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 <linux/kvm_irqfd.h>
31 #include <linux/irqbypass.h>
32 #include <trace/events/kvm.h>
33 #include <kvm/arm_pmu.h>
34 #include <kvm/arm_psci.h>
36 #define CREATE_TRACE_POINTS
39 #include <linux/uaccess.h>
40 #include <asm/ptrace.h>
42 #include <asm/tlbflush.h>
43 #include <asm/cacheflush.h>
45 #include <asm/kvm_arm.h>
46 #include <asm/kvm_asm.h>
47 #include <asm/kvm_mmu.h>
48 #include <asm/kvm_emulate.h>
49 #include <asm/kvm_coproc.h>
50 #include <asm/sections.h>
53 __asm__(".arch_extension virt");
56 DEFINE_PER_CPU(kvm_cpu_context_t
, kvm_host_cpu_state
);
57 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
59 /* Per-CPU variable containing the currently running vcpu. */
60 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
62 /* The VMID used in the VTTBR */
63 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
64 static u32 kvm_next_vmid
;
65 static unsigned int kvm_vmid_bits __read_mostly
;
66 static DEFINE_SPINLOCK(kvm_vmid_lock
);
68 static bool vgic_present
;
70 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled
);
72 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
74 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
77 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use
);
80 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
81 * Must be called from non-preemptible context
83 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
85 return __this_cpu_read(kvm_arm_running_vcpu
);
89 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
91 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
93 return &kvm_arm_running_vcpu
;
96 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
98 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
101 int kvm_arch_hardware_setup(void)
106 void kvm_arch_check_processor_compat(void *rtn
)
113 * kvm_arch_init_vm - initializes a VM data structure
114 * @kvm: pointer to the KVM struct
116 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
123 kvm
->arch
.last_vcpu_ran
= alloc_percpu(typeof(*kvm
->arch
.last_vcpu_ran
));
124 if (!kvm
->arch
.last_vcpu_ran
)
127 for_each_possible_cpu(cpu
)
128 *per_cpu_ptr(kvm
->arch
.last_vcpu_ran
, cpu
) = -1;
130 ret
= kvm_alloc_stage2_pgd(kvm
);
134 ret
= create_hyp_mappings(kvm
, kvm
+ 1, PAGE_HYP
);
136 goto out_free_stage2_pgd
;
138 kvm_vgic_early_init(kvm
);
140 /* Mark the initial VMID generation invalid */
141 kvm
->arch
.vmid_gen
= 0;
143 /* The maximum number of VCPUs is limited by the host's GIC model */
144 kvm
->arch
.max_vcpus
= vgic_present
?
145 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
149 kvm_free_stage2_pgd(kvm
);
151 free_percpu(kvm
->arch
.last_vcpu_ran
);
152 kvm
->arch
.last_vcpu_ran
= NULL
;
156 bool kvm_arch_has_vcpu_debugfs(void)
161 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
166 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
168 return VM_FAULT_SIGBUS
;
173 * kvm_arch_destroy_vm - destroy the VM data structure
174 * @kvm: pointer to the KVM struct
176 void kvm_arch_destroy_vm(struct kvm
*kvm
)
180 kvm_vgic_destroy(kvm
);
182 free_percpu(kvm
->arch
.last_vcpu_ran
);
183 kvm
->arch
.last_vcpu_ran
= NULL
;
185 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
187 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
188 kvm
->vcpus
[i
] = NULL
;
191 atomic_set(&kvm
->online_vcpus
, 0);
194 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
198 case KVM_CAP_IRQCHIP
:
201 case KVM_CAP_IOEVENTFD
:
202 case KVM_CAP_DEVICE_CTRL
:
203 case KVM_CAP_USER_MEMORY
:
204 case KVM_CAP_SYNC_MMU
:
205 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
206 case KVM_CAP_ONE_REG
:
207 case KVM_CAP_ARM_PSCI
:
208 case KVM_CAP_ARM_PSCI_0_2
:
209 case KVM_CAP_READONLY_MEM
:
210 case KVM_CAP_MP_STATE
:
211 case KVM_CAP_IMMEDIATE_EXIT
:
214 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
217 case KVM_CAP_NR_VCPUS
:
218 r
= num_online_cpus();
220 case KVM_CAP_MAX_VCPUS
:
223 case KVM_CAP_NR_MEMSLOTS
:
224 r
= KVM_USER_MEM_SLOTS
;
226 case KVM_CAP_MSI_DEVID
:
230 r
= kvm
->arch
.vgic
.msis_require_devid
;
232 case KVM_CAP_ARM_USER_IRQ
:
234 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
235 * (bump this number if adding more devices)
240 r
= kvm_arch_dev_ioctl_check_extension(kvm
, ext
);
246 long kvm_arch_dev_ioctl(struct file
*filp
,
247 unsigned int ioctl
, unsigned long arg
)
253 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
256 struct kvm_vcpu
*vcpu
;
258 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
263 if (id
>= kvm
->arch
.max_vcpus
) {
268 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
274 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
278 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
284 kvm_vcpu_uninit(vcpu
