4 * Copyright (C) 2006-2008 Qumranet Technologies
6 * Licensed under the terms of the GNU GPL version 2 or higher.
9 #include "config-host.h"
15 #include "qemu-common.h"
26 #include <sys/utsname.h>
27 #include <sys/syscall.h>
29 #include <sys/ioctl.h>
31 #include <sys/prctl.h>
37 #define PR_MCE_KILL 33
41 #define BUS_MCEERR_AR 4
44 #define BUS_MCEERR_AO 5
47 #define EXPECTED_KVM_API_VERSION 12
49 #if EXPECTED_KVM_API_VERSION != KVM_API_VERSION
50 #error libkvm: userspace and kernel version mismatch
55 int kvm_pit_reinject
= 1;
60 kvm_context_t kvm_context
;
62 pthread_mutex_t qemu_mutex
= PTHREAD_MUTEX_INITIALIZER
;
63 pthread_cond_t qemu_vcpu_cond
= PTHREAD_COND_INITIALIZER
;
64 pthread_cond_t qemu_system_cond
= PTHREAD_COND_INITIALIZER
;
65 pthread_cond_t qemu_pause_cond
= PTHREAD_COND_INITIALIZER
;
66 pthread_cond_t qemu_work_cond
= PTHREAD_COND_INITIALIZER
;
67 __thread CPUState
*current_env
;
69 static int qemu_system_ready
;
71 #define SIG_IPI (SIGRTMIN+4)
74 static int io_thread_sigfd
= -1;
76 static CPUState
*kvm_debug_cpu_requested
;
78 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
79 /* The list of ioperm_data */
80 static QLIST_HEAD(, ioperm_data
) ioperm_head
;
83 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
85 int kvm_abi
= EXPECTED_KVM_API_VERSION
;
88 #ifdef KVM_CAP_SET_GUEST_DEBUG
89 static int kvm_debug(CPUState
*env
,
90 struct kvm_debug_exit_arch
*arch_info
)
92 int handle
= kvm_arch_debug(arch_info
);
95 kvm_debug_cpu_requested
= env
;
102 static int handle_unhandled(uint64_t reason
)
104 fprintf(stderr
, "kvm: unhandled exit %" PRIx64
"\n", reason
);
108 #define VMX_INVALID_GUEST_STATE 0x80000021
110 static int handle_failed_vmentry(uint64_t reason
)
112 fprintf(stderr
, "kvm: vm entry failed with error 0x%" PRIx64
"\n\n", reason
);
114 /* Perhaps we will need to check if this machine is intel since exit reason 0x21
115 has a different interpretation on SVM */
116 if (reason
== VMX_INVALID_GUEST_STATE
) {
117 fprintf(stderr
, "If you're runnning a guest on an Intel machine without\n");
118 fprintf(stderr
, "unrestricted mode support, the failure can be most likely\n");
119 fprintf(stderr
, "due to the guest entering an invalid state for Intel VT.\n");
120 fprintf(stderr
, "For example, the guest maybe running in big real mode\n");
121 fprintf(stderr
, "which is not supported on less recent Intel processors.\n\n");
127 static inline void set_gsi(kvm_context_t kvm
, unsigned int gsi
)
129 uint32_t *bitmap
= kvm
->used_gsi_bitmap
;
131 if (gsi
< kvm
->max_gsi
)
132 bitmap
[gsi
/ 32] |= 1U << (gsi
% 32);
134 DPRINTF("Invalid GSI %u\n", gsi
);
137 static inline void clear_gsi(kvm_context_t kvm
, unsigned int gsi
)
139 uint32_t *bitmap
= kvm
->used_gsi_bitmap
;
141 if (gsi
< kvm
->max_gsi
)
142 bitmap
[gsi
/ 32] &= ~(1U << (gsi
% 32));
144 DPRINTF("Invalid GSI %u\n", gsi
);
147 static int kvm_create_context(void);
149 int kvm_init(int smp_cpus
)
155 fd
= open("/dev/kvm", O_RDWR
);
157 perror("open /dev/kvm");
160 r
= ioctl(fd
, KVM_GET_API_VERSION
, 0);
163 "kvm kernel version too old: "
164 "KVM_GET_API_VERSION ioctl not supported\n");
167 if (r
< EXPECTED_KVM_API_VERSION
) {
168 fprintf(stderr
, "kvm kernel version too old: "
169 "We expect API version %d or newer, but got "
170 "version %d\n", EXPECTED_KVM_API_VERSION
, r
);
173 if (r
> EXPECTED_KVM_API_VERSION
) {
174 fprintf(stderr
, "kvm userspace version too old\n");
178 kvm_page_size
= getpagesize();
179 kvm_state
= qemu_mallocz(sizeof(*kvm_state
));
180 kvm_context
= &kvm_state
->kvm_context
;
183 kvm_state
->vmfd
= -1;
184 kvm_context
->opaque
= cpu_single_env
;
185 kvm_context
->dirty_pages_log_all
= 0;
186 kvm_context
->no_irqchip_creation
= 0;
187 kvm_context
->no_pit_creation
= 0;
189 #ifdef KVM_CAP_SET_GUEST_DEBUG
190 QTAILQ_INIT(&kvm_state
->kvm_sw_breakpoints
);
193 gsi_count
= kvm_get_gsi_count(kvm_context
);
197 /* Round up so we can search ints using ffs */
198 gsi_bits
= ALIGN(gsi_count
, 32);
199 kvm_context
->used_gsi_bitmap
= qemu_mallocz(gsi_bits
/ 8);
200 kvm_context
->max_gsi
= gsi_bits
;
