2 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
3 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2, as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
19 #include <linux/cpu_pm.h>
20 #include <linux/errno.h>
21 #include <linux/err.h>
22 #include <linux/kvm_host.h>
23 #include <linux/list.h>
24 #include <linux/module.h>
25 #include <linux/vmalloc.h>
27 #include <linux/mman.h>
28 #include <linux/sched.h>
29 #include <linux/kvm.h>
30 #include <trace/events/kvm.h>
31 #include <kvm/arm_pmu.h>
32 #include <kvm/arm_psci.h>
34 #define CREATE_TRACE_POINTS
37 #include <linux/uaccess.h>
38 #include <asm/ptrace.h>
40 #include <asm/tlbflush.h>
41 #include <asm/cacheflush.h>
43 #include <asm/kvm_arm.h>
44 #include <asm/kvm_asm.h>
45 #include <asm/kvm_mmu.h>
46 #include <asm/kvm_emulate.h>
47 #include <asm/kvm_coproc.h>
48 #include <asm/sections.h>
51 __asm__(".arch_extension virt");
54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
55 static kvm_cpu_context_t __percpu
*kvm_host_cpu_state
;
57 /* Per-CPU variable containing the currently running vcpu. */
58 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
60 /* The VMID used in the VTTBR */
61 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
62 static u32 kvm_next_vmid
;
63 static unsigned int kvm_vmid_bits __read_mostly
;
64 static DEFINE_SPINLOCK(kvm_vmid_lock
);
66 static bool vgic_present
;
68 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled
);
70 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
72 BUG_ON(preemptible());
73 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
77 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
78 * Must be called from non-preemptible context
80 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
82 BUG_ON(preemptible());
83 return __this_cpu_read(kvm_arm_running_vcpu
);
87 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
89 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
91 return &kvm_arm_running_vcpu
;
94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
96 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
99 int kvm_arch_hardware_setup(void)
104 void kvm_arch_check_processor_compat(void *rtn
)
111 * kvm_arch_init_vm - initializes a VM data structure
112 * @kvm: pointer to the KVM struct
114 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
121 kvm
->arch
.last_vcpu_ran
= alloc_percpu(typeof(*kvm
->arch
.last_vcpu_ran
));
122 if (!kvm
->arch
.last_vcpu_ran
)
125 for_each_possible_cpu(cpu
)
126 *per_cpu_ptr(kvm
->arch
.last_vcpu_ran
, cpu
) = -1;
128 ret
= kvm_alloc_stage2_pgd(kvm
);
132 ret
= create_hyp_mappings(kvm
, kvm
+ 1, PAGE_HYP
);
134 goto out_free_stage2_pgd
;
136 kvm_vgic_early_init(kvm
);
138 /* Mark the initial VMID generation invalid */
139 kvm
->arch
.vmid_gen
= 0;
141 /* The maximum number of VCPUs is limited by the host's GIC model */
142 kvm
->arch
.max_vcpus
= vgic_present
?