);
286 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
291 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
293 kvm_vgic_vcpu_early_init(vcpu
);
296 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
298 if (vcpu
->arch
.has_run_once
&& unlikely(!irqchip_in_kernel(vcpu
->kvm
)))
299 static_branch_dec(&userspace_irqchip_in_use
);
301 kvm_mmu_free_memory_caches(vcpu
);
302 kvm_timer_vcpu_terminate(vcpu
);
303 kvm_pmu_vcpu_destroy(vcpu
);
304 kvm_vcpu_uninit(vcpu
);
305 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
308 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
310 kvm_arch_vcpu_free(vcpu
);
313 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
315 return kvm_timer_is_pending(vcpu
);
318 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
320 kvm_timer_schedule(vcpu
);
321 kvm_vgic_v4_enable_doorbell(vcpu
);
324 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
326 kvm_timer_unschedule(vcpu
);
327 kvm_vgic_v4_disable_doorbell(vcpu
);
330 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
332 /* Force users to call KVM_ARM_VCPU_INIT */
333 vcpu
->arch
.target
= -1;
334 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
336 /* Set up the timer */
337 kvm_timer_vcpu_init(vcpu
);
339 kvm_arm_reset_debug_ptr(vcpu
);
341 return kvm_vgic_vcpu_init(vcpu
);
344 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
348 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
351 * We might get preempted before the vCPU actually runs, but
352 * over-invalidation doesn't affect correctness.
354 if (*last_ran
!= vcpu
->vcpu_id
) {
355 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
356 *last_ran
= vcpu
->vcpu_id
;
360 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(&kvm_host_cpu_state
);
362 kvm_arm_set_running_vcpu(vcpu
);
364 kvm_timer_vcpu_load(vcpu
);
367 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
369 kvm_timer_vcpu_put(vcpu
);
374 kvm_arm_set_running_vcpu(NULL
);
377 static void vcpu_power_off(struct kvm_vcpu
*vcpu
)
379 vcpu
->arch
.power_off
= true;
380 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
384 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
385 struct kvm_mp_state
*mp_state
)
387 if (vcpu
->arch
.power_off
)
388 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
390 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
395 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
396 struct kvm_mp_state
*mp_state
)
400 switch (mp_state
->mp_state
) {
401 case KVM_MP_STATE_RUNNABLE
:
402 vcpu
->arch
.power_off
= false;
404 case KVM_MP_STATE_STOPPED
:
405 vcpu_power_off(vcpu
);
415 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
416 * @v: The VCPU pointer
418 * If the guest CPU is not waiting for interrupts or an interrupt line is
419 * asserted, the CPU is by definition runnable.
421 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
423 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
424 && !v
->arch
.power_off
&& !v
->arch
.pause
);
427 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
429 return vcpu_mode_priv(vcpu
);
432 /* Just ensure a guest exit from a particular CPU */
433 static void exit_vm_noop(void *info
)
437 void force_vm_exit(const cpumask_t
*mask
)
440 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
445 * need_new_vmid_gen - check that the VMID is still valid
446 * @kvm: The VM's VMID to check
448 * return true if there is a new generation of VMIDs being used
450 * The hardware supports only 256 values with the value zero reserved for the
451 * host, so we check if an assigned value belongs to a previous generation,
452 * which which requires us to assign a new value. If we're the first to use a
453 * VMID for the new generation, we must flush necessary caches and TLBs on all
456 static bool need_new_vmid_gen(struct kvm
*kvm
)
458 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
462 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
463 * @kvm The guest that we are about to run
465 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
466 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
469 static void update_vttbr(struct kvm
*kvm
)
471 phys_addr_t pgd_phys
;
474 if (!need_new_vmid_gen(kvm
))
477 spin_lock(&kvm_vmid_lock
);
480 * We need to re-check the vmid_gen here to ensure that if another vcpu
481 * already allocated a valid vmid for this vm, then this vcpu should
484 if (!need_new_vmid_gen(kvm
)) {
485 spin_unlock(&kvm_vmid_lock
);
489 /* First user of a new VMID generation? */
490 if (unlikely(kvm_next_vmid
== 0)) {
491 atomic64_inc(&kvm_vmid_gen
);
495 * On SMP we know no other CPUs can use this CPU's or each
496 * other's VMID after force_vm_exit returns since the
497 * kvm_vmid_lock blocks them from reentry to the guest.
