_gsi_bitmap
);
1292 void kvm_init_irq_routing(KVMState
*s
)
1296 gsi_count
= kvm_check_extension(s
, KVM_CAP_IRQ_ROUTING
) - 1;
1297 if (gsi_count
> 0) {
1298 /* Round up so we can search ints using ffs */
1299 s
->used_gsi_bitmap
= bitmap_new(gsi_count
);
1300 s
->gsi_count
= gsi_count
;
1303 s
->irq_routes
= g_malloc0(sizeof(*s
->irq_routes
));
1304 s
->nr_allocated_irq_routes
= 0;
1306 if (!kvm_direct_msi_allowed
) {
1307 for (i
= 0; i
< KVM_MSI_HASHTAB_SIZE
; i
++) {
1308 QTAILQ_INIT(&s
->msi_hashtab
[i
]);
1312 kvm_arch_init_irq_routing(s
);
1315 void kvm_irqchip_commit_routes(KVMState
*s
)
1319 if (kvm_gsi_direct_mapping()) {
1323 if (!kvm_gsi_routing_enabled()) {
1327 s
->irq_routes
->flags
= 0;
1328 trace_kvm_irqchip_commit_routes();
1329 ret
= kvm_vm_ioctl(s
, KVM_SET_GSI_ROUTING
, s
->irq_routes
);
1333 static void kvm_add_routing_entry(KVMState
*s
,
1334 struct kvm_irq_routing_entry
*entry
)
1336 struct kvm_irq_routing_entry
*new;
1339 if (s
->irq_routes
->nr
== s
->nr_allocated_irq_routes
) {
1340 n
= s
->nr_allocated_irq_routes
* 2;
1344 size
= sizeof(struct kvm_irq_routing
);
1345 size
+= n
* sizeof(*new);
1346 s
->irq_routes
= g_realloc(s
->irq_routes
, size
);
1347 s
->nr_allocated_irq_routes
= n
;
1349 n
= s
->irq_routes
->nr
++;
1350 new = &s
->irq_routes
->entries
[n
];
1354 set_gsi(s
, entry
->gsi
);
1357 static int kvm_update_routing_entry(KVMState
*s
,
1358 struct kvm_irq_routing_entry
*new_entry
)
1360 struct kvm_irq_routing_entry
*entry
;
1363 for (n
= 0; n
< s
->irq_routes
->nr
; n
++) {
1364 entry
= &s
->irq_routes
->entries
[n
];
1365 if (entry
->gsi
!= new_entry
->gsi
) {
1369 if(!memcmp(entry
, new_entry
, sizeof *entry
)) {
1373 *entry
= *new_entry
;
1381 void kvm_irqchip_add_irq_route(KVMState
*s
, int irq
, int irqchip
, int pin
)
1383 struct kvm_irq_routing_entry e
= {};
1385 assert(pin
< s
->gsi_count
);
1388 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
1390 e
.u
.irqchip
.irqchip
= irqchip
;
1391 e
.u
.irqchip
.pin
= pin
;
1392 kvm_add_routing_entry(s
, &e
);
1395 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1397 struct kvm_irq_routing_entry
*e
;
1400 if (kvm_gsi_direct_mapping()) {
1404 for (i
= 0; i
< s
->irq_routes
->nr
; i
++) {
1405 e
= &s
->irq_routes
->entries
[i
];
1406 if (e
->gsi
== virq
) {
1407 s
->irq_routes
->nr
--;
1408 *e
= s
->irq_routes
->entries
[s
->irq_routes
->nr
];
1412 kvm_arch_release_virq_post(virq
);
1413 trace_kvm_irqchip_release_virq(virq
);
1416 void kvm_irqchip_add_change_notifier(Notifier
*n
)
1418 notifier_list_add(&kvm_irqchip_change_notifiers
, n
);
1421 void kvm_irqchip_remove_change_notifier(Notifier
*n
)
1426 void kvm_irqchip_change_notify(void)
1428 notifier_list_notify(&kvm_irqchip_change_notifiers
, NULL
);
1431 static unsigned int kvm_hash_msi(uint32_t data
)
1433 /* This is optimized for IA32 MSI layout. However, no other arch shall
1434 * repeat the mistake of not providing a direct MSI injection API. */
1438 static void kvm_flush_dynamic_msi_routes(KVMState
*s
)
1440 KVMMSIRoute
*route
, *next
;
1443 for (hash
= 0; hash
< KVM_MSI_HASHTAB_SIZE
; hash
++) {
1444 QTAILQ_FOREACH_SAFE(route
, &s
->msi_hashtab
[hash
], entry
, next
) {
1445 kvm_irqchip_release_virq(s
, route
->kroute
.gsi
);
1446 QTAILQ_REMOVE(&s
->msi_hashtab
[hash
], route
, entry
);
1452 static int kvm_irqchip_get_virq(KVMState
*s
)
1457 * PIC and IOAPIC share the first 16 GSI numbers, thus the available
1458 * GSI numbers are more than the number of IRQ route. Allocating a GSI
1459 * number can succeed even though a new route entry cannot be added.
1460 * When this happens, flush dynamic MSI entries to free IRQ route entries.