143 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
147 kvm_free_stage2_pgd(kvm
);
149 free_percpu(kvm
->arch
.last_vcpu_ran
);
150 kvm
->arch
.last_vcpu_ran
= NULL
;
154 bool kvm_arch_has_vcpu_debugfs(void)
159 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
164 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
166 return VM_FAULT_SIGBUS
;
171 * kvm_arch_destroy_vm - destroy the VM data structure
172 * @kvm: pointer to the KVM struct
174 void kvm_arch_destroy_vm(struct kvm
*kvm
)
178 free_percpu(kvm
->arch
.last_vcpu_ran
);
179 kvm
->arch
.last_vcpu_ran
= NULL
;
181 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
183 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
184 kvm
->vcpus
[i
] = NULL
;
188 kvm_vgic_destroy(kvm
);
191 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
195 case KVM_CAP_IRQCHIP
:
198 case KVM_CAP_IOEVENTFD
:
199 case KVM_CAP_DEVICE_CTRL
:
200 case KVM_CAP_USER_MEMORY
:
201 case KVM_CAP_SYNC_MMU
:
202 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
203 case KVM_CAP_ONE_REG
:
204 case KVM_CAP_ARM_PSCI
:
205 case KVM_CAP_ARM_PSCI_0_2
:
206 case KVM_CAP_READONLY_MEM
:
207 case KVM_CAP_MP_STATE
:
208 case KVM_CAP_IMMEDIATE_EXIT
:
211 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
214 case KVM_CAP_NR_VCPUS
:
215 r
= num_online_cpus();
217 case KVM_CAP_MAX_VCPUS
:
220 case KVM_CAP_NR_MEMSLOTS
:
221 r
= KVM_USER_MEM_SLOTS
;
223 case KVM_CAP_MSI_DEVID
:
227 r
= kvm
->arch
.vgic
.msis_require_devid
;
229 case KVM_CAP_ARM_USER_IRQ
:
231 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
232 * (bump this number if adding more devices)
237 r
= kvm_arch_dev_ioctl_check_extension(kvm
, ext
);
243 long kvm_arch_dev_ioctl(struct file
*filp
,
244 unsigned int ioctl
, unsigned long arg
)
250 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
253 struct kvm_vcpu
*vcpu
;
255 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
260 if (id
>= kvm
->arch
.max_vcpus
) {
265 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
271 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
275 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
281 kvm_vcpu_uninit(vcpu
);
283 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
288 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
290 kvm_vgic_vcpu_early_init(vcpu
);
293 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
295 kvm_mmu_free_memory_caches(vcpu
);
296 kvm_timer_vcpu_terminate(vcpu
);
297 kvm_vgic_vcpu_destroy(vcpu
);
298 kvm_pmu_vcpu_destroy(vcpu
);
299 kvm_vcpu_uninit(vcpu
);
300 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
303 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
305 kvm_arch_vcpu_free(vcpu
);
308 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
310 return kvm_timer_should_fire(vcpu_vtimer(vcpu
)) ||
311 kvm_timer_should_fire(vcpu_ptimer(vcpu
));
314 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
316 kvm_timer_schedule(vcpu
);
319 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
321 kvm_timer_unschedule(vcpu
);
324 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
326 /* Force users to call KVM_ARM_VCPU_INIT */
327 vcpu
->arch
.target
= -1;
328 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
330 /* Set up the timer */
331 kvm_timer_vcpu_init(vcpu
);
333 kvm_arm_reset_debug_ptr(vcpu
);
335 return kvm_vgic_vcpu_init(vcpu
);
338 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
342 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
345 * We might get preempted before the vCPU actually runs, but
346 * over-invalidation doesn't affect correctness.
348 if (*last_ran
!= vcpu
->vcpu_id
) {
349 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
350 *last_ran
= vcpu
->vcpu_id
;
354 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
356 kvm_arm_set_running_vcpu(vcpu
);
361 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
367 kvm_arm_set_running_vcpu(NULL
);
368 kvm_timer_vcpu_put(vcpu
);
371 static void vcpu_power_off(struct kvm_vcpu
*vcpu
)
373 vcpu
->arch
.power_off
= true;
374 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
378 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
379 struct kvm_mp_state
*mp_state
)
381 if (vcpu
->arch
.power_off
)
382 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
384 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
389 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
390 struct kvm_mp_state
*mp_state
)
392 switch (mp_state
->mp_state
) {
393 case KVM_MP_STATE_RUNNABLE
:
394 vcpu
->arch
.power_off
= false;
396 case KVM_MP_STATE_STOPPED
:
397 vcpu_power_off(vcpu
);
407 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
408 * @v: The VCPU pointer
410 * If the guest CPU is not waiting for interrupts or an interrupt line is
411 * asserted, the CPU is by definition runnable.