202 /* Mark any over-allocated bits as already in use */
203 for (i
= gsi_count
; i
< gsi_bits
; i
++) {
204 set_gsi(kvm_context
, i
);
208 kvm_cpu_register_phys_memory_client();
210 pthread_mutex_lock(&qemu_mutex
);
211 return kvm_create_context();
218 static void kvm_finalize(KVMState
*s
)
221 if (kvm->vcpu_fd[0] != -1)
222 close(kvm->vcpu_fd[0]);
223 if (kvm->vm_fd != -1)
230 void kvm_disable_irqchip_creation(kvm_context_t kvm
)
232 kvm
->no_irqchip_creation
= 1;
235 void kvm_disable_pit_creation(kvm_context_t kvm
)
237 kvm
->no_pit_creation
= 1;
240 static void kvm_reset_vcpu(void *opaque
)
242 CPUState
*env
= opaque
;
244 kvm_arch_cpu_reset(env
);
247 static void kvm_create_vcpu(CPUState
*env
, int id
)
251 KVMState
*s
= kvm_state
;
253 r
= kvm_vm_ioctl(kvm_state
, KVM_CREATE_VCPU
, id
);
255 fprintf(stderr
, "kvm_create_vcpu: %m\n");
256 fprintf(stderr
, "Failed to create vCPU. Check the -smp parameter.\n");
261 env
->kvm_state
= kvm_state
;
263 mmap_size
= kvm_ioctl(kvm_state
, KVM_GET_VCPU_MMAP_SIZE
, 0);
265 fprintf(stderr
, "get vcpu mmap size: %m\n");
269 mmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
, env
->kvm_fd
,
271 if (env
->kvm_run
== MAP_FAILED
) {
272 fprintf(stderr
, "mmap vcpu area: %m\n");
276 #ifdef KVM_CAP_COALESCED_MMIO
277 if (s
->coalesced_mmio
&& !s
->coalesced_mmio_ring
)
278 s
->coalesced_mmio_ring
= (void *) env
->kvm_run
+
279 s
->coalesced_mmio
* PAGE_SIZE
;
282 r
= kvm_arch_init_vcpu(env
);
284 qemu_register_reset(kvm_reset_vcpu
, env
);
291 /* We're no good with semi-broken states. */
295 static int kvm_set_boot_vcpu_id(kvm_context_t kvm
, uint32_t id
)
297 #ifdef KVM_CAP_SET_BOOT_CPU_ID
298 int r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_SET_BOOT_CPU_ID
);
300 return kvm_vm_ioctl(kvm_state
, KVM_SET_BOOT_CPU_ID
, id
);
308 int kvm_create_vm(kvm_context_t kvm
)
311 #ifdef KVM_CAP_IRQ_ROUTING
312 kvm
->irq_routes
= qemu_mallocz(sizeof(*kvm
->irq_routes
));
313 kvm
->nr_allocated_irq_routes
= 0;
316 fd
= kvm_ioctl(kvm_state
, KVM_CREATE_VM
, 0);
318 fprintf(stderr
, "kvm_create_vm: %m\n");
321 kvm_state
->vmfd
= fd
;
325 static int kvm_create_default_phys_mem(kvm_context_t kvm
,
326 unsigned long phys_mem_bytes
,
329 #ifdef KVM_CAP_USER_MEMORY
330 int r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_USER_MEMORY
);
334 "Hypervisor too old: KVM_CAP_USER_MEMORY extension not supported\n");
336 #error Hypervisor too old: KVM_CAP_USER_MEMORY extension not supported
341 void kvm_create_irqchip(kvm_context_t kvm
)
345 kvm
->irqchip_in_kernel
= 0;
346 #ifdef KVM_CAP_IRQCHIP
347 if (!kvm
->no_irqchip_creation
) {
348 r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_IRQCHIP
);
349 if (r
> 0) { /* kernel irqchip supported */
350 r
= kvm_vm_ioctl(kvm_state
, KVM_CREATE_IRQCHIP
);
352 kvm
->irqchip_inject_ioctl
= KVM_IRQ_LINE
;
353 #if defined(KVM_CAP_IRQ_INJECT_STATUS) && defined(KVM_IRQ_LINE_STATUS)
354 r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
,
355 KVM_CAP_IRQ_INJECT_STATUS
);
357 kvm
->irqchip_inject_ioctl
= KVM_IRQ_LINE_STATUS
;
360 kvm
->irqchip_in_kernel
= 1;
362 fprintf(stderr
, "Create kernel PIC irqchip failed\n");
366 kvm_state
->irqchip_in_kernel
= kvm
->irqchip_in_kernel
;
369 int kvm_create(kvm_context_t kvm
, unsigned long phys_mem_bytes
, void **vm_mem
)
373 r
= kvm_create_vm(kvm
);
377 r
= kvm_arch_create(kvm
, phys_mem_bytes
, vm_mem
);
381 for (i
= 0; i
< ARRAY_SIZE(kvm_state
->slots
); i
++) {
382 kvm_state
->slots
[i
].slot
= i
;
385 r
= kvm_create_default_phys_mem(kvm
, phys_mem_bytes
, vm_mem
);
390 kvm_create_irqchip(kvm
);
395 #ifdef KVM_CAP_IRQCHIP
397 int kvm_set_irq_level(kvm_context_t kvm
, int irq
, int level
, int *status
)
399 struct kvm_irq_level event
;
402 if (!kvm
->irqchip_in_kernel
) {
407 r
= kvm_vm_ioctl(kvm_state
, kvm
->irqchip_inject_ioctl
, &event
);
409 perror("kvm_set_irq_level");
413 #ifdef KVM_CAP_IRQ_INJECT_STATUS
415 (kvm
->irqchip_inject_ioctl
== KVM_IRQ_LINE
) ? 1 : event
.status
;
424 int kvm_get_irqchip(kvm_context_t kvm
, struct kvm_irqchip
*chip
)
428 if (!kvm
->irqchip_in_kernel
) {
431 r
= kvm_vm_ioctl(kvm_state