1462 if (!kvm_direct_msi_allowed
&& s
->irq_routes
->nr
== s
->gsi_count
) {
1463 kvm_flush_dynamic_msi_routes(s
);
1466 /* Return the lowest unused GSI in the bitmap */
1467 next_virq
= find_first_zero_bit(s
->used_gsi_bitmap
, s
->gsi_count
);
1468 if (next_virq
>= s
->gsi_count
) {
1475 static KVMMSIRoute
*kvm_lookup_msi_route(KVMState
*s
, MSIMessage msg
)
1477 unsigned int hash
= kvm_hash_msi(msg
.data
);
1480 QTAILQ_FOREACH(route
, &s
->msi_hashtab
[hash
], entry
) {
1481 if (route
->kroute
.u
.msi
.address_lo
== (uint32_t)msg
.address
&&
1482 route
->kroute
.u
.msi
.address_hi
== (msg
.address
>> 32) &&
1483 route
->kroute
.u
.msi
.data
== le32_to_cpu(msg
.data
)) {
1490 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1495 if (kvm_direct_msi_allowed
) {
1496 msi
.address_lo
= (uint32_t)msg
.address
;
1497 msi
.address_hi
= msg
.address
>> 32;
1498 msi
.data
= le32_to_cpu(msg
.data
);
1500 memset(msi
.pad
, 0, sizeof(msi
.pad
));
1502 return kvm_vm_ioctl(s
, KVM_SIGNAL_MSI
, &msi
);
1505 route
= kvm_lookup_msi_route(s
, msg
);
1509 virq
= kvm_irqchip_get_virq(s
);
1514 route
= g_malloc0(sizeof(KVMMSIRoute
));
1515 route
->kroute
.gsi
= virq
;
1516 route
->kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1517 route
->kroute
.flags
= 0;
1518 route
->kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1519 route
->kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1520 route
->kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1522 kvm_add_routing_entry(s
, &route
->kroute
);
1523 kvm_irqchip_commit_routes(s
);
1525 QTAILQ_INSERT_TAIL(&s
->msi_hashtab
[kvm_hash_msi(msg
.data
)], route
,
1529 assert(route
->kroute
.type
== KVM_IRQ_ROUTING_MSI
);
1531 return kvm_set_irq(s
, route
->kroute
.gsi
, 1);
1534 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1536 struct kvm_irq_routing_entry kroute
= {};
1538 MSIMessage msg
= {0, 0};
1540 if (pci_available
&& dev
) {
1541 msg
= pci_get_msi_message(dev
, vector
);
1544 if (kvm_gsi_direct_mapping()) {
1545 return kvm_arch_msi_data_to_gsi(msg
.data
);
1548 if (!kvm_gsi_routing_enabled()) {
1552 virq
= kvm_irqchip_get_virq(s
);
1558 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1560 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1561 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1562 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1563 if (pci_available
&& kvm_msi_devid_required()) {
1564 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1565 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1567 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1568 kvm_irqchip_release_virq(s
, virq
);
1572 trace_kvm_irqchip_add_msi_route(dev
? dev
->name
: (char *)"N/A",
1575 kvm_add_routing_entry(s
, &kroute
);
1576 kvm_arch_add_msi_route_post(&kroute
, vector
, dev
);
1577 kvm_irqchip_commit_routes(s
);
1582 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
,
1585 struct kvm_irq_routing_entry kroute
= {};
1587 if (kvm_gsi_direct_mapping()) {
1591 if (!kvm_irqchip_in_kernel()) {
1596 kroute
.type
= KVM_IRQ_ROUTING_MSI
;
1598 kroute
.u
.msi
.address_lo
= (uint32_t)msg
.address
;
1599 kroute
.u
.msi
.address_hi
= msg
.address
>> 32;
1600 kroute
.u
.msi
.data
= le32_to_cpu(msg
.data
);
1601 if (pci_available
&& kvm_msi_devid_required()) {
1602 kroute
.flags
= KVM_MSI_VALID_DEVID
;
1603 kroute
.u
.msi
.devid
= pci_requester_id(dev
);
1605 if (kvm_arch_fixup_msi_route(&kroute
, msg
.address
, msg
.data
, dev
)) {
1609 trace_kvm_irqchip_update_msi_route(virq
);
1611 return kvm_update_routing_entry(s
, &kroute
);
1614 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int rfd
, int virq
,
1617 struct kvm_irqfd irqfd
= {
1620 .flags
= assign
? 0 : KVM_IRQFD_FLAG_DEASSIGN
,
1624 irqfd
.flags
|= KVM_IRQFD_FLAG_RESAMPLE
;
1625 irqfd
.resamplefd
= rfd
;
1628 if (!kvm_irqfds_enabled()) {
1632 return kvm_vm_ioctl(s
, KVM_IRQFD
, &irqfd
);
1635 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1637 struct kvm_irq_routing_entry kroute
= {};
1640 if (!kvm_gsi_routing_enabled()) {
1644 virq
= kvm_irqchip_get_virq(s
);
1650 kroute
.type
= KVM_IRQ_ROUTING_S390_ADAPTER
;
1652 kroute
.u
.adapter
.summary_addr
= adapter
->summary_addr
;
1653 kroute
.u
.adapter
.ind_addr
= adapter
->ind_addr
;
1654 kroute
.u
.adapter
.summary_offset
= adapter
->summary_offset
;
1655 kroute
.u
.adapter
.ind_offset
= adapter
->ind_offset
;
1656 kroute
.u
.adapter
.adapter_id
= adapter
->adapter_id
;
1658 kvm_add_routing_entry(s
, &kroute
);
1663 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1665 struct kvm_irq_routing_entry kroute
= {};
1668 if (!kvm_gsi_routing_enabled()) {