413 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
415 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
416 && !v
->arch
.power_off
&& !v
->arch
.pause
);
419 /* Just ensure a guest exit from a particular CPU */
420 static void exit_vm_noop(void *info
)
424 void force_vm_exit(const cpumask_t
*mask
)
427 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
432 * need_new_vmid_gen - check that the VMID is still valid
433 * @kvm: The VM's VMID to check
435 * return true if there is a new generation of VMIDs being used
437 * The hardware supports only 256 values with the value zero reserved for the
438 * host, so we check if an assigned value belongs to a previous generation,
439 * which which requires us to assign a new value. If we're the first to use a
440 * VMID for the new generation, we must flush necessary caches and TLBs on all
443 static bool need_new_vmid_gen(struct kvm
*kvm
)
445 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
449 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
450 * @kvm The guest that we are about to run
452 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
453 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
456 static void update_vttbr(struct kvm
*kvm
)
458 phys_addr_t pgd_phys
;
461 if (!need_new_vmid_gen(kvm
))
464 spin_lock(&kvm_vmid_lock
);
467 * We need to re-check the vmid_gen here to ensure that if another vcpu
468 * already allocated a valid vmid for this vm, then this vcpu should
471 if (!need_new_vmid_gen(kvm
)) {
472 spin_unlock(&kvm_vmid_lock
);
476 /* First user of a new VMID generation? */
477 if (unlikely(kvm_next_vmid
== 0)) {
478 atomic64_inc(&kvm_vmid_gen
);
482 * On SMP we know no other CPUs can use this CPU's or each
483 * other's VMID after force_vm_exit returns since the
484 * kvm_vmid_lock blocks them from reentry to the guest.
486 force_vm_exit(cpu_all_mask
);
488 * Now broadcast TLB + ICACHE invalidation over the inner
489 * shareable domain to make sure all data structures are
492 kvm_call_hyp(__kvm_flush_vm_context
);
495 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
496 kvm
->arch
.vmid
= kvm_next_vmid
;
498 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
500 /* update vttbr to be used with the new vmid */
501 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
502 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
503 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
504 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
506 spin_unlock(&kvm_vmid_lock
);
509 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
511 struct kvm
*kvm
= vcpu
->kvm
;
514 if (likely(vcpu
->arch
.has_run_once
))
517 vcpu
->arch
.has_run_once
= true;
520 * Map the VGIC hardware resources before running a vcpu the first
523 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
524 ret
= kvm_vgic_map_resources(kvm
);
529 ret
= kvm_timer_enable(vcpu
);
533 ret
= kvm_arm_pmu_v3_enable(vcpu
);
538 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
540 return vgic_initialized(kvm
);
543 void kvm_arm_halt_guest(struct kvm
*kvm
)
546 struct kvm_vcpu
*vcpu
;
548 kvm_for_each_vcpu(i
, vcpu
, kvm
)
549 vcpu
->arch
.pause
= true;
550 kvm_make_all_cpus_request(kvm
, KVM_REQ_SLEEP
);
553 void kvm_arm_resume_guest(struct kvm
*kvm
)
556 struct kvm_vcpu
*vcpu
;
558 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
559 vcpu
->arch
.pause
= false;
560 swake_up(kvm_arch_vcpu_wq(vcpu
));
564 static void vcpu_req_sleep(struct kvm_vcpu
*vcpu
)
566 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
568 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
569 (!vcpu
->arch
.pause
)));
571 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
572 /* Awaken to handle a signal, request we sleep again later. */
573 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
577 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
579 return vcpu
->arch
.target
>= 0;
582 static void check_vcpu_requests(struct kvm_vcpu
*vcpu
)
584 if (kvm_request_pending(vcpu
)) {
585 if (kvm_check_request(KVM_REQ_SLEEP
, vcpu
))
586 vcpu_req_sleep(vcpu
);
589 * Clear IRQ_PENDING requests that were made to guarantee
590 * that a VCPU sees new virtual interrupts.
592 kvm_check_request(KVM_REQ_IRQ_PENDING
, vcpu
);
597 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
598 * @vcpu: The VCPU pointer
599 * @run: The kvm_run structure pointer used for userspace state exchange
601 * This function is called through the VCPU_RUN ioctl called from user space. It
602 * will execute VM code in a loop until the time slice for the process is used
603 * or some emulation is needed from user space in which case the function will
604 * return with return value 0 and with the kvm_run structure filled in with the
605 * required data for the requested emulation.