, KVM_GET_IRQCHIP
, chip
);
433 perror("kvm_get_irqchip\n");
438 int kvm_set_irqchip(kvm_context_t kvm
, struct kvm_irqchip
*chip
)
442 if (!kvm
->irqchip_in_kernel
) {
445 r
= kvm_vm_ioctl(kvm_state
, KVM_SET_IRQCHIP
, chip
);
447 perror("kvm_set_irqchip\n");
454 static int handle_debug(CPUState
*env
)
456 #ifdef KVM_CAP_SET_GUEST_DEBUG
457 struct kvm_run
*run
= env
->kvm_run
;
459 return kvm_debug(env
, &run
->debug
.arch
);
465 int kvm_get_regs(CPUState
*env
, struct kvm_regs
*regs
)
467 return kvm_vcpu_ioctl(env
, KVM_GET_REGS
, regs
);
470 int kvm_set_regs(CPUState
*env
, struct kvm_regs
*regs
)
472 return kvm_vcpu_ioctl(env
, KVM_SET_REGS
, regs
);
475 #ifdef KVM_CAP_MP_STATE
476 int kvm_get_mpstate(CPUState
*env
, struct kvm_mp_state
*mp_state
)
480 r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_MP_STATE
);
482 return kvm_vcpu_ioctl(env
, KVM_GET_MP_STATE
, mp_state
);
487 int kvm_set_mpstate(CPUState
*env
, struct kvm_mp_state
*mp_state
)
491 r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_MP_STATE
);
493 return kvm_vcpu_ioctl(env
, KVM_SET_MP_STATE
, mp_state
);
499 static int handle_mmio(CPUState
*env
)
501 unsigned long addr
= env
->kvm_run
->mmio
.phys_addr
;
502 struct kvm_run
*kvm_run
= env
->kvm_run
;
503 void *data
= kvm_run
->mmio
.data
;
505 /* hack: Red Hat 7.1 generates these weird accesses. */
506 if ((addr
> 0xa0000 - 4 && addr
<= 0xa0000) && kvm_run
->mmio
.len
== 3) {
510 cpu_physical_memory_rw(addr
, data
, kvm_run
->mmio
.len
, kvm_run
->mmio
.is_write
);
514 int handle_io_window(kvm_context_t kvm
)
519 int handle_shutdown(kvm_context_t kvm
, CPUState
*env
)
521 /* stop the current vcpu from going back to guest mode */
524 qemu_system_reset_request();
528 static inline void push_nmi(kvm_context_t kvm
)
530 #ifdef KVM_CAP_USER_NMI
531 kvm_arch_push_nmi(kvm
->opaque
);
532 #endif /* KVM_CAP_USER_NMI */
535 void post_kvm_run(kvm_context_t kvm
, CPUState
*env
)
537 pthread_mutex_lock(&qemu_mutex
);
538 kvm_arch_post_run(env
, env
->kvm_run
);
539 cpu_single_env
= env
;
542 int pre_kvm_run(kvm_context_t kvm
, CPUState
*env
)
544 kvm_arch_pre_run(env
, env
->kvm_run
);
546 pthread_mutex_unlock(&qemu_mutex
);
550 int kvm_is_ready_for_interrupt_injection(CPUState
*env
)
552 return env
->kvm_run
->ready_for_interrupt_injection
;
555 int kvm_run(CPUState
*env
)
558 kvm_context_t kvm
= &env
->kvm_state
->kvm_context
;
559 struct kvm_run
*run
= env
->kvm_run
;
560 int fd
= env
->kvm_fd
;
563 if (env
->kvm_vcpu_dirty
) {
564 kvm_arch_load_regs(env
, KVM_PUT_RUNTIME_STATE
);
565 env
->kvm_vcpu_dirty
= 0;
568 #if !defined(__s390__)
569 if (!kvm
->irqchip_in_kernel
) {
570 run
->request_interrupt_window
= kvm_arch_try_push_interrupts(env
);
574 r
= pre_kvm_run(kvm
, env
);
578 if (env
->exit_request
) {
579 env
->exit_request
= 0;
580 pthread_kill(env
->kvm_cpu_state
.thread
, SIG_IPI
);
582 r
= ioctl(fd
, KVM_RUN
, 0);
584 if (r
== -1 && errno
!= EINTR
&& errno
!= EAGAIN
) {
586 post_kvm_run(kvm
, env
);
587 fprintf(stderr
, "kvm_run: %s\n", strerror(-r
));
591 post_kvm_run(kvm
, env
);
593 kvm_flush_coalesced_mmio_buffer();
595 #if !defined(__s390__)
597 r
= handle_io_window(kvm
);
602 switch (run
->exit_reason
) {
603 case KVM_EXIT_UNKNOWN
:
604 r
= handle_unhandled(run
->hw
.hardware_exit_reason
);
606 case KVM_EXIT_FAIL_ENTRY
:
607 r
= handle_failed_vmentry(run
->fail_entry
.hardware_entry_failure_reason
);
609 case KVM_EXIT_EXCEPTION
:
610 fprintf(stderr
, "exception %d (%x)\n", run
->ex
.exception
,
617 r
= kvm_handle_io(run
->io
.port
,
618 (uint8_t *)run
+ run
->io
.data_offset
,
625 r
= handle_debug(env
);
628 r
= handle_mmio(env
);
631 r
= kvm_arch_halt(env
);
633 case KVM_EXIT_IRQ_WINDOW_OPEN
:
635 case KVM_EXIT_SHUTDOWN
:
636 r
= handle_shutdown(kvm
, env
);
638 #if defined(__s390__)
639 case KVM_EXIT_S390_SIEIC
:
640 r
= kvm_s390_handle_intercept(kvm
, env
, run
);
642 case KVM_EXIT_S390_RESET
:
643 r
= kvm_s390_handle_reset(kvm
, env
, run
);
646 case KVM_EXIT_INTERNAL_ERROR
:
647 kvm_handle_internal_error(env
, run
);
651 if (kvm_arch_run(env
)) {
652 fprintf(stderr