1671 if (!kvm_check_extension(s
, KVM_CAP_HYPERV_SYNIC
)) {
1674 virq
= kvm_irqchip_get_virq(s
);
1680 kroute
.type
= KVM_IRQ_ROUTING_HV_SINT
;
1682 kroute
.u
.hv_sint
.vcpu
= vcpu
;
1683 kroute
.u
.hv_sint
.sint
= sint
;
1685 kvm_add_routing_entry(s
, &kroute
);
1686 kvm_irqchip_commit_routes(s
);
1691 #else /* !KVM_CAP_IRQ_ROUTING */
1693 void kvm_init_irq_routing(KVMState
*s
)
1697 void kvm_irqchip_release_virq(KVMState
*s
, int virq
)
1701 int kvm_irqchip_send_msi(KVMState
*s
, MSIMessage msg
)
1706 int kvm_irqchip_add_msi_route(KVMState
*s
, int vector
, PCIDevice
*dev
)
1711 int kvm_irqchip_add_adapter_route(KVMState
*s
, AdapterInfo
*adapter
)
1716 int kvm_irqchip_add_hv_sint_route(KVMState
*s
, uint32_t vcpu
, uint32_t sint
)
1721 static int kvm_irqchip_assign_irqfd(KVMState
*s
, int fd
, int virq
, bool assign
)
1726 int kvm_irqchip_update_msi_route(KVMState
*s
, int virq
, MSIMessage msg
)
1730 #endif /* !KVM_CAP_IRQ_ROUTING */
1732 int kvm_irqchip_add_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1733 EventNotifier
*rn
, int virq
)
1735 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
),
1736 rn
? event_notifier_get_fd(rn
) : -1, virq
, true);
1739 int kvm_irqchip_remove_irqfd_notifier_gsi(KVMState
*s
, EventNotifier
*n
,
1742 return kvm_irqchip_assign_irqfd(s
, event_notifier_get_fd(n
), -1, virq
,
1746 int kvm_irqchip_add_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1747 EventNotifier
*rn
, qemu_irq irq
)
1750 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1755 return kvm_irqchip_add_irqfd_notifier_gsi(s
, n
, rn
, GPOINTER_TO_INT(gsi
));
1758 int kvm_irqchip_remove_irqfd_notifier(KVMState
*s
, EventNotifier
*n
,
1762 gboolean found
= g_hash_table_lookup_extended(s
->gsimap
, irq
, &key
, &gsi
);
1767 return kvm_irqchip_remove_irqfd_notifier_gsi(s
, n
, GPOINTER_TO_INT(gsi
));
1770 void kvm_irqchip_set_qemuirq_gsi(KVMState
*s
, qemu_irq irq
, int gsi
)
1772 g_hash_table_insert(s
->gsimap
, irq
, GINT_TO_POINTER(gsi
));
1775 static void kvm_irqchip_create(MachineState
*machine
, KVMState
*s
)
1779 if (kvm_check_extension(s
, KVM_CAP_IRQCHIP
)) {
1781 } else if (kvm_check_extension(s
, KVM_CAP_S390_IRQCHIP
)) {
1782 ret
= kvm_vm_enable_cap(s
, KVM_CAP_S390_IRQCHIP
, 0);
1784 fprintf(stderr
, "Enable kernel irqchip failed: %s\n", strerror(-ret
));
1791 /* First probe and see if there's a arch-specific hook to create the
1792 * in-kernel irqchip for us */
1793 ret
= kvm_arch_irqchip_create(machine
, s
);
1795 if (machine_kernel_irqchip_split(machine
)) {
1796 perror("Split IRQ chip mode not supported.");
1799 ret
= kvm_vm_ioctl(s
, KVM_CREATE_IRQCHIP
);
1803 fprintf(stderr
, "Create kernel irqchip failed: %s\n", strerror(-ret
));
1807 kvm_kernel_irqchip
= true;
1808 /* If we have an in-kernel IRQ chip then we must have asynchronous
1809 * interrupt delivery (though the reverse is not necessarily true)
1811 kvm_async_interrupts_allowed
= true;
1812 kvm_halt_in_kernel_allowed
= true;
1814 kvm_init_irq_routing(s
);
1816 s
->gsimap
= g_hash_table_new(g_direct_hash
, g_direct_equal
);
1819 /* Find number of supported CPUs using the recommended
1820 * procedure from the kernel API documentation to cope with
1821 * older kernels that may be missing capabilities.
1823 static int kvm_recommended_vcpus(KVMState
*s
)
1825 int ret
= kvm_vm_check_extension(s
, KVM_CAP_NR_VCPUS
);
1826 return (ret
) ? ret
: 4;
1829 static int kvm_max_vcpus(KVMState
*s
)
1831 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPUS
);
1832 return (ret
) ? ret
: kvm_recommended_vcpus(s
);
1835 static int kvm_max_vcpu_id(KVMState
*s
)
1837 int ret
= kvm_check_extension(s
, KVM_CAP_MAX_VCPU_ID
);
1838 return (ret
) ? ret
: kvm_max_vcpus(s
);
1841 bool kvm_vcpu_id_is_valid(int vcpu_id
)
1843 KVMState
*s
= KVM_STATE(current_machine
->accelerator
);
1844 return vcpu_id
>= 0 && vcpu_id
< kvm_max_vcpu_id(s
);
1847 static int kvm_init(MachineState
*ms
)
1849 MachineClass
*mc
= MACHINE_GET_CLASS(ms
);
1850 static const char upgrade_note
[] =
1851 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1852 "(see http://sourceforge.net/projects/kvm).\n";
1857 { "SMP", ms
->smp
.cpus
},
1858 { "hotpluggable", ms
->smp
.max_cpus
},
1861 int soft_vcpus_limit
, hard_vcpus_limit
;
1863 const KVMCapabilityInfo
*missing_cap
;
1866 const char *kvm_type
;
1868 s
= KVM_STATE(ms
->accelerator
);
1871 * On systems where the kernel can support different base page
1872 * sizes, host page size may be different from TARGET_PAGE_SIZE,
1873 * even with KVM. TARGET_PAGE_SIZE is assumed to be the minimum
1874 * page size for the system though.