607 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
612 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
615 ret
= kvm_vcpu_first_run_init(vcpu
);
619 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
620 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
625 if (run
->immediate_exit
)
628 if (vcpu
->sigset_active
)
629 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
632 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
635 * Check conditions before entering the guest
639 update_vttbr(vcpu
->kvm
);
641 check_vcpu_requests(vcpu
);
644 * Preparing the interrupts to be injected also
645 * involves poking the GIC, which must be done in a
646 * non-preemptible context.
650 kvm_pmu_flush_hwstate(vcpu
);
652 kvm_timer_flush_hwstate(vcpu
);
653 kvm_vgic_flush_hwstate(vcpu
);
658 * If we have a singal pending, or need to notify a userspace
659 * irqchip about timer or PMU level changes, then we exit (and
660 * update the timer level state in kvm_timer_update_run
663 if (signal_pending(current
) ||
664 kvm_timer_should_notify_user(vcpu
) ||
665 kvm_pmu_should_notify_user(vcpu
)) {
667 run
->exit_reason
= KVM_EXIT_INTR
;
671 * Ensure we set mode to IN_GUEST_MODE after we disable
672 * interrupts and before the final VCPU requests check.
673 * See the comment in kvm_vcpu_exiting_guest_mode() and
674 * Documentation/virtual/kvm/vcpu-requests.rst
676 smp_store_mb(vcpu
->mode
, IN_GUEST_MODE
);
678 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
679 kvm_request_pending(vcpu
)) {
680 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
682 kvm_pmu_sync_hwstate(vcpu
);
683 kvm_timer_sync_hwstate(vcpu
);
684 kvm_vgic_sync_hwstate(vcpu
);
689 kvm_arm_setup_debug(vcpu
);
691 /**************************************************************
694 trace_kvm_entry(*vcpu_pc(vcpu
));
695 guest_enter_irqoff();
697 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
699 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
703 *************************************************************/
705 kvm_arm_clear_debug(vcpu
);
708 * We may have taken a host interrupt in HYP mode (ie
709 * while executing the guest). This interrupt is still
710 * pending, as we haven't serviced it yet!
712 * We're now back in SVC mode, with interrupts
713 * disabled. Enabling the interrupts now will have
714 * the effect of taking the interrupt again, in SVC
720 * We do local_irq_enable() before calling guest_exit() so
721 * that if a timer interrupt hits while running the guest we
722 * account that tick as being spent in the guest. We enable
723 * preemption after calling guest_exit() so that if we get
724 * preempted we make sure ticks after that is not counted as
728 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
731 * We must sync the PMU and timer state before the vgic state so
732 * that the vgic can properly sample the updated state of the
735 kvm_pmu_sync_hwstate(vcpu
);
736 kvm_timer_sync_hwstate(vcpu
);
738 kvm_vgic_sync_hwstate(vcpu
);
742 ret
= handle_exit(vcpu
, run
, ret
);
745 /* Tell userspace about in-kernel device output levels */
746 if (unlikely(!irqchip_in_kernel(vcpu
->kvm
))) {
747 kvm_timer_update_run(vcpu
);
748 kvm_pmu_update_run(vcpu
);
751 if (vcpu
->sigset_active
)
752 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
756 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
762 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
763 bit_index
= __ffs(HCR_VI
);
764 else /* KVM_ARM_IRQ_CPU_FIQ */
765 bit_index
= __ffs(HCR_VF
);
767 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
769 set
= test_and_set_bit(bit_index
, ptr
);
771 set
= test_and_clear_bit(bit_index
, ptr
);
774 * If we didn't change anything, no need to wake up or kick other CPUs
780 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
781 * trigger a world-switch round on the running physical CPU to set the
782 * virtual IRQ/FIQ fields in the HCR appropriately.