, "unhandled vm exit: 0x%x\n", run
->exit_reason
);
666 int kvm_inject_irq(CPUState
*env
, unsigned irq
)
668 struct kvm_interrupt intr
;
671 return kvm_vcpu_ioctl(env
, KVM_INTERRUPT
, &intr
);
674 int kvm_inject_nmi(CPUState
*env
)
676 #ifdef KVM_CAP_USER_NMI
677 return kvm_vcpu_ioctl(env
, KVM_NMI
);
683 int kvm_init_coalesced_mmio(kvm_context_t kvm
)
686 kvm_state
->coalesced_mmio
= 0;
687 #ifdef KVM_CAP_COALESCED_MMIO
688 r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_COALESCED_MMIO
);
690 kvm_state
->coalesced_mmio
= r
;
697 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
698 int kvm_assign_pci_device(kvm_context_t kvm
,
699 struct kvm_assigned_pci_dev
*assigned_dev
)
701 return kvm_vm_ioctl(kvm_state
, KVM_ASSIGN_PCI_DEVICE
, assigned_dev
);
704 static int kvm_old_assign_irq(kvm_context_t kvm
,
705 struct kvm_assigned_irq
*assigned_irq
)
707 return kvm_vm_ioctl(kvm_state
, KVM_ASSIGN_IRQ
, assigned_irq
);
710 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
711 int kvm_assign_irq(kvm_context_t kvm
, struct kvm_assigned_irq
*assigned_irq
)
715 ret
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_ASSIGN_DEV_IRQ
);
717 return kvm_vm_ioctl(kvm_state
, KVM_ASSIGN_DEV_IRQ
, assigned_irq
);
720 return kvm_old_assign_irq(kvm
, assigned_irq
);
723 int kvm_deassign_irq(kvm_context_t kvm
, struct kvm_assigned_irq
*assigned_irq
)
725 return kvm_vm_ioctl(kvm_state
, KVM_DEASSIGN_DEV_IRQ
, assigned_irq
);
728 int kvm_assign_irq(kvm_context_t kvm
, struct kvm_assigned_irq
*assigned_irq
)
730 return kvm_old_assign_irq(kvm
, assigned_irq
);
735 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
736 int kvm_deassign_pci_device(kvm_context_t kvm
,
737 struct kvm_assigned_pci_dev
*assigned_dev
)
739 return kvm_vm_ioctl(kvm_state
, KVM_DEASSIGN_PCI_DEVICE
, assigned_dev
);
743 int kvm_reinject_control(kvm_context_t kvm
, int pit_reinject
)
745 #ifdef KVM_CAP_REINJECT_CONTROL
747 struct kvm_reinject_control control
;
749 control
.pit_reinject
= pit_reinject
;
751 r
= kvm_ioctl(kvm_state
, KVM_CHECK_EXTENSION
, KVM_CAP_REINJECT_CONTROL
);
753 return kvm_vm_ioctl(kvm_state
, KVM_REINJECT_CONTROL
, &control
);
759 int kvm_has_gsi_routing(void)
763 #ifdef KVM_CAP_IRQ_ROUTING
764 r
= kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
769 int kvm_get_gsi_count(kvm_context_t kvm
)
771 #ifdef KVM_CAP_IRQ_ROUTING
772 return kvm_check_extension(kvm_state
, KVM_CAP_IRQ_ROUTING
);
778 int kvm_clear_gsi_routes(void)
780 #ifdef KVM_CAP_IRQ_ROUTING
781 kvm_context_t kvm
= kvm_context
;
783 kvm
->irq_routes
->nr
= 0;
790 int kvm_add_routing_entry(struct kvm_irq_routing_entry
*entry
)
792 #ifdef KVM_CAP_IRQ_ROUTING
793 kvm_context_t kvm
= kvm_context
;
794 struct kvm_irq_routing
*z
;
795 struct kvm_irq_routing_entry
*new;
798 if (kvm
->irq_routes
->nr
== kvm
->nr_allocated_irq_routes
) {
799 n
= kvm
->nr_allocated_irq_routes
* 2;
803 size
= sizeof(struct kvm_irq_routing
);
804 size
+= n
* sizeof(*new);
805 z
= realloc(kvm
->irq_routes
, size
);
809 kvm
->nr_allocated_irq_routes
= n
;
812 n
= kvm
->irq_routes
->nr
++;
813 new = &kvm
->irq_routes
->entries
[n
];
814 memset(new, 0, sizeof(*new));
815 new->gsi
= entry
->gsi
;
816 new->type
= entry
->type
;
817 new->flags
= entry
->flags
;
820 set_gsi(kvm
, entry
->gsi
);
828 int kvm_add_irq_route(int gsi
, int irqchip
, int pin
)
830 #ifdef KVM_CAP_IRQ_ROUTING
831 struct kvm_irq_routing_entry e
;
834 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
836 e
.u
.irqchip
.irqchip
= irqchip
;
837 e
.u
.irqchip
.pin
= pin
;
838 return kvm_add_routing_entry(&e
);
844 int kvm_del_routing_entry(struct kvm_irq_routing_entry
*entry
)
846 #ifdef KVM_CAP_IRQ_ROUTING
847 kvm_context_t kvm
= kvm_context
;
848 struct kvm_irq_routing_entry
*e
, *p
;
849 int i
, gsi
, found
= 0;
853 for (i
= 0; i
< kvm
->irq_routes
->nr
; ++i
) {
854 e
= &kvm
->irq_routes
->entries
[i
];
855 if (e
->type
== entry
->type
&& e
->gsi
== gsi
) {
857 case KVM_IRQ_ROUTING_IRQCHIP
:{
858 if (e
->u
.irqchip
.irqchip
==
859 entry
->u
.irqchip
.irqchip
860 && e
->u
.irqchip
.pin
== entry
->u
.irqchip
.pin