1876 assert(TARGET_PAGE_SIZE
<= qemu_real_host_page_size
);
1880 #ifdef KVM_CAP_SET_GUEST_DEBUG
1881 QTAILQ_INIT(&s
->kvm_sw_breakpoints
);
1883 QLIST_INIT(&s
->kvm_parked_vcpus
);
1885 s
->fd
= qemu_open("/dev/kvm", O_RDWR
);
1887 fprintf(stderr
, "Could not access KVM kernel module: %m\n");
1892 ret
= kvm_ioctl(s
, KVM_GET_API_VERSION
, 0);
1893 if (ret
< KVM_API_VERSION
) {
1897 fprintf(stderr
, "kvm version too old\n");
1901 if (ret
> KVM_API_VERSION
) {
1903 fprintf(stderr
, "kvm version not supported\n");
1907 kvm_immediate_exit
= kvm_check_extension(s
, KVM_CAP_IMMEDIATE_EXIT
);
1908 s
->nr_slots
= kvm_check_extension(s
, KVM_CAP_NR_MEMSLOTS
);
1910 /* If unspecified, use the default value */
1915 s
->nr_as
= kvm_check_extension(s
, KVM_CAP_MULTI_ADDRESS_SPACE
);
1916 if (s
->nr_as
<= 1) {
1919 s
->as
= g_new0(struct KVMAs
, s
->nr_as
);
1921 kvm_type
= qemu_opt_get(qemu_get_machine_opts(), "kvm-type");
1923 type
= mc
->kvm_type(ms
, kvm_type
);
1924 } else if (kvm_type
) {
1926 fprintf(stderr
, "Invalid argument kvm-type=%s\n", kvm_type
);
1931 ret
= kvm_ioctl(s
, KVM_CREATE_VM
, type
);
1932 } while (ret
== -EINTR
);
1935 fprintf(stderr
, "ioctl(KVM_CREATE_VM) failed: %d %s\n", -ret
,
1939 if (ret
== -EINVAL
) {
1941 "Host kernel setup problem detected. Please verify:\n");
1942 fprintf(stderr
, "- for kernels supporting the switch_amode or"
1943 " user_mode parameters, whether\n");
1945 " user space is running in primary address space\n");
1947 "- for kernels supporting the vm.allocate_pgste sysctl, "
1948 "whether it is enabled\n");
1956 /* check the vcpu limits */
1957 soft_vcpus_limit
= kvm_recommended_vcpus(s
);
1958 hard_vcpus_limit
= kvm_max_vcpus(s
);
1961 if (nc
->num
> soft_vcpus_limit
) {
1962 warn_report("Number of %s cpus requested (%d) exceeds "
1963 "the recommended cpus supported by KVM (%d)",
1964 nc
->name
, nc
->num
, soft_vcpus_limit
);
1966 if (nc
->num
> hard_vcpus_limit
) {
1967 fprintf(stderr
, "Number of %s cpus requested (%d) exceeds "
1968 "the maximum cpus supported by KVM (%d)\n",
1969 nc
->name
, nc
->num
, hard_vcpus_limit
);
1976 missing_cap
= kvm_check_extension_list(s
, kvm_required_capabilites
);
1979 kvm_check_extension_list(s
, kvm_arch_required_capabilities
);
1983 fprintf(stderr
, "kvm does not support %s\n%s",
1984 missing_cap
->name
, upgrade_note
);
1988 s
->coalesced_mmio
= kvm_check_extension(s
, KVM_CAP_COALESCED_MMIO
);
1989 s
->coalesced_pio
= s
->coalesced_mmio
&&
1990 kvm_check_extension(s
, KVM_CAP_COALESCED_PIO
);
1992 s
->manual_dirty_log_protect
=
1993 kvm_check_extension(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
);
1994 if (s
->manual_dirty_log_protect
) {
1995 ret
= kvm_vm_enable_cap(s
, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2
, 0, 1);
1997 warn_report("Trying to enable KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2 "
1998 "but failed. Falling back to the legacy mode. ");
1999 s
->manual_dirty_log_protect
= false;
2003 #ifdef KVM_CAP_VCPU_EVENTS
2004 s
->vcpu_events
= kvm_check_extension(s
, KVM_CAP_VCPU_EVENTS
);
2007 s
->robust_singlestep
=
2008 kvm_check_extension(s
, KVM_CAP_X86_ROBUST_SINGLESTEP
);
2010 #ifdef KVM_CAP_DEBUGREGS
2011 s
->debugregs
= kvm_check_extension(s
, KVM_CAP_DEBUGREGS
);
2014 s
->max_nested_state_len
= kvm_check_extension(s
, KVM_CAP_NESTED_STATE
);
2016 #ifdef KVM_CAP_IRQ_ROUTING
2017 kvm_direct_msi_allowed
= (kvm_check_extension(s
, KVM_CAP_SIGNAL_MSI
) > 0);
2020 s
->intx_set_mask
= kvm_check_extension(s
, KVM_CAP_PCI_2_3
);
2022 s
->irq_set_ioctl
= KVM_IRQ_LINE
;
2023 if (kvm_check_extension(s
, KVM_CAP_IRQ_INJECT_STATUS
)) {
2024 s
->irq_set_ioctl
= KVM_IRQ_LINE_STATUS
;
2027 kvm_readonly_mem_allowed
=
2028 (kvm_check_extension(s
, KVM_CAP_READONLY_MEM
) > 0);
2030 kvm_eventfds_allowed
=
2031 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD
) > 0);
2033 kvm_irqfds_allowed
=
2034 (kvm_check_extension(s
, KVM_CAP_IRQFD
) > 0);
2036 kvm_resamplefds_allowed
=
2037 (kvm_check_extension(s
, KVM_CAP_IRQFD_RESAMPLE
) > 0);
2039 kvm_vm_attributes_allowed
=
2040 (kvm_check_extension(s
, KVM_CAP_VM_ATTRIBUTES
) > 0);
2042 kvm_ioeventfd_any_length_allowed
=
2043 (kvm_check_extension(s
, KVM_CAP_IOEVENTFD_ANY_LENGTH
) > 0);
2048 * if memory encryption object is specified then initialize the memory
2049 * encryption context.