784 kvm_make_request(KVM_REQ_IRQ_PENDING
, vcpu
);
790 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
793 u32 irq
= irq_level
->irq
;
794 unsigned int irq_type
, vcpu_idx
, irq_num
;
795 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
796 struct kvm_vcpu
*vcpu
= NULL
;
797 bool level
= irq_level
->level
;
799 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
800 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
801 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
803 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
806 case KVM_ARM_IRQ_TYPE_CPU
:
807 if (irqchip_in_kernel(kvm
))
810 if (vcpu_idx
>= nrcpus
)
813 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
817 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
820 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
821 case KVM_ARM_IRQ_TYPE_PPI
:
822 if (!irqchip_in_kernel(kvm
))
825 if (vcpu_idx
>= nrcpus
)
828 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
832 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
835 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
, NULL
);
836 case KVM_ARM_IRQ_TYPE_SPI
:
837 if (!irqchip_in_kernel(kvm
))
840 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
843 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
, NULL
);
849 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
850 const struct kvm_vcpu_init
*init
)
853 int phys_target
= kvm_target_cpu();
855 if (init
->target
!= phys_target
)
859 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
860 * use the same target.
862 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
865 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
866 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
867 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
869 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
873 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
874 * use the same feature set.
876 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
877 test_bit(i
, vcpu
->arch
.features
) != set
)
881 set_bit(i
, vcpu
->arch
.features
);
884 vcpu
->arch
.target
= phys_target
;
886 /* Now we know what it is, we can reset it. */
887 return kvm_reset_vcpu(vcpu
);
891 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
892 struct kvm_vcpu_init
*init
)
896 ret
= kvm_vcpu_set_target(vcpu
, init
);
901 * Ensure a rebooted VM will fault in RAM pages and detect if the
902 * guest MMU is turned off and flush the caches as needed.
904 if (vcpu
->arch
.has_run_once
)
905 stage2_unmap_vm(vcpu
->kvm
);
907 vcpu_reset_hcr(vcpu
);
910 * Handle the "start in power-off" case.
912 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
913 vcpu_power_off(vcpu
);
915 vcpu
->arch
.power_off
= false;
920 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
921 struct kvm_device_attr
*attr
)
925 switch (attr
->group
) {
927 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
934 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
935 struct kvm_device_attr
*attr
)
939 switch (attr
->group
) {
941 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
948 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
949 struct kvm_device_attr
*attr
)
953 switch (attr
->group
) {
955 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
962 long kvm_arch_vcpu_ioctl(struct file
*filp
,
963 unsigned int ioctl
, unsigned long arg
)
965 struct kvm_vcpu
*vcpu
= filp
->private_data
;
966 void __user
*argp
= (void __user
*)arg
;
967 struct kvm_device_attr attr
;
970 case KVM_ARM_VCPU_INIT
: {
971 struct kvm_vcpu_init init
;
973 if (copy_from_user(&init
, argp
, sizeof(init
)))
976 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
978 case KVM_SET_ONE_REG
:
979 case KVM_GET_ONE_REG
: {
980 struct kvm_one_reg reg
;
982 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
985 if (copy_from_user(®
, argp
, sizeof(reg
)))
987 if (ioctl
== KVM_SET_ONE_REG
)
988 return kvm_arm_set_reg(vcpu
, ®
);
990 return kvm_arm_get_reg(vcpu
, ®
);
992 case KVM_GET_REG_LIST
: {
993 struct kvm_reg_list __user
*user_list
= argp
;
994 struct kvm_reg_list reg_list
;
997 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1000 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
1003 reg_list
.n
= kvm_arm_num_regs(vcpu
);
1004 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
1008 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
1010 case KVM_SET_DEVICE_ATTR
: {
1011 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1013 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
1015 case KVM_GET_DEVICE_ATTR
: {
1016 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1018 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
1020 case KVM_HAS_DEVICE_ATTR
: {
1021 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1023 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
1031 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1032 * @kvm: kvm instance
1033 * @log: slot id and address to which we copy the log
1035 * Steps 1-4 below provide general overview of dirty page logging. See
1036 * kvm_get_dirty_log_protect() function description for additional details.