499 force_vm_exit(cpu_all_mask
);
501 * Now broadcast TLB + ICACHE invalidation over the inner
502 * shareable domain to make sure all data structures are
505 kvm_call_hyp(__kvm_flush_vm_context
);
508 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
509 kvm
->arch
.vmid
= kvm_next_vmid
;
511 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
513 /* update vttbr to be used with the new vmid */
514 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
515 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
516 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
517 kvm
->arch
.vttbr
= kvm_phys_to_vttbr(pgd_phys
) | vmid
;
519 spin_unlock(&kvm_vmid_lock
);
522 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
524 struct kvm
*kvm
= vcpu
->kvm
;
527 if (likely(vcpu
->arch
.has_run_once
))
530 vcpu
->arch
.has_run_once
= true;
532 if (likely(irqchip_in_kernel(kvm
))) {
534 * Map the VGIC hardware resources before running a vcpu the
535 * first time on this VM.
537 if (unlikely(!vgic_ready(kvm
))) {
538 ret
= kvm_vgic_map_resources(kvm
);
544 * Tell the rest of the code that there are userspace irqchip
547 static_branch_inc(&userspace_irqchip_in_use
);
550 ret
= kvm_timer_enable(vcpu
);
554 ret
= kvm_arm_pmu_v3_enable(vcpu
);
559 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
561 return vgic_initialized(kvm
);
564 void kvm_arm_halt_guest(struct kvm
*kvm
)
567 struct kvm_vcpu
*vcpu
;
569 kvm_for_each_vcpu(i
, vcpu
, kvm
)
570 vcpu
->arch
.pause
= true;
571 kvm_make_all_cpus_request(kvm
, KVM_REQ_SLEEP
);
574 void kvm_arm_resume_guest(struct kvm
*kvm
)
577 struct kvm_vcpu
*vcpu
;
579 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
580 vcpu
->arch
.pause
= false;
581 swake_up(kvm_arch_vcpu_wq(vcpu
));
585 static void vcpu_req_sleep(struct kvm_vcpu
*vcpu
)
587 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
589 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
590 (!vcpu
->arch
.pause
)));
592 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
593 /* Awaken to handle a signal, request we sleep again later. */
594 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
598 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
600 return vcpu
->arch
.target
>= 0;
603 static void check_vcpu_requests(struct kvm_vcpu
*vcpu
)
605 if (kvm_request_pending(vcpu
)) {
606 if (kvm_check_request(KVM_REQ_SLEEP
, vcpu
))
607 vcpu_req_sleep(vcpu
);
610 * Clear IRQ_PENDING requests that were made to guarantee
611 * that a VCPU sees new virtual interrupts.
613 kvm_check_request(KVM_REQ_IRQ_PENDING
, vcpu
);
618 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
619 * @vcpu: The VCPU pointer
620 * @run: The kvm_run structure pointer used for userspace state exchange
622 * This function is called through the VCPU_RUN ioctl called from user space. It
623 * will execute VM code in a loop until the time slice for the process is used
624 * or some emulation is needed from user space in which case the function will
625 * return with return value 0 and with the kvm_run structure filled in with the
626 * required data for the requested emulation.
628 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
632 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
637 ret
= kvm_vcpu_first_run_init(vcpu
);
641 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
642 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
645 if (kvm_arm_handle_step_debug(vcpu
, vcpu
->run
)) {
652 if (run
->immediate_exit
) {
657 kvm_sigset_activate(vcpu
);
660 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
663 * Check conditions before entering the guest
667 update_vttbr(vcpu
->kvm
);
669 check_vcpu_requests(vcpu
);
672 * Preparing the interrupts to be injected also
673 * involves poking the GIC, which must be done in a
674 * non-preemptible context.