2051 if (ms
->memory_encryption
) {
2052 kvm_state
->memcrypt_handle
= sev_guest_init(ms
->memory_encryption
);
2053 if (!kvm_state
->memcrypt_handle
) {
2058 kvm_state
->memcrypt_encrypt_data
= sev_encrypt_data
;
2061 ret
= kvm_arch_init(ms
, s
);
2066 if (machine_kernel_irqchip_allowed(ms
)) {
2067 kvm_irqchip_create(ms
, s
);
2070 if (kvm_eventfds_allowed
) {
2071 s
->memory_listener
.listener
.eventfd_add
= kvm_mem_ioeventfd_add
;
2072 s
->memory_listener
.listener
.eventfd_del
= kvm_mem_ioeventfd_del
;
2074 s
->memory_listener
.listener
.coalesced_io_add
= kvm_coalesce_mmio_region
;
2075 s
->memory_listener
.listener
.coalesced_io_del
= kvm_uncoalesce_mmio_region
;
2077 kvm_memory_listener_register(s
, &s
->memory_listener
,
2078 &address_space_memory
, 0);
2079 memory_listener_register(&kvm_io_listener
,
2081 memory_listener_register(&kvm_coalesced_pio_listener
,
2084 s
->many_ioeventfds
= kvm_check_many_ioeventfds();
2086 s
->sync_mmu
= !!kvm_vm_check_extension(kvm_state
, KVM_CAP_SYNC_MMU
);
2088 qemu_balloon_inhibit(true);
2101 g_free(s
->memory_listener
.slots
);
2106 void kvm_set_sigmask_len(KVMState
*s
, unsigned int sigmask_len
)
2108 s
->sigmask_len
= sigmask_len
;
2111 static void kvm_handle_io(uint16_t port
, MemTxAttrs attrs
, void *data
, int direction
,
2112 int size
, uint32_t count
)
2115 uint8_t *ptr
= data
;
2117 for (i
= 0; i
< count
; i
++) {
2118 address_space_rw(&address_space_io
, port
, attrs
,
2120 direction
== KVM_EXIT_IO_OUT
);
2125 static int kvm_handle_internal_error(CPUState
*cpu
, struct kvm_run
*run
)
2127 fprintf(stderr
, "KVM internal error. Suberror: %d\n",
2128 run
->internal
.suberror
);
2130 if (kvm_check_extension(kvm_state
, KVM_CAP_INTERNAL_ERROR_DATA
)) {
2133 for (i
= 0; i
< run
->internal
.ndata
; ++i
) {
2134 fprintf(stderr
, "extra data[%d]: %"PRIx64
"\n",
2135 i
, (uint64_t)run
->internal
.data
[i
]);
2138 if (run
->internal
.suberror
== KVM_INTERNAL_ERROR_EMULATION
) {
2139 fprintf(stderr
, "emulation failure\n");
2140 if (!kvm_arch_stop_on_emulation_error(cpu
)) {
2141 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2142 return EXCP_INTERRUPT
;
2145 /* FIXME: Should trigger a qmp message to let management know
2146 * something went wrong.
2151 void kvm_flush_coalesced_mmio_buffer(void)
2153 KVMState
*s
= kvm_state
;
2155 if (s
->coalesced_flush_in_progress
) {
2159 s
->coalesced_flush_in_progress
= true;
2161 if (s
->coalesced_mmio_ring
) {
2162 struct kvm_coalesced_mmio_ring
*ring
= s
->coalesced_mmio_ring
;
2163 while (ring
->first
!= ring
->last
) {
2164 struct kvm_coalesced_mmio
*ent
;
2166 ent
= &ring
->coalesced_mmio
[ring
->first
];
2168 if (ent
->pio
== 1) {
2169 address_space_rw(&address_space_io
, ent
->phys_addr
,
2170 MEMTXATTRS_UNSPECIFIED
, ent
->data
,
2173 cpu_physical_memory_write(ent
->phys_addr
, ent
->data
, ent
->len
);
2176 ring
->first
= (ring
->first
+ 1) % KVM_COALESCED_MMIO_MAX
;
2180 s
->coalesced_flush_in_progress
= false;
2183 static void do_kvm_cpu_synchronize_state(CPUState
*cpu
, run_on_cpu_data arg
)
2185 if (!cpu
->vcpu_dirty
) {
2186 kvm_arch_get_registers(cpu
);
2187 cpu
->vcpu_dirty
= true;
2191 void kvm_cpu_synchronize_state(CPUState
*cpu
)
2193 if (!cpu
->vcpu_dirty
) {
2194 run_on_cpu(cpu
, do_kvm_cpu_synchronize_state
, RUN_ON_CPU_NULL
);
2198 static void do_kvm_cpu_synchronize_post_reset(CPUState
*cpu
, run_on_cpu_data arg
)
2200 kvm_arch_put_registers(cpu
, KVM_PUT_RESET_STATE
);
2201 cpu
->vcpu_dirty
= false;
2204 void kvm_cpu_synchronize_post_reset(CPUState
*cpu
)
2206 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_reset
, RUN_ON_CPU_NULL
);
2209 static void do_kvm_cpu_synchronize_post_init(CPUState
*cpu
, run_on_cpu_data arg
)
2211 kvm_arch_put_registers(cpu
, KVM_PUT_FULL_STATE
);
2212 cpu
->vcpu_dirty
= false;
2215 void kvm_cpu_synchronize_post_init(CPUState
*cpu
)
2217 run_on_cpu(cpu
, do_kvm_cpu_synchronize_post_init
, RUN_ON_CPU_NULL
);
2220 static void do_kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
, run_on_cpu_data arg
)
2222 cpu
->vcpu_dirty
= true;
2225 void kvm_cpu_synchronize_pre_loadvm(CPUState
*cpu
)
2227 run_on_cpu(cpu
, do_kvm_cpu_synchronize_pre_loadvm
, RUN_ON_CPU_NULL
);
2230 #ifdef KVM_HAVE_MCE_INJECTION
2231 static __thread
void *pending_sigbus_addr
;
2232 static __thread
int pending_sigbus_code
;
2233 static __thread
bool have_sigbus_pending
;
2236 static void kvm_cpu_kick(CPUState
*cpu
)
2238 atomic_set(&cpu
->kvm_run
->immediate_exit
, 1);
2241 static void kvm_cpu_kick_self(void)
2243 if (kvm_immediate_exit
) {
2244 kvm_cpu_kick(current_cpu
);
2246 qemu_cpu_kick_self();
2250 static void kvm_eat_signals(CPUState
*cpu
)
2252 struct timespec ts
= { 0, 0 };
2258 if (kvm_immediate_exit
) {
2259 atomic_set(&cpu
->kvm_run
->immediate_exit
, 0);
2260 /* Write kvm_run->immediate_exit before the cpu->exit_request
2261 * write in kvm_cpu_exec.