1038 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1039 * always flush the TLB (step 4) even if previous step failed and the dirty
1040 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1041 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1042 * writes will be marked dirty for next log read.
1044 * 1. Take a snapshot of the bit and clear it if needed.
1045 * 2. Write protect the corresponding page.
1046 * 3. Copy the snapshot to the userspace.
1047 * 4. Flush TLB's if needed.
1049 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1051 bool is_dirty
= false;
1054 mutex_lock(&kvm
->slots_lock
);
1056 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
1059 kvm_flush_remote_tlbs(kvm
);
1061 mutex_unlock(&kvm
->slots_lock
);
1065 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1066 struct kvm_arm_device_addr
*dev_addr
)
1068 unsigned long dev_id
, type
;
1070 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1071 KVM_ARM_DEVICE_ID_SHIFT
;
1072 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1073 KVM_ARM_DEVICE_TYPE_SHIFT
;
1076 case KVM_ARM_DEVICE_VGIC_V2
:
1079 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1085 long kvm_arch_vm_ioctl(struct file
*filp
,
1086 unsigned int ioctl
, unsigned long arg
)
1088 struct kvm
*kvm
= filp
->private_data
;
1089 void __user
*argp
= (void __user
*)arg
;
1092 case KVM_CREATE_IRQCHIP
: {
1096 mutex_lock(&kvm
->lock
);
1097 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1098 mutex_unlock(&kvm
->lock
);
1101 case KVM_ARM_SET_DEVICE_ADDR
: {
1102 struct kvm_arm_device_addr dev_addr
;
1104 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1106 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1108 case KVM_ARM_PREFERRED_TARGET
: {
1110 struct kvm_vcpu_init init
;
1112 err
= kvm_vcpu_preferred_target(&init
);
1116 if (copy_to_user(argp
, &init
, sizeof(init
)))
1126 static void cpu_init_hyp_mode(void *dummy
)
1128 phys_addr_t pgd_ptr
;
1129 unsigned long hyp_stack_ptr
;
1130 unsigned long stack_page
;
1131 unsigned long vector_ptr
;
1133 /* Switch from the HYP stub to our own HYP init vector */
1134 __hyp_set_vectors(kvm_get_idmap_vector());
1136 pgd_ptr
= kvm_mmu_get_httbr();
1137 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1138 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1139 vector_ptr
= (unsigned long)kvm_get_hyp_vector();
1141 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1142 __cpu_init_stage2();
1144 kvm_arm_init_debug();
1147 static void cpu_hyp_reset(void)
1149 if (!is_kernel_in_hyp_mode())
1150 __hyp_reset_vectors();
1153 static void cpu_hyp_reinit(void)
1157 if (is_kernel_in_hyp_mode()) {
1159 * __cpu_init_stage2() is safe to call even if the PM
1160 * event was cancelled before the CPU was reset.
1162 __cpu_init_stage2();
1163 kvm_timer_init_vhe();
1165 cpu_init_hyp_mode(NULL
);
1169 kvm_vgic_init_cpu_hardware();
1172 static void _kvm_arch_hardware_enable(void *discard
)
1174 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1176 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1180 int kvm_arch_hardware_enable(void)
1182 _kvm_arch_hardware_enable(NULL
);
1186 static void _kvm_arch_hardware_disable(void *discard
)
1188 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1190 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1194 void kvm_arch_hardware_disable(void)
1196 _kvm_arch_hardware_disable(NULL
);
1199 #ifdef CONFIG_CPU_PM
1200 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1205 * kvm_arm_hardware_enabled is left with its old value over
1206 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1211 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1213 * don't update kvm_arm_hardware_enabled here
1214 * so that the hardware will be re-enabled
1215 * when we resume. See below.