) {
861 p
= &kvm
->irq_routes
->entries
[--kvm
->irq_routes
->nr
];
867 case KVM_IRQ_ROUTING_MSI
:{
868 if (e
->u
.msi
.address_lo
==
869 entry
->u
.msi
.address_lo
870 && e
->u
.msi
.address_hi
==
871 entry
->u
.msi
.address_hi
872 && e
->u
.msi
.data
== entry
->u
.msi
.data
) {
873 p
= &kvm
->irq_routes
->entries
[--kvm
->irq_routes
->nr
];
883 /* If there are no other users of this GSI
884 * mark it available in the bitmap */
885 for (i
= 0; i
< kvm
->irq_routes
->nr
; i
++) {
886 e
= &kvm
->irq_routes
->entries
[i
];
890 if (i
== kvm
->irq_routes
->nr
) {
904 int kvm_update_routing_entry(struct kvm_irq_routing_entry
*entry
,
905 struct kvm_irq_routing_entry
*newentry
)
907 #ifdef KVM_CAP_IRQ_ROUTING
908 kvm_context_t kvm
= kvm_context
;
909 struct kvm_irq_routing_entry
*e
;
912 if (entry
->gsi
!= newentry
->gsi
|| entry
->type
!= newentry
->type
) {
916 for (i
= 0; i
< kvm
->irq_routes
->nr
; ++i
) {
917 e
= &kvm
->irq_routes
->entries
[i
];
918 if (e
->type
!= entry
->type
|| e
->gsi
!= entry
->gsi
) {
922 case KVM_IRQ_ROUTING_IRQCHIP
:
923 if (e
->u
.irqchip
.irqchip
== entry
->u
.irqchip
.irqchip
&&
924 e
->u
.irqchip
.pin
== entry
->u
.irqchip
.pin
) {
925 memcpy(&e
->u
.irqchip
, &newentry
->u
.irqchip
,
926 sizeof e
->u
.irqchip
);
930 case KVM_IRQ_ROUTING_MSI
:
931 if (e
->u
.msi
.address_lo
== entry
->u
.msi
.address_lo
&&
932 e
->u
.msi
.address_hi
== entry
->u
.msi
.address_hi
&&
933 e
->u
.msi
.data
== entry
->u
.msi
.data
) {
934 memcpy(&e
->u
.msi
, &newentry
->u
.msi
, sizeof e
->u
.msi
);
948 int kvm_del_irq_route(int gsi
, int irqchip
, int pin
)
950 #ifdef KVM_CAP_IRQ_ROUTING
951 struct kvm_irq_routing_entry e
;
954 e
.type
= KVM_IRQ_ROUTING_IRQCHIP
;
956 e
.u
.irqchip
.irqchip
= irqchip
;
957 e
.u
.irqchip
.pin
= pin
;
958 return kvm_del_routing_entry(&e
);
964 int kvm_commit_irq_routes(void)
966 #ifdef KVM_CAP_IRQ_ROUTING
967 kvm_context_t kvm
= kvm_context
;
969 kvm
->irq_routes
->flags
= 0;
970 return kvm_vm_ioctl(kvm_state
, KVM_SET_GSI_ROUTING
, kvm
->irq_routes
);
976 int kvm_get_irq_route_gsi(void)
978 kvm_context_t kvm
= kvm_context
;
980 uint32_t *buf
= kvm
->used_gsi_bitmap
;
982 /* Return the lowest unused GSI in the bitmap */
983 for (i
= 0; i
< kvm
->max_gsi
/ 32; i
++) {
989 return bit
- 1 + i
* 32;
995 static void kvm_msix_routing_entry(struct kvm_irq_routing_entry
*e
,
996 uint32_t gsi
, uint32_t addr_lo
,
997 uint32_t addr_hi
, uint32_t data
)
1001 e
->type
= KVM_IRQ_ROUTING_MSI
;
1003 e
->u
.msi
.address_lo
= addr_lo
;
1004 e
->u
.msi
.address_hi
= addr_hi
;
1005 e
->u
.msi
.data
= data
;
1008 int kvm_add_msix(uint32_t gsi
, uint32_t addr_lo
,
1009 uint32_t addr_hi
, uint32_t data
)
1011 struct kvm_irq_routing_entry e
;
1013 kvm_msix_routing_entry(&e
, gsi
, addr_lo
, addr_hi
, data
);
1014 return kvm_add_routing_entry(&e
);
1017 int kvm_del_msix(uint32_t gsi
, uint32_t addr_lo
,
1018 uint32_t addr_hi
, uint32_t data
)
1020 struct kvm_irq_routing_entry e
;
1022 kvm_msix_routing_entry(&e
, gsi
, addr_lo
, addr_hi
, data
);
1023 return kvm_del_routing_entry(&e
);
1026 int kvm_update_msix(uint32_t old_gsi
, uint32_t old_addr_lo
,
1027 uint32_t old_addr_hi
, uint32_t old_data
,
1028 uint32_t new_gsi
, uint32_t new_addr_lo
,
1029 uint32_t new_addr_hi
, uint32_t new_data
)
1031 struct kvm_irq_routing_entry e1
, e2
;
1033 kvm_msix_routing_entry(&e1
, old_gsi
, old_addr_lo
, old_addr_hi
, old_data
);
1034 kvm_msix_routing_entry(&e2
, new_gsi
, new_addr_lo
, new_addr_hi
, new_data
);
1035 return kvm_update_routing_entry(&e1
, &e2
);
1039 #ifdef KVM_CAP_DEVICE_MSIX
1040 int kvm_assign_set_msix_nr(kvm_context_t kvm
,
1041 struct kvm_assigned_msix_nr
*msix_nr
)
1043 return kvm_vm_ioctl(kvm_state
, KVM_ASSIGN_SET_MSIX_NR
, msix_nr
);
1046 int kvm_assign_set_msix_entry(kvm_context_t kvm
,
1047 struct kvm_assigned_msix_entry
*entry
)
1049 return kvm_vm_ioctl(kvm_state
, KVM_ASSIGN_SET_MSIX_ENTRY
, entry
);
1053 #if defined(KVM_CAP_IRQFD) && defined(CONFIG_EVENTFD)
1055 #include <sys/eventfd.h>
1057 static int _kvm_irqfd(kvm_context_t kvm
, int fd
, int gsi
, int flags
)
1059 struct kvm_irqfd data