2267 sigemptyset(&waitset
);
2268 sigaddset(&waitset
, SIG_IPI
);
2271 r
= sigtimedwait(&waitset
, &siginfo
, &ts
);
2272 if (r
== -1 && !(errno
== EAGAIN
|| errno
== EINTR
)) {
2273 perror("sigtimedwait");
2277 r
= sigpending(&chkset
);
2279 perror("sigpending");
2282 } while (sigismember(&chkset
, SIG_IPI
));
2285 int kvm_cpu_exec(CPUState
*cpu
)
2287 struct kvm_run
*run
= cpu
->kvm_run
;
2290 DPRINTF("kvm_cpu_exec()\n");
2292 if (kvm_arch_process_async_events(cpu
)) {
2293 atomic_set(&cpu
->exit_request
, 0);
2297 qemu_mutex_unlock_iothread();
2298 cpu_exec_start(cpu
);
2303 if (cpu
->vcpu_dirty
) {
2304 kvm_arch_put_registers(cpu
, KVM_PUT_RUNTIME_STATE
);
2305 cpu
->vcpu_dirty
= false;
2308 kvm_arch_pre_run(cpu
, run
);
2309 if (atomic_read(&cpu
->exit_request
)) {
2310 DPRINTF("interrupt exit requested\n");
2312 * KVM requires us to reenter the kernel after IO exits to complete
2313 * instruction emulation. This self-signal will ensure that we
2316 kvm_cpu_kick_self();
2319 /* Read cpu->exit_request before KVM_RUN reads run->immediate_exit.
2320 * Matching barrier in kvm_eat_signals.
2324 run_ret
= kvm_vcpu_ioctl(cpu
, KVM_RUN
, 0);
2326 attrs
= kvm_arch_post_run(cpu
, run
);
2328 #ifdef KVM_HAVE_MCE_INJECTION
2329 if (unlikely(have_sigbus_pending
)) {
2330 qemu_mutex_lock_iothread();
2331 kvm_arch_on_sigbus_vcpu(cpu
, pending_sigbus_code
,
2332 pending_sigbus_addr
);
2333 have_sigbus_pending
= false;
2334 qemu_mutex_unlock_iothread();
2339 if (run_ret
== -EINTR
|| run_ret
== -EAGAIN
) {
2340 DPRINTF("io window exit\n");
2341 kvm_eat_signals(cpu
);
2342 ret
= EXCP_INTERRUPT
;
2345 fprintf(stderr
, "error: kvm run failed %s\n",
2346 strerror(-run_ret
));
2348 if (run_ret
== -EBUSY
) {
2350 "This is probably because your SMT is enabled.\n"
2351 "VCPU can only run on primary threads with all "
2352 "secondary threads offline.\n");
2359 trace_kvm_run_exit(cpu
->cpu_index
, run
->exit_reason
);
2360 switch (run
->exit_reason
) {
2362 DPRINTF("handle_io\n");
2363 /* Called outside BQL */
2364 kvm_handle_io(run
->io
.port
, attrs
,
2365 (uint8_t *)run
+ run
->io
.data_offset
,
2372 DPRINTF("handle_mmio\n");
2373 /* Called outside BQL */
2374 address_space_rw(&address_space_memory
,
2375 run
->mmio
.phys_addr
, attrs
,
2378 run
->mmio
.is_write
);
2381 case KVM_EXIT_IRQ_WINDOW_OPEN
:
2382 DPRINTF("irq_window_open\n");
2383 ret
= EXCP_INTERRUPT
;
2385 case KVM_EXIT_SHUTDOWN
:
2386 DPRINTF("shutdown\n");
2387 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2388 ret
= EXCP_INTERRUPT
;
2390 case KVM_EXIT_UNKNOWN
:
2391 fprintf(stderr
, "KVM: unknown exit, hardware reason %" PRIx64
"\n",
2392 (uint64_t)run
->hw
.hardware_exit_reason
);
2395 case KVM_EXIT_INTERNAL_ERROR
:
2396 ret
= kvm_handle_internal_error(cpu
, run
);
2398 case KVM_EXIT_SYSTEM_EVENT
:
2399 switch (run
->system_event
.type
) {
2400 case KVM_SYSTEM_EVENT_SHUTDOWN
:
2401 qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN
);
2402 ret
= EXCP_INTERRUPT
;
2404 case KVM_SYSTEM_EVENT_RESET
:
2405 qemu_system_reset_request(SHUTDOWN_CAUSE_GUEST_RESET
);
2406 ret
= EXCP_INTERRUPT
;
2408 case KVM_SYSTEM_EVENT_CRASH
:
2409 kvm_cpu_synchronize_state(cpu
);
2410 qemu_mutex_lock_iothread();
2411 qemu_system_guest_panicked(cpu_get_crash_info(cpu
));
2412 qemu_mutex_unlock_iothread();
2416 DPRINTF("kvm_arch_handle_exit\n");
2417 ret
= kvm_arch_handle_exit(cpu
, run
);
2422 DPRINTF("kvm_arch_handle_exit\n");
2423 ret
= kvm_arch_handle_exit(cpu
, run
);
2429 qemu_mutex_lock_iothread();
2432 cpu_dump_state(cpu
, stderr
, CPU_DUMP_CODE
);
2433 vm_stop(RUN_STATE_INTERNAL_ERROR
);
2436 atomic_set(&cpu
->exit_request
, 0);
2440 int kvm_ioctl(KVMState
*s
, int type
, ...)
2447 arg
= va_arg(ap
, void *);
2450 trace_kvm_ioctl(type
, arg
);
2451 ret
= ioctl(s
->fd
, type
, arg
);
2458 int kvm_vm_ioctl(KVMState
*s
, int type
, ...)
2465 arg
= va_arg(ap
, void *);
2468 trace_kvm_vm_ioctl(type
, arg
);
2469 ret
= ioctl(s
->vmfd
, type
, arg
);
2476 int kvm_vcpu_ioctl(CPUState
*cpu
, int type
, ...)
2483 arg
= va_arg(ap
, void *);
2486 trace_kvm_vcpu_ioctl(cpu
->cpu_index
, type
, arg
);
2487 ret
= ioctl(cpu
->kvm_fd
, type
, arg
);
2494 int kvm_device_ioctl(int fd
, int type
, ...)