1221 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1222 /* The hardware was enabled before suspend. */
1232 static struct notifier_block hyp_init_cpu_pm_nb
= {
1233 .notifier_call
= hyp_init_cpu_pm_notifier
,
1236 static void __init
hyp_cpu_pm_init(void)
1238 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1240 static void __init
hyp_cpu_pm_exit(void)
1242 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1245 static inline void hyp_cpu_pm_init(void)
1248 static inline void hyp_cpu_pm_exit(void)
1253 static void teardown_common_resources(void)
1255 free_percpu(kvm_host_cpu_state
);
1258 static int init_common_resources(void)
1260 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1261 if (!kvm_host_cpu_state
) {
1262 kvm_err("Cannot allocate host CPU state\n");
1266 /* set size of VMID supported by CPU */
1267 kvm_vmid_bits
= kvm_get_vmid_bits();
1268 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1273 static int init_subsystems(void)
1278 * Enable hardware so that subsystem initialisation can access EL2.
1280 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1283 * Register CPU lower-power notifier
1288 * Init HYP view of VGIC
1290 err
= kvm_vgic_hyp_init();
1293 vgic_present
= true;
1297 vgic_present
= false;
1305 * Init HYP architected timer support
1307 err
= kvm_timer_hyp_init();
1312 kvm_coproc_table_init();
1315 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1320 static void teardown_hyp_mode(void)
1324 if (is_kernel_in_hyp_mode())
1328 for_each_possible_cpu(cpu
)
1329 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1333 static int init_vhe_mode(void)
1335 kvm_info("VHE mode initialized successfully\n");
1340 * Inits Hyp-mode on all online CPUs
1342 static int init_hyp_mode(void)
1348 * Allocate Hyp PGD and setup Hyp identity mapping
1350 err
= kvm_mmu_init();
1355 * Allocate stack pages for Hypervisor-mode
1357 for_each_possible_cpu(cpu
) {
1358 unsigned long stack_page
;
1360 stack_page
= __get_free_page(GFP_KERNEL
);
1366 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1370 * Map the Hyp-code called directly from the host
1372 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1373 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1375 kvm_err("Cannot map world-switch code\n");
1379 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1380 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1382 kvm_err("Cannot map rodata section\n");
1386 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1387 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1389 kvm_err("Cannot map bss section\n");
1393 err
= kvm_map_vectors();
1395 kvm_err("Cannot map vectors\n");
1400 * Map the Hyp stack pages
1402 for_each_possible_cpu(cpu
) {
1403 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1404 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1408 kvm_err("Cannot map hyp stack\n");
1413 for_each_possible_cpu(cpu
) {
1414 kvm_cpu_context_t
*cpu_ctxt
;
1416 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1417 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1420 kvm_err("Cannot map host CPU state: %d\n", err
);
1425 kvm_info("Hyp mode initialized successfully\n");
1430 teardown_hyp_mode();
1431 kvm_err("error initializing Hyp mode: %d\n", err
);
1435 static void check_kvm_target_cpu(void *ret
)
1437 *(int *)ret
= kvm_target_cpu();
1440 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1442 struct kvm_vcpu
*vcpu
;
1445 mpidr
&= MPIDR_HWID_BITMASK
;
1446 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1447 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1454 * Initialize Hyp-mode and memory mappings on all CPUs.
1456 int kvm_arch_init(void *opaque
)
1461 if (!is_hyp_mode_available()) {
1462 kvm_err("HYP mode not available\n");
1466 for_each_online_cpu(cpu
) {
1467 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1469 kvm_err("Error, CPU %d not supported!\n", cpu
);
1474 err
= init_common_resources();
1478 if (is_kernel_in_hyp_mode())
1479 err
= init_vhe_mode();
1481 err
= init_hyp_mode();
1485 err
= init_subsystems();
1492 teardown_hyp_mode();
1494 teardown_common_resources();
1498 /* NOP: Compiling as a module not supported */
1499 void kvm_arch_exit(void)
1501 kvm_perf_teardown();
1504 static int arm_init(void)
1506 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1510 module_init(arm_init
);