= {
1065 return kvm_vm_ioctl(kvm_state
, KVM_IRQFD
, &data
);
1068 int kvm_irqfd(kvm_context_t kvm
, int gsi
, int flags
)
1073 if (!kvm_check_extension(kvm_state
, KVM_CAP_IRQFD
))
1081 r
= _kvm_irqfd(kvm
, fd
, gsi
, 0);
1090 #else /* KVM_CAP_IRQFD */
1092 int kvm_irqfd(kvm_context_t kvm
, int gsi
, int flags
)
1097 #endif /* KVM_CAP_IRQFD */
1098 unsigned long kvm_get_thread_id(void)
1100 return syscall(SYS_gettid
);
1103 static void qemu_cond_wait(pthread_cond_t
*cond
)
1105 CPUState
*env
= cpu_single_env
;
1107 pthread_cond_wait(cond
, &qemu_mutex
);
1108 cpu_single_env
= env
;
1111 static void sig_ipi_handler(int n
)
1115 static void sigbus_reraise(void)
1118 struct sigaction action
;
1120 memset(&action
, 0, sizeof(action
));
1121 action
.sa_handler
= SIG_DFL
;
1122 if (!sigaction(SIGBUS
, &action
, NULL
)) {
1125 sigaddset(&set
, SIGBUS
);
1126 sigprocmask(SIG_UNBLOCK
, &set
, NULL
);
1128 perror("Failed to re-raise SIGBUS!\n");
1132 static void sigbus_handler(int n
, struct qemu_signalfd_siginfo
*siginfo
,
1135 if (kvm_on_sigbus(siginfo
->ssi_code
, (void *)(intptr_t)siginfo
->ssi_addr
))
1139 void on_vcpu(CPUState
*env
, void (*func
)(void *data
), void *data
)
1141 struct qemu_work_item wi
;
1143 if (env
== current_env
) {
1150 if (!env
->kvm_cpu_state
.queued_work_first
) {
1151 env
->kvm_cpu_state
.queued_work_first
= &wi
;
1153 env
->kvm_cpu_state
.queued_work_last
->next
= &wi
;
1155 env
->kvm_cpu_state
.queued_work_last
= &wi
;
1159 pthread_kill(env
->kvm_cpu_state
.thread
, SIG_IPI
);
1161 qemu_cond_wait(&qemu_work_cond
);
1165 static void do_kvm_cpu_synchronize_state(void *_env
)
1167 CPUState
*env
= _env
;
1169 if (!env
->kvm_vcpu_dirty
) {
1170 kvm_arch_save_regs(env
);
1171 env
->kvm_vcpu_dirty
= 1;
1175 void kvm_cpu_synchronize_state(CPUState
*env
)
1177 if (!env
->kvm_vcpu_dirty
) {
1178 on_vcpu(env
, do_kvm_cpu_synchronize_state
, env
);
1182 void kvm_cpu_synchronize_post_reset(CPUState
*env
)
1184 kvm_arch_load_regs(env
, KVM_PUT_RESET_STATE
);
1185 env
->kvm_vcpu_dirty
= 0;
1188 void kvm_cpu_synchronize_post_init(CPUState
*env
)
1190 kvm_arch_load_regs(env
, KVM_PUT_FULL_STATE
);
1191 env
->kvm_vcpu_dirty
= 0;
1194 static void inject_interrupt(void *data
)
1196 cpu_interrupt(current_env
, (long) data
);
1199 void kvm_inject_interrupt(CPUState
*env
, int mask
)
1201 on_vcpu(env
, inject_interrupt
, (void *) (long) mask
);
1204 void kvm_update_interrupt_request(CPUState
*env
)
1209 if (!current_env
|| !current_env
->created
) {
1213 * Testing for created here is really redundant
1215 if (current_env
&& current_env
->created
&&
1216 env
!= current_env
&& !env
->kvm_cpu_state
.signalled
) {
1221 env
->kvm_cpu_state
.signalled
= 1;
1222 if (env
->kvm_cpu_state
.thread
) {
1223 pthread_kill(env
->kvm_cpu_state
.thread
, SIG_IPI
);
1229 int kvm_cpu_exec(CPUState
*env
)
1235 printf("kvm_run returned %d\n", r
);
1242 int kvm_cpu_is_stopped(CPUState
*env
)
1244 return !vm_running
|| env
->stopped
;
1247 static void flush_queued_work(CPUState
*env
)
1249 struct qemu_work_item
*wi
;
1251 if (!env
->kvm_cpu_state
.queued_work_first
) {
1255 while ((wi
= env
->kvm_cpu_state
.queued_work_first
)) {
1256 env
->kvm_cpu_state
.queued_work_first
= wi
->next
;
1260 env
->kvm_cpu_state
.queued_work_last
= NULL
;
1261 pthread_cond_broadcast(&qemu_work_cond
);
1264 static void kvm_main_loop_wait(CPUState
*env
, int timeout
)
1272 ts
.tv_sec
= timeout
/ 1000;
1273 ts
.tv_nsec
= (timeout
% 1000) * 1000000;
1274 sigemptyset(&waitset
);
1275 sigaddset(&waitset
, SIG_IPI
);
1276 sigaddset(&waitset
, SIGBUS
);
1279 pthread_mutex_unlock(&qemu_mutex
);
1281 r
= sigtimedwait(&waitset
, &siginfo
, &ts
);
1284 pthread_mutex_lock(&qemu_mutex
);
1286 if (r
== -1 && !(e
== EAGAIN
|| e
== EINTR
)) {
1287 printf("sigtimedwait: %s\n", strerror(e
));
1293 if (kvm_on_sigbus_vcpu(env
, siginfo
.si_code
, siginfo
.si_addr
))
1300 r
= sigpending(&chkset
);
1302 printf("sigpending: %s\n", strerror(e
));
1305 } while (sigismember(&chkset
, SIG_IPI
) || sigismember(&chkset
, SIGBUS
));
1307 cpu_single_env
= env
;