2501 arg
= va_arg(ap
, void *);
2504 trace_kvm_device_ioctl(fd
, type
, arg
);
2505 ret
= ioctl(fd
, type
, arg
);
2512 int kvm_vm_check_attr(KVMState
*s
, uint32_t group
, uint64_t attr
)
2515 struct kvm_device_attr attribute
= {
2520 if (!kvm_vm_attributes_allowed
) {
2524 ret
= kvm_vm_ioctl(s
, KVM_HAS_DEVICE_ATTR
, &attribute
);
2525 /* kvm returns 0 on success for HAS_DEVICE_ATTR */
2529 int kvm_device_check_attr(int dev_fd
, uint32_t group
, uint64_t attr
)
2531 struct kvm_device_attr attribute
= {
2537 return kvm_device_ioctl(dev_fd
, KVM_HAS_DEVICE_ATTR
, &attribute
) ? 0 : 1;
2540 int kvm_device_access(int fd
, int group
, uint64_t attr
,
2541 void *val
, bool write
, Error
**errp
)
2543 struct kvm_device_attr kvmattr
;
2547 kvmattr
.group
= group
;
2548 kvmattr
.attr
= attr
;
2549 kvmattr
.addr
= (uintptr_t)val
;
2551 err
= kvm_device_ioctl(fd
,
2552 write
? KVM_SET_DEVICE_ATTR
: KVM_GET_DEVICE_ATTR
,
2555 error_setg_errno(errp
, -err
,
2556 "KVM_%s_DEVICE_ATTR failed: Group %d "
2557 "attr 0x%016" PRIx64
,
2558 write
? "SET" : "GET", group
, attr
);
2563 bool kvm_has_sync_mmu(void)
2565 return kvm_state
->sync_mmu
;
2568 int kvm_has_vcpu_events(void)
2570 return kvm_state
->vcpu_events
;
2573 int kvm_has_robust_singlestep(void)
2575 return kvm_state
->robust_singlestep
;
2578 int kvm_has_debugregs(void)
2580 return kvm_state
->debugregs
;
2583 int kvm_max_nested_state_length(void)
2585 return kvm_state
->max_nested_state_len
;
2588 int kvm_has_many_ioeventfds(void)
2590 if (!kvm_enabled()) {
2593 return kvm_state
->many_ioeventfds
;
2596 int kvm_has_gsi_routing(void)
2598 #ifdef KVM_CAP_IRQ_ROUTING
2599 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
2605 int kvm_has_intx_set_mask(void)
2607 return kvm_state
->intx_set_mask
;
2610 bool kvm_arm_supports_user_irq(void)
2612 return kvm_check_extension(kvm_state
, KVM_CAP_ARM_USER_IRQ
);
2615 #ifdef KVM_CAP_SET_GUEST_DEBUG
2616 struct kvm_sw_breakpoint
*kvm_find_sw_breakpoint(CPUState
*cpu
,
2619 struct kvm_sw_breakpoint
*bp
;
2621 QTAILQ_FOREACH(bp
, &cpu
->kvm_state
->kvm_sw_breakpoints
, entry
) {
2629 int kvm_sw_breakpoints_active(CPUState
*cpu
)
2631 return !QTAILQ_EMPTY(&cpu
->kvm_state
->kvm_sw_breakpoints
);
2634 struct kvm_set_guest_debug_data
{
2635 struct kvm_guest_debug dbg
;
2639 static void kvm_invoke_set_guest_debug(CPUState
*cpu
, run_on_cpu_data data
)
2641 struct kvm_set_guest_debug_data
*dbg_data
=
2642 (struct kvm_set_guest_debug_data
*) data
.host_ptr
;
2644 dbg_data
->err
= kvm_vcpu_ioctl(cpu
, KVM_SET_GUEST_DEBUG
,
2648 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2650 struct kvm_set_guest_debug_data data
;
2652 data
.dbg
.control
= reinject_trap
;
2654 if (cpu
->singlestep_enabled
) {
2655 data
.dbg
.control
|= KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_SINGLESTEP
;
2657 kvm_arch_update_guest_debug(cpu
, &data
.dbg
);
2659 run_on_cpu(cpu
, kvm_invoke_set_guest_debug
,
2660 RUN_ON_CPU_HOST_PTR(&data
));
2664 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2665 target_ulong len
, int type
)
2667 struct kvm_sw_breakpoint
*bp
;
2670 if (type
== GDB_BREAKPOINT_SW
) {
2671 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2677 bp
= g_malloc(sizeof(struct kvm_sw_breakpoint
));
2680 err
= kvm_arch_insert_sw_breakpoint(cpu
, bp
);
2686 QTAILQ_INSERT_HEAD(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2688 err
= kvm_arch_insert_hw_breakpoint(addr
, len
, type
);
2695 err
= kvm_update_guest_debug(cpu
, 0);
2703 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2704 target_ulong len
, int type
)
2706 struct kvm_sw_breakpoint
*bp
;
2709 if (type
== GDB_BREAKPOINT_SW
) {
2710 bp
= kvm_find_sw_breakpoint(cpu
, addr
);
2715 if (bp
->use_count
> 1) {
2720 err
= kvm_arch_remove_sw_breakpoint(cpu
, bp
);
2725 QTAILQ_REMOVE(&cpu
->kvm_state
->kvm_sw_breakpoints
, bp
, entry
);
2728 err
= kvm_arch_remove_hw_breakpoint(addr
, len
, type
);
2735 err
= kvm_update_guest_debug(cpu
, 0);
2743 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2745 struct kvm_sw_breakpoint
*bp
, *next
;
2746 KVMState
*s
= cpu
->kvm_state
;
2749 QTAILQ_FOREACH_SAFE(bp
, &s
->kvm_sw_breakpoints
, entry
, next
) {
2750 if (kvm_arch_remove_sw_breakpoint(cpu
, bp
) != 0) {
2751 /* Try harder to find a CPU that currently sees the breakpoint. */
2752 CPU_FOREACH(tmpcpu
) {
2753 if (kvm_arch_remove_sw_breakpoint(tmpcpu
, bp
) == 0) {
2758 QTAILQ_REMOVE(&s