678 /* Flush FP/SIMD state that can't survive guest entry/exit */
679 kvm_fpsimd_flush_cpu_state();
681 kvm_pmu_flush_hwstate(vcpu
);
685 kvm_vgic_flush_hwstate(vcpu
);
688 * Exit if we have a signal pending so that we can deliver the
689 * signal to user space.
691 if (signal_pending(current
)) {
693 run
->exit_reason
= KVM_EXIT_INTR
;
697 * If we're using a userspace irqchip, then check if we need
698 * to tell a userspace irqchip about timer or PMU level
699 * changes and if so, exit to userspace (the actual level
700 * state gets updated in kvm_timer_update_run and
701 * kvm_pmu_update_run below).
703 if (static_branch_unlikely(&userspace_irqchip_in_use
)) {
704 if (kvm_timer_should_notify_user(vcpu
) ||
705 kvm_pmu_should_notify_user(vcpu
)) {
707 run
->exit_reason
= KVM_EXIT_INTR
;
712 * Ensure we set mode to IN_GUEST_MODE after we disable
713 * interrupts and before the final VCPU requests check.
714 * See the comment in kvm_vcpu_exiting_guest_mode() and
715 * Documentation/virtual/kvm/vcpu-requests.rst
717 smp_store_mb(vcpu
->mode
, IN_GUEST_MODE
);
719 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
720 kvm_request_pending(vcpu
)) {
721 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
722 kvm_pmu_sync_hwstate(vcpu
);
723 if (static_branch_unlikely(&userspace_irqchip_in_use
))
724 kvm_timer_sync_hwstate(vcpu
);
725 kvm_vgic_sync_hwstate(vcpu
);
731 kvm_arm_setup_debug(vcpu
);
733 /**************************************************************
736 trace_kvm_entry(*vcpu_pc(vcpu
));
737 guest_enter_irqoff();
739 kvm_arm_vhe_guest_enter();
741 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
744 kvm_arm_vhe_guest_exit();
745 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
749 *************************************************************/
751 kvm_arm_clear_debug(vcpu
);
754 * We must sync the PMU state before the vgic state so
755 * that the vgic can properly sample the updated state of the
758 kvm_pmu_sync_hwstate(vcpu
);
761 * Sync the vgic state before syncing the timer state because
762 * the timer code needs to know if the virtual timer
763 * interrupts are active.
765 kvm_vgic_sync_hwstate(vcpu
);
768 * Sync the timer hardware state before enabling interrupts as
769 * we don't want vtimer interrupts to race with syncing the
770 * timer virtual interrupt state.
772 if (static_branch_unlikely(&userspace_irqchip_in_use
))
773 kvm_timer_sync_hwstate(vcpu
);
776 * We may have taken a host interrupt in HYP mode (ie
777 * while executing the guest). This interrupt is still
778 * pending, as we haven't serviced it yet!
780 * We're now back in SVC mode, with interrupts
781 * disabled. Enabling the interrupts now will have
782 * the effect of taking the interrupt again, in SVC
788 * We do local_irq_enable() before calling guest_exit() so
789 * that if a timer interrupt hits while running the guest we
790 * account that tick as being spent in the guest. We enable
791 * preemption after calling guest_exit() so that if we get
792 * preempted we make sure ticks after that is not counted as
796 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
798 /* Exit types that need handling before we can be preempted */
799 handle_exit_early(vcpu
, run
, ret
);
803 ret
= handle_exit(vcpu
, run
, ret
);
806 /* Tell userspace about in-kernel device output levels */
807 if (unlikely(!irqchip_in_kernel(vcpu
->kvm
))) {
808 kvm_timer_update_run(vcpu
);
809 kvm_pmu_update_run(vcpu
);
812 kvm_sigset_deactivate(vcpu
);
819 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
825 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
826 bit_index
= __ffs(HCR_VI
);
827 else /* KVM_ARM_IRQ_CPU_FIQ */
828 bit_index
= __ffs(HCR_VF
);
830 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
832 set
= test_and_set_bit(bit_index
, ptr
);
834 set
= test_and_clear_bit(bit_index
, ptr
);
837 * If we didn't change anything, no need to wake up or kick other CPUs
843 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
844 * trigger a world-switch round on the running physical CPU to set the
845 * virtual IRQ/FIQ fields in the HCR appropriately.