1308 flush_queued_work(env
);
1313 pthread_cond_signal(&qemu_pause_cond
);
1316 env
->kvm_cpu_state
.signalled
= 0;
1319 static int all_threads_paused(void)
1321 CPUState
*penv
= first_cpu
;
1327 penv
= (CPUState
*) penv
->next_cpu
;
1333 static void pause_all_threads(void)
1335 CPUState
*penv
= first_cpu
;
1338 if (penv
!= cpu_single_env
) {
1340 pthread_kill(penv
->kvm_cpu_state
.thread
, SIG_IPI
);
1346 penv
= (CPUState
*) penv
->next_cpu
;
1349 while (!all_threads_paused()) {
1350 qemu_cond_wait(&qemu_pause_cond
);
1354 static void resume_all_threads(void)
1356 CPUState
*penv
= first_cpu
;
1358 assert(!cpu_single_env
);
1363 pthread_kill(penv
->kvm_cpu_state
.thread
, SIG_IPI
);
1364 penv
= (CPUState
*) penv
->next_cpu
;
1368 static void kvm_vm_state_change_handler(void *context
, int running
, int reason
)
1371 resume_all_threads();
1373 pause_all_threads();
1377 static void setup_kernel_sigmask(CPUState
*env
)
1382 sigaddset(&set
, SIGUSR2
);
1383 sigaddset(&set
, SIGIO
);
1384 sigaddset(&set
, SIGALRM
);
1385 sigprocmask(SIG_BLOCK
, &set
, NULL
);
1387 sigprocmask(SIG_BLOCK
, NULL
, &set
);
1388 sigdelset(&set
, SIG_IPI
);
1389 sigdelset(&set
, SIGBUS
);
1391 kvm_set_signal_mask(env
, &set
);
1394 static void qemu_kvm_system_reset(void)
1396 pause_all_threads();
1398 qemu_system_reset();
1400 resume_all_threads();
1403 static void process_irqchip_events(CPUState
*env
)
1405 kvm_arch_process_irqchip_events(env
);
1406 if (kvm_arch_has_work(env
))
1410 static int kvm_main_loop_cpu(CPUState
*env
)
1413 int run_cpu
= !kvm_cpu_is_stopped(env
);
1414 if (run_cpu
&& !kvm_irqchip_in_kernel()) {
1415 process_irqchip_events(env
);
1416 run_cpu
= !env
->halted
;
1420 kvm_main_loop_wait(env
, 0);
1422 kvm_main_loop_wait(env
, 1000);
1425 pthread_mutex_unlock(&qemu_mutex
);
1429 static void *ap_main_loop(void *_env
)
1431 CPUState
*env
= _env
;
1433 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
1434 struct ioperm_data
*data
= NULL
;
1438 env
->thread_id
= kvm_get_thread_id();
1439 sigfillset(&signals
);
1440 sigprocmask(SIG_BLOCK
, &signals
, NULL
);
1442 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
1443 /* do ioperm for io ports of assigned devices */
1444 QLIST_FOREACH(data
, &ioperm_head
, entries
)
1445 on_vcpu(env
, kvm_arch_do_ioperm
, data
);
1448 pthread_mutex_lock(&qemu_mutex
);
1449 cpu_single_env
= env
;
1451 kvm_create_vcpu(env
, env
->cpu_index
);
1452 setup_kernel_sigmask(env
);
1454 /* signal VCPU creation */
1455 current_env
->created
= 1;
1456 pthread_cond_signal(&qemu_vcpu_cond
);
1458 /* and wait for machine initialization */
1459 while (!qemu_system_ready
) {
1460 qemu_cond_wait(&qemu_system_cond
);
1463 /* re-initialize cpu_single_env after re-acquiring qemu_mutex */
1464 cpu_single_env
= env
;
1466 kvm_main_loop_cpu(env
);
1470 int kvm_init_vcpu(CPUState
*env
)
1472 pthread_create(&env
->kvm_cpu_state
.thread
, NULL
, ap_main_loop
, env
);
1474 while (env
->created
== 0) {
1475 qemu_cond_wait(&qemu_vcpu_cond
);
1481 int kvm_vcpu_inited(CPUState
*env
)
1483 return env
->created
;
1487 void kvm_hpet_disable_kpit(void)
1489 struct kvm_pit_state2 ps2
;
1491 kvm_get_pit2(kvm_context
, &ps2
);
1492 ps2
.flags
|= KVM_PIT_FLAGS_HPET_LEGACY
;
1493 kvm_set_pit2(kvm_context
, &ps2
);
1496 void kvm_hpet_enable_kpit(void)
1498 struct kvm_pit_state2 ps2
;
1500 kvm_get_pit2(kvm_context
, &ps2
);
1501 ps2
.flags
&= ~KVM_PIT_FLAGS_HPET_LEGACY
;
1502 kvm_set_pit2(kvm_context
, &ps2
);
1506 int kvm_init_ap(void)
1508 struct sigaction action
;
1510 qemu_add_vm_change_state_handler(kvm_vm_state_change_handler
, NULL
);
1512 signal(SIG_IPI
, sig_ipi_handler
);
1514 memset(&action
, 0, sizeof(action
));
1515 action
.sa_flags
= SA_SIGINFO
;
1516 action
.sa_sigaction
= (void (*)(int, siginfo_t
*, void*))sigbus_handler
;
1517 sigaction(SIGBUS
, &action
, NULL
);
1518 prctl(PR_MCE_KILL
, 1, 1, 0, 0);
1522 /* If we have signalfd, we mask out the signals we want to handle and then
1523 * use signalfd to listen for them. We rely on whatever the current signal
1524 * handler is to dispatch the signals when we receive them.