->kvm_sw_breakpoints
, bp
, entry
);
2761 kvm_arch_remove_all_hw_breakpoints();
2764 kvm_update_guest_debug(cpu
, 0);
2768 #else /* !KVM_CAP_SET_GUEST_DEBUG */
2770 int kvm_update_guest_debug(CPUState
*cpu
, unsigned long reinject_trap
)
2775 int kvm_insert_breakpoint(CPUState
*cpu
, target_ulong addr
,
2776 target_ulong len
, int type
)
2781 int kvm_remove_breakpoint(CPUState
*cpu
, target_ulong addr
,
2782 target_ulong len
, int type
)
2787 void kvm_remove_all_breakpoints(CPUState
*cpu
)
2790 #endif /* !KVM_CAP_SET_GUEST_DEBUG */
2792 static int kvm_set_signal_mask(CPUState
*cpu
, const sigset_t
*sigset
)
2794 KVMState
*s
= kvm_state
;
2795 struct kvm_signal_mask
*sigmask
;
2798 sigmask
= g_malloc(sizeof(*sigmask
) + sizeof(*sigset
));
2800 sigmask
->len
= s
->sigmask_len
;
2801 memcpy(sigmask
->sigset
, sigset
, sizeof(*sigset
));
2802 r
= kvm_vcpu_ioctl(cpu
, KVM_SET_SIGNAL_MASK
, sigmask
);
2808 static void kvm_ipi_signal(int sig
)
2811 assert(kvm_immediate_exit
);
2812 kvm_cpu_kick(current_cpu
);
2816 void kvm_init_cpu_signals(CPUState
*cpu
)
2820 struct sigaction sigact
;
2822 memset(&sigact
, 0, sizeof(sigact
));
2823 sigact
.sa_handler
= kvm_ipi_signal
;
2824 sigaction(SIG_IPI
, &sigact
, NULL
);
2826 pthread_sigmask(SIG_BLOCK
, NULL
, &set
);
2827 #if defined KVM_HAVE_MCE_INJECTION
2828 sigdelset(&set
, SIGBUS
);
2829 pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2831 sigdelset(&set
, SIG_IPI
);
2832 if (kvm_immediate_exit
) {
2833 r
= pthread_sigmask(SIG_SETMASK
, &set
, NULL
);
2835 r
= kvm_set_signal_mask(cpu
, &set
);
2838 fprintf(stderr
, "kvm_set_signal_mask: %s\n", strerror(-r
));
2843 /* Called asynchronously in VCPU thread. */
2844 int kvm_on_sigbus_vcpu(CPUState
*cpu
, int code
, void *addr
)
2846 #ifdef KVM_HAVE_MCE_INJECTION
2847 if (have_sigbus_pending
) {
2850 have_sigbus_pending
= true;
2851 pending_sigbus_addr
= addr
;
2852 pending_sigbus_code
= code
;
2853 atomic_set(&cpu
->exit_request
, 1);
2860 /* Called synchronously (via signalfd) in main thread. */
2861 int kvm_on_sigbus(int code
, void *addr
)
2863 #ifdef KVM_HAVE_MCE_INJECTION
2864 /* Action required MCE kills the process if SIGBUS is blocked. Because
2865 * that's what happens in the I/O thread, where we handle MCE via signalfd,
2866 * we can only get action optional here.
2868 assert(code
!= BUS_MCEERR_AR
);
2869 kvm_arch_on_sigbus_vcpu(first_cpu
, code
, addr
);
2876 int kvm_create_device(KVMState
*s
, uint64_t type
, bool test
)
2879 struct kvm_create_device create_dev
;
2881 create_dev
.type
= type
;
2883 create_dev
.flags
= test
? KVM_CREATE_DEVICE_TEST
: 0;
2885 if (!kvm_check_extension(s
, KVM_CAP_DEVICE_CTRL
)) {
2889 ret
= kvm_vm_ioctl(s
, KVM_CREATE_DEVICE
, &create_dev
);
2894 return test
? 0 : create_dev
.fd
;
2897 bool kvm_device_supported(int vmfd
, uint64_t type
)
2899 struct kvm_create_device create_dev
= {
2902 .flags
= KVM_CREATE_DEVICE_TEST
,
2905 if (ioctl(vmfd
, KVM_CHECK_EXTENSION
, KVM_CAP_DEVICE_CTRL
) <= 0) {
2909 return (ioctl(vmfd
, KVM_CREATE_DEVICE
, &create_dev
) >= 0);
2912 int kvm_set_one_reg(CPUState
*cs
, uint64_t id
, void *source
)
2914 struct kvm_one_reg reg
;
2918 reg
.addr
= (uintptr_t) source
;
2919 r
= kvm_vcpu_ioctl(cs
, KVM_SET_ONE_REG
, ®
);
2921 trace_kvm_failed_reg_set(id
, strerror(-r
));
2926 int kvm_get_one_reg(CPUState
*cs
, uint64_t id
, void *target
)
2928 struct kvm_one_reg reg
;
2932 reg
.addr
= (uintptr_t) target
;
2933 r
= kvm_vcpu_ioctl(cs
, KVM_GET_ONE_REG
, ®
);
2935 trace_kvm_failed_reg_get(id
, strerror(-r
));
2940 static bool kvm_accel_has_memory(MachineState
*ms
, AddressSpace
*as
,
2941 hwaddr start_addr
, hwaddr size
)
2943 KVMState
*kvm
= KVM_STATE(ms
->accelerator
);
2946 for (i
= 0; i
< kvm
->nr_as
; ++i
) {
2947 if (kvm
->as
[i
].as
== as
&& kvm
->as
[i
].ml
) {
2948 size
= MIN(kvm_max_slot_size
, size
);
2949 return NULL
!= kvm_lookup_matching_slot(kvm
->as
[i
].ml
,
2957 static void kvm_accel_class_init(ObjectClass
*oc
, void *data
)
2959 AccelClass
*ac
= ACCEL_CLASS(oc
);
2961 ac
->init_machine
= kvm_init
;
2962 ac
->has_memory
= kvm_accel_has_memory
;
2963 ac
->allowed
= &kvm_allowed
;
2966 static const TypeInfo kvm_accel_type
= {
2967 .name
= TYPE_KVM_ACCEL
,
2968 .parent
= TYPE_ACCEL
,
2969 .class_init
= kvm_accel_class_init
,
2970 .instance_size
= sizeof(KVMState
),
2973 static void kvm_type_init(void)
2975 type_register_static(&kvm_accel_type
);
2978 type_init(kvm_type_init
);