847 kvm_make_request(KVM_REQ_IRQ_PENDING
, vcpu
);
853 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
856 u32 irq
= irq_level
->irq
;
857 unsigned int irq_type
, vcpu_idx
, irq_num
;
858 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
859 struct kvm_vcpu
*vcpu
= NULL
;
860 bool level
= irq_level
->level
;
862 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
863 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
864 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
866 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
869 case KVM_ARM_IRQ_TYPE_CPU
:
870 if (irqchip_in_kernel(kvm
))
873 if (vcpu_idx
>= nrcpus
)
876 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
880 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
883 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
884 case KVM_ARM_IRQ_TYPE_PPI
:
885 if (!irqchip_in_kernel(kvm
))
888 if (vcpu_idx
>= nrcpus
)
891 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
895 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
898 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
, NULL
);
899 case KVM_ARM_IRQ_TYPE_SPI
:
900 if (!irqchip_in_kernel(kvm
))
903 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
906 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
, NULL
);
912 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
913 const struct kvm_vcpu_init
*init
)
916 int phys_target
= kvm_target_cpu();
918 if (init
->target
!= phys_target
)
922 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
923 * use the same target.
925 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
928 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
929 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
930 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
932 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
936 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
937 * use the same feature set.
939 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
940 test_bit(i
, vcpu
->arch
.features
) != set
)
944 set_bit(i
, vcpu
->arch
.features
);
947 vcpu
->arch
.target
= phys_target
;
949 /* Now we know what it is, we can reset it. */
950 return kvm_reset_vcpu(vcpu
);
954 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
955 struct kvm_vcpu_init
*init
)
959 ret
= kvm_vcpu_set_target(vcpu
, init
);
964 * Ensure a rebooted VM will fault in RAM pages and detect if the
965 * guest MMU is turned off and flush the caches as needed.
967 if (vcpu
->arch
.has_run_once
)
968 stage2_unmap_vm(vcpu
->kvm
);
970 vcpu_reset_hcr(vcpu
);
973 * Handle the "start in power-off" case.
975 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
976 vcpu_power_off(vcpu
);
978 vcpu
->arch
.power_off
= false;
983 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
984 struct kvm_device_attr
*attr
)
988 switch (attr
->group
) {
990 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
997 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
998 struct kvm_device_attr
*attr
)
1002 switch (attr
->group
) {
1004 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
1011 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
1012 struct kvm_device_attr
*attr
)
1016 switch (attr
->group
) {
1018 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
1025 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1026 unsigned int ioctl
, unsigned long arg
)
1028 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1029 void __user
*argp
= (void __user
*)arg
;
1030 struct kvm_device_attr attr
;
1034 case KVM_ARM_VCPU_INIT
: {
1035 struct kvm_vcpu_init init
;
1038 if (copy_from_user(&init
, argp
, sizeof(init
)))
1041 r
= kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
1044 case KVM_SET_ONE_REG
:
1045 case KVM_GET_ONE_REG
: {
1046 struct kvm_one_reg reg
;
1049 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1053 if (copy_from_user(®
, argp
, sizeof(reg
)))
1056 if (ioctl
== KVM_SET_ONE_REG
)
1057 r
= kvm_arm_set_reg(vcpu
, ®
);
1059 r
= kvm_arm_get_reg(vcpu
, ®
);
1062 case KVM_GET_REG_LIST
: {
1063 struct kvm_reg_list __user
*user_list
= argp
;
1064 struct kvm_reg_list reg_list
;
1068 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1072 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
1075 reg_list
.n
= kvm_arm_num_regs(vcpu
);
1076 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
1081 r
= kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
1084 case KVM_SET_DEVICE_ATTR
: {
1086 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1088 r
= kvm_arm_vcpu_set_attr(vcpu
, &attr
);
1091 case KVM_GET_DEVICE_ATTR
: {
1093 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1095 r
= kvm_arm_vcpu_get_attr(vcpu
, &attr
);
1098 case KVM_HAS_DEVICE_ATTR
: {
1100 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1102 r
= kvm_arm_vcpu_has_attr(vcpu
, &attr
);
1113 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1114 * @kvm: kvm instance
1115 * @log: slot id and address to which we copy the log
1117 * Steps 1-4 below provide general overview of dirty page logging. See
1118 * kvm_get_dirty_log_protect() function description for additional details.