1527 static void sigfd_handler(void *opaque
)
1529 int fd
= (unsigned long) opaque
;
1530 struct qemu_signalfd_siginfo info
;
1531 struct sigaction action
;
1536 len
= read(fd
, &info
, sizeof(info
));
1537 } while (len
== -1 && errno
== EINTR
);
1539 if (len
== -1 && errno
== EAGAIN
) {
1543 if (len
!= sizeof(info
)) {
1544 printf("read from sigfd returned %zd: %m\n", len
);
1548 sigaction(info
.ssi_signo
, NULL
, &action
);
1549 if ((action
.sa_flags
& SA_SIGINFO
) && action
.sa_sigaction
) {
1550 action
.sa_sigaction(info
.ssi_signo
,
1551 (siginfo_t
*)&info
, NULL
);
1552 } else if (action
.sa_handler
) {
1553 action
.sa_handler(info
.ssi_signo
);
1558 int kvm_main_loop(void)
1563 io_thread
= pthread_self();
1564 qemu_system_ready
= 1;
1567 sigaddset(&mask
, SIGIO
);
1568 sigaddset(&mask
, SIGALRM
);
1569 sigaddset(&mask
, SIGBUS
);
1570 sigprocmask(SIG_BLOCK
, &mask
, NULL
);
1572 sigfd
= qemu_signalfd(&mask
);
1574 fprintf(stderr
, "failed to create signalfd\n");
1578 fcntl(sigfd
, F_SETFL
, O_NONBLOCK
);
1580 qemu_set_fd_handler2(sigfd
, NULL
, sigfd_handler
, NULL
,
1581 (void *)(unsigned long) sigfd
);
1583 pthread_cond_broadcast(&qemu_system_cond
);
1585 io_thread_sigfd
= sigfd
;
1586 cpu_single_env
= NULL
;
1590 if (qemu_shutdown_requested()) {
1591 monitor_protocol_event(QEVENT_SHUTDOWN
, NULL
);
1592 if (qemu_no_shutdown()) {
1597 } else if (qemu_powerdown_requested()) {
1598 monitor_protocol_event(QEVENT_POWERDOWN
, NULL
);
1599 qemu_irq_raise(qemu_system_powerdown
);
1600 } else if (qemu_reset_requested()) {
1601 qemu_kvm_system_reset();
1602 } else if (kvm_debug_cpu_requested
) {
1603 gdb_set_stop_cpu(kvm_debug_cpu_requested
);
1604 vm_stop(EXCP_DEBUG
);
1605 kvm_debug_cpu_requested
= NULL
;
1609 pause_all_threads();
1610 pthread_mutex_unlock(&qemu_mutex
);
1615 #if !defined(TARGET_I386)
1616 int kvm_arch_init_irq_routing(void)
1624 static int kvm_create_context(void)
1626 static const char upgrade_note
[] =
1627 "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1628 "(see http://sourceforge.net/projects/kvm).\n";
1633 kvm_disable_irqchip_creation(kvm_context
);
1636 kvm_disable_pit_creation(kvm_context
);
1638 if (kvm_create(kvm_context
, 0, NULL
) < 0) {
1639 kvm_finalize(kvm_state
);
1642 r
= kvm_arch_qemu_create_context();
1644 kvm_finalize(kvm_state
);
1647 if (kvm_pit
&& !kvm_pit_reinject
) {
1648 if (kvm_reinject_control(kvm_context
, 0)) {
1649 fprintf(stderr
, "failure to disable in-kernel PIT reinjection\n");
1654 /* There was a nasty bug in < kvm-80 that prevents memory slots from being
1655 * destroyed properly. Since we rely on this capability, refuse to work
1656 * with any kernel without this capability. */
1657 if (!kvm_check_extension(kvm_state
, KVM_CAP_DESTROY_MEMORY_REGION_WORKS
)) {
1659 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
1664 r
= kvm_arch_init_irq_routing();
1669 kvm_state
->vcpu_events
= 0;
1670 #ifdef KVM_CAP_VCPU_EVENTS
1671 kvm_state
->vcpu_events
= kvm_check_extension(kvm_state
, KVM_CAP_VCPU_EVENTS
);
1674 kvm_state
->debugregs
= 0;
1675 #ifdef KVM_CAP_DEBUGREGS
1676 kvm_state
->debugregs
= kvm_check_extension(kvm_state
, KVM_CAP_DEBUGREGS
);
1679 kvm_state
->xsave
= 0;
1680 #ifdef KVM_CAP_XSAVE
1681 kvm_state
->xsave
= kvm_check_extension(kvm_state
, KVM_CAP_XSAVE
);
1684 kvm_state
->xcrs
= 0;
1686 kvm_state
->xcrs
= kvm_check_extension(kvm_state
, KVM_CAP_XCRS
);
1691 if (!qemu_kvm_has_gsi_routing()) {
1694 /* if kernel can't do irq routing, interrupt source
1695 * override 0->2 can not be set up as required by hpet,
1699 } else if (!qemu_kvm_has_pit_state2()) {
1710 #ifdef KVM_CAP_IRQCHIP
1712 int kvm_set_irq(int irq
, int level
, int *status
)
1714 return kvm_set_irq_level(kvm_context
, irq
, level
, status
);
1719 static void kvm_mutex_unlock(void)
1721 assert(!cpu_single_env
);
1722 pthread_mutex_unlock(&qemu_mutex
);
1725 static void kvm_mutex_lock(void)
1727 pthread_mutex_lock(&qemu_mutex
);
1728 cpu_single_env
= NULL
;
1731 void qemu_mutex_unlock_iothread(void)
1733 if (kvm_enabled()) {
1738 void qemu_mutex_lock_iothread(void)
1740 if (kvm_enabled()) {
1745 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
1746 void kvm_add_ioperm_data(struct ioperm_data
*data
)
1748 QLIST_INSERT_HEAD(&ioperm_head
, data
, entries
);
1751 void kvm_remove_ioperm_data(unsigned long start_port
, unsigned long num
)
1753 struct ioperm_data
*data
;
1755 data
= QLIST_FIRST(&ioperm_head
);
1757 struct ioperm_data
*next
= QLIST_NEXT(data
, entries
);
1759 if (data
->start_port
== start_port
&& data
->num
== num
) {
1760 QLIST_REMOVE(data
, entries
);
1768 void kvm_ioperm(CPUState
*env
, void *data
)
1770 if (kvm_enabled() && qemu_system_ready
) {
1771 on_vcpu(env
, kvm_arch_do_ioperm
, data
);
1777 int kvm_set_boot_cpu_id(uint32_t id
)
1779 return kvm_set_boot_vcpu_id(kvm_context
, id
);