1120 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1121 * always flush the TLB (step 4) even if previous step failed and the dirty
1122 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1123 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1124 * writes will be marked dirty for next log read.
1126 * 1. Take a snapshot of the bit and clear it if needed.
1127 * 2. Write protect the corresponding page.
1128 * 3. Copy the snapshot to the userspace.
1129 * 4. Flush TLB's if needed.
1131 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1133 bool is_dirty
= false;
1136 mutex_lock(&kvm
->slots_lock
);
1138 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
1141 kvm_flush_remote_tlbs(kvm
);
1143 mutex_unlock(&kvm
->slots_lock
);
1147 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1148 struct kvm_arm_device_addr
*dev_addr
)
1150 unsigned long dev_id
, type
;
1152 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1153 KVM_ARM_DEVICE_ID_SHIFT
;
1154 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1155 KVM_ARM_DEVICE_TYPE_SHIFT
;
1158 case KVM_ARM_DEVICE_VGIC_V2
:
1161 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1167 long kvm_arch_vm_ioctl(struct file
*filp
,
1168 unsigned int ioctl
, unsigned long arg
)
1170 struct kvm
*kvm
= filp
->private_data
;
1171 void __user
*argp
= (void __user
*)arg
;
1174 case KVM_CREATE_IRQCHIP
: {
1178 mutex_lock(&kvm
->lock
);
1179 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1180 mutex_unlock(&kvm
->lock
);
1183 case KVM_ARM_SET_DEVICE_ADDR
: {
1184 struct kvm_arm_device_addr dev_addr
;
1186 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1188 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1190 case KVM_ARM_PREFERRED_TARGET
: {
1192 struct kvm_vcpu_init init
;
1194 err
= kvm_vcpu_preferred_target(&init
);
1198 if (copy_to_user(argp
, &init
, sizeof(init
)))
1208 static void cpu_init_hyp_mode(void *dummy
)
1210 phys_addr_t pgd_ptr
;
1211 unsigned long hyp_stack_ptr
;
1212 unsigned long stack_page
;
1213 unsigned long vector_ptr
;
1215 /* Switch from the HYP stub to our own HYP init vector */
1216 __hyp_set_vectors(kvm_get_idmap_vector());
1218 pgd_ptr
= kvm_mmu_get_httbr();
1219 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1220 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1221 vector_ptr
= (unsigned long)kvm_get_hyp_vector();
1223 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1224 __cpu_init_stage2();
1226 kvm_arm_init_debug();
1229 static void cpu_hyp_reset(void)
1231 if (!is_kernel_in_hyp_mode())
1232 __hyp_reset_vectors();
1235 static void cpu_hyp_reinit(void)
1239 if (is_kernel_in_hyp_mode()) {
1241 * __cpu_init_stage2() is safe to call even if the PM
1242 * event was cancelled before the CPU was reset.
1244 __cpu_init_stage2();
1245 kvm_timer_init_vhe();
1247 cpu_init_hyp_mode(NULL
);
1251 kvm_vgic_init_cpu_hardware();
1254 static void _kvm_arch_hardware_enable(void *discard
)
1256 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1258 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1262 int kvm_arch_hardware_enable(void)
1264 _kvm_arch_hardware_enable(NULL
);
1268 static void _kvm_arch_hardware_disable(void *discard
)
1270 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1272 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1276 void kvm_arch_hardware_disable(void)
1278 _kvm_arch_hardware_disable(NULL
);
1281 #ifdef CONFIG_CPU_PM
1282 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1287 * kvm_arm_hardware_enabled is left with its old value over
1288 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1293 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1295 * don't update kvm_arm_hardware_enabled here
1296 * so that the hardware will be re-enabled
1297 * when we resume. See below.
1302 case CPU_PM_ENTER_FAILED
:
1304 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1305 /* The hardware was enabled before suspend. */
1315 static struct notifier_block hyp_init_cpu_pm_nb
= {
1316 .notifier_call
= hyp_init_cpu_pm_notifier
,
1319 static void __init
hyp_cpu_pm_init(void)
1321 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1323 static void __init
hyp_cpu_pm_exit(void)
1325 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1328 static inline void hyp_cpu_pm_init(void)
1331 static inline void hyp_cpu_pm_exit(void)
1336 static int init_common_resources(void)
1338 /* set size of VMID supported by CPU */
1339 kvm_vmid_bits
= kvm_get_vmid_bits();
1340 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1345 static int init_subsystems(void)
1350 * Enable hardware so that subsystem initialisation can access EL2.
1352 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1355 * Register CPU lower-power notifier
1360 * Init HYP view of VGIC
1362 err
= kvm_vgic_hyp_init();
1365 vgic_present
= true;
1369 vgic_present
= false;
1377 * Init HYP architected timer support
1379 err
= kvm_timer_hyp_init(vgic_present
);
1384 kvm_coproc_table_init();
1387 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1392 static void teardown_hyp_mode(void)
1397 for_each_possible_cpu(cpu
)
1398 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1403 * Inits Hyp-mode on all online CPUs
1405 static int init_hyp_mode(void)
1411 * Allocate Hyp PGD and setup Hyp identity mapping
1413 err
= kvm_mmu_init();
1418 * Allocate stack pages for Hypervisor-mode
1420 for_each_possible_cpu(cpu
) {
1421 unsigned long stack_page
;
1423 stack_page
= __get_free_page(GFP_KERNEL
);
1429 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1433 * Map the Hyp-code called directly from the host
1435 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1436 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1438 kvm_err("Cannot map world-switch code\n");
1442 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1443 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1445 kvm_err("Cannot map rodata section\n");
1449 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1450 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1452 kvm_err("Cannot map bss section\n");
1456 err
= kvm_map_vectors();
1458 kvm_err("Cannot map vectors\n");
1463 * Map the Hyp stack pages
1465 for_each_possible_cpu(cpu
) {
1466 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1467 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1471 kvm_err("Cannot map hyp stack\n");
1476 for_each_possible_cpu(cpu
) {
1477 kvm_cpu_context_t
*cpu_ctxt
;
1479 cpu_ctxt
= per_cpu_ptr(&kvm_host_cpu_state
, cpu
);
1480 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1483 kvm_err("Cannot map host CPU state: %d\n", err
);
1491 teardown_hyp_mode();
1492 kvm_err("error initializing Hyp mode: %d\n", err
);
1496 static void check_kvm_target_cpu(void *ret
)
1498 *(int *)ret
= kvm_target_cpu();
1501 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1503 struct kvm_vcpu
*vcpu
;
1506 mpidr
&= MPIDR_HWID_BITMASK
;
1507 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1508 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1514 bool kvm_arch_has_irq_bypass(void)
1519 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
1520 struct irq_bypass_producer
*prod
)
1522 struct kvm_kernel_irqfd
*irqfd
=
1523 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1525 return kvm_vgic_v4_set_forwarding(irqfd
->kvm
, prod
->irq
,
1528 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
1529 struct irq_bypass_producer
*prod
)
1531 struct kvm_kernel_irqfd
*irqfd
=
1532 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1534 kvm_vgic_v4_unset_forwarding(irqfd
->kvm
, prod
->irq
,
1538 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer
*cons
)
1540 struct kvm_kernel_irqfd
*irqfd
=
1541 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1543 kvm_arm_halt_guest(irqfd
->kvm
);
1546 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer
*cons
)
1548 struct kvm_kernel_irqfd
*irqfd
=
1549 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1551 kvm_arm_resume_guest(irqfd
->kvm
);
1555 * Initialize Hyp-mode and memory mappings on all CPUs.
1557 int kvm_arch_init(void *opaque
)
1563 if (!is_hyp_mode_available()) {
1564 kvm_info("HYP mode not available\n");
1568 for_each_online_cpu(cpu
) {
1569 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1571 kvm_err("Error, CPU %d not supported!\n", cpu
);
1576 err
= init_common_resources();
1580 in_hyp_mode
= is_kernel_in_hyp_mode();
1583 err
= init_hyp_mode();
1588 err
= init_subsystems();
1593 kvm_info("VHE mode initialized successfully\n");
1595 kvm_info("Hyp mode initialized successfully\n");
1601 teardown_hyp_mode();
1606 /* NOP: Compiling as a module not supported */
1607 void kvm_arch_exit(void)
1609 kvm_perf_teardown();
1612 static int arm_init(void)
1614 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1618 module_init(arm_init
);