1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
8 #include <linux/cpu_pm.h>
9 #include <linux/errno.h>
10 #include <linux/err.h>
11 #include <linux/kvm_host.h>
12 #include <linux/list.h>
13 #include <linux/module.h>
14 #include <linux/vmalloc.h>
16 #include <linux/mman.h>
17 #include <linux/sched.h>
18 #include <linux/kvm.h>
19 #include <linux/kvm_irqfd.h>
20 #include <linux/irqbypass.h>
21 #include <linux/sched/stat.h>
22 #include <trace/events/kvm.h>
23 #include <kvm/arm_pmu.h>
24 #include <kvm/arm_psci.h>
26 #define CREATE_TRACE_POINTS
29 #include <linux/uaccess.h>
30 #include <asm/ptrace.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cacheflush.h>
34 #include <asm/cpufeature.h>
36 #include <asm/kvm_arm.h>
37 #include <asm/kvm_asm.h>
38 #include <asm/kvm_mmu.h>
39 #include <asm/kvm_emulate.h>
40 #include <asm/kvm_coproc.h>
41 #include <asm/sections.h>
44 __asm__(".arch_extension virt");
47 DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data);
48 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
50 /* Per-CPU variable containing the currently running vcpu. */
51 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
53 /* The VMID used in the VTTBR */
54 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
55 static u32 kvm_next_vmid;
56 static DEFINE_SPINLOCK(kvm_vmid_lock);
58 static bool vgic_present;
60 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
62 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
64 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
67 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
70 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
71 * Must be called from non-preemptible context
73 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
75 return __this_cpu_read(kvm_arm_running_vcpu);
79 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
81 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
83 return &kvm_arm_running_vcpu;
86 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
88 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
91 int kvm_arch_hardware_setup(void)
96 int kvm_arch_check_processor_compat(void)
103 * kvm_arch_init_vm - initializes a VM data structure
104 * @kvm: pointer to the KVM struct
106 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
110 ret = kvm_arm_setup_stage2(kvm, type);
114 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
115 if (!kvm->arch.last_vcpu_ran)
118 for_each_possible_cpu(cpu)
119 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
121 ret = kvm_alloc_stage2_pgd(kvm);
125 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
127 goto out_free_stage2_pgd;
129 kvm_vgic_early_init(kvm);
131 /* Mark the initial VMID generation invalid */
132 kvm->arch.vmid.vmid_gen = 0;
134 /* The maximum number of VCPUs is limited by the host's GIC model */
135 kvm->arch.max_vcpus = vgic_present ?
136 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
140 kvm_free_stage2_pgd(kvm);
142 free_percpu(kvm->arch.last_vcpu_ran);
143 kvm->arch.last_vcpu_ran = NULL;
147 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
152 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
154 return VM_FAULT_SIGBUS;
159 * kvm_arch_destroy_vm - destroy the VM data structure
160 * @kvm: pointer to the KVM struct
162 void kvm_arch_destroy_vm(struct kvm *kvm)
166 kvm_vgic_destroy(kvm);
168 free_percpu(kvm->arch.last_vcpu_ran);
169 kvm->arch.last_vcpu_ran = NULL;
171 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
173 kvm_arch_vcpu_free(kvm->vcpus[i]);
174 kvm->vcpus[i] = NULL;
177 atomic_set(&kvm->online_vcpus, 0);
180 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
184 case KVM_CAP_IRQCHIP:
187 case KVM_CAP_IOEVENTFD:
188 case KVM_CAP_DEVICE_CTRL:
189 case KVM_CAP_USER_MEMORY:
190 case KVM_CAP_SYNC_MMU:
191 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
192 case KVM_CAP_ONE_REG:
193 case KVM_CAP_ARM_PSCI:
194 case KVM_CAP_ARM_PSCI_0_2:
195 case KVM_CAP_READONLY_MEM:
196 case KVM_CAP_MP_STATE:
197 case KVM_CAP_IMMEDIATE_EXIT:
198 case KVM_CAP_VCPU_EVENTS:
201 case KVM_CAP_ARM_SET_DEVICE_ADDR:
204 case KVM_CAP_NR_VCPUS:
205 r = num_online_cpus();
207 case KVM_CAP_MAX_VCPUS:
210 case KVM_CAP_MAX_VCPU_ID:
213 case KVM_CAP_MSI_DEVID:
217 r = kvm->arch.vgic.msis_require_devid;
219 case KVM_CAP_ARM_USER_IRQ:
221 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
222 * (bump this number if adding more devices)
227 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
233 long kvm_arch_dev_ioctl(struct file *filp,
234 unsigned int ioctl, unsigned long arg)
239 struct kvm *kvm_arch_alloc_vm(void)
242 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
244 return vzalloc(sizeof(struct kvm));
247 void kvm_arch_free_vm(struct kvm *kvm)
255 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
258 struct kvm_vcpu *vcpu;
260 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
265 if (id >= kvm->arch.max_vcpus) {
270 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
276 err = kvm_vcpu_init(vcpu, kvm, id);
280 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
286 kvm_vcpu_uninit(vcpu);
288 kmem_cache_free(kvm_vcpu_cache, vcpu);
293 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
297 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
299 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
300 static_branch_dec(&userspace_irqchip_in_use);
302 kvm_mmu_free_memory_caches(vcpu);
303 kvm_timer_vcpu_terminate(vcpu);
304 kvm_pmu_vcpu_destroy(vcpu);
305 kvm_vcpu_uninit(vcpu);
306 kmem_cache_free(kvm_vcpu_cache, vcpu);
309 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
311 kvm_arch_vcpu_free(vcpu);
314 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
316 return kvm_timer_is_pending(vcpu);
319 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
322 * If we're about to block (most likely because we've just hit a
323 * WFI), we need to sync back the state of the GIC CPU interface
324 * so that we have the lastest PMR and group enables. This ensures
325 * that kvm_arch_vcpu_runnable has up-to-date data to decide
326 * whether we have pending interrupts.
329 kvm_vgic_vmcr_sync(vcpu);
332 kvm_vgic_v4_enable_doorbell(vcpu);
335 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
337 kvm_vgic_v4_disable_doorbell(vcpu);
340 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
342 /* Force users to call KVM_ARM_VCPU_INIT */
343 vcpu->arch.target = -1;
344 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
346 /* Set up the timer */
347 kvm_timer_vcpu_init(vcpu);
349 kvm_pmu_vcpu_init(vcpu);
351 kvm_arm_reset_debug_ptr(vcpu);
353 return kvm_vgic_vcpu_init(vcpu);
356 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
359 kvm_host_data_t *cpu_data;
361 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
362 cpu_data = this_cpu_ptr(&kvm_host_data);
365 * We might get preempted before the vCPU actually runs, but
366 * over-invalidation doesn't affect correctness.
368 if (*last_ran != vcpu->vcpu_id) {
369 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
370 *last_ran = vcpu->vcpu_id;
374 vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
376 kvm_arm_set_running_vcpu(vcpu);
378 kvm_timer_vcpu_load(vcpu);
379 kvm_vcpu_load_sysregs(vcpu);
380 kvm_arch_vcpu_load_fp(vcpu);
381 kvm_vcpu_pmu_restore_guest(vcpu);
383 if (single_task_running())
384 vcpu_clear_wfe_traps(vcpu);
386 vcpu_set_wfe_traps(vcpu);
388 vcpu_ptrauth_setup_lazy(vcpu);
391 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
393 kvm_arch_vcpu_put_fp(vcpu);
394 kvm_vcpu_put_sysregs(vcpu);
395 kvm_timer_vcpu_put(vcpu);
397 kvm_vcpu_pmu_restore_host(vcpu);
401 kvm_arm_set_running_vcpu(NULL);
404 static void vcpu_power_off(struct kvm_vcpu *vcpu)
406 vcpu->arch.power_off = true;
407 kvm_make_request(KVM_REQ_SLEEP, vcpu);
411 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
412 struct kvm_mp_state *mp_state)
414 if (vcpu->arch.power_off)
415 mp_state->mp_state = KVM_MP_STATE_STOPPED;
417 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
422 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
423 struct kvm_mp_state *mp_state)
427 switch (mp_state->mp_state) {
428 case KVM_MP_STATE_RUNNABLE:
429 vcpu->arch.power_off = false;
431 case KVM_MP_STATE_STOPPED:
432 vcpu_power_off(vcpu);
442 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
443 * @v: The VCPU pointer
445 * If the guest CPU is not waiting for interrupts or an interrupt line is
446 * asserted, the CPU is by definition runnable.
448 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
450 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
451 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
452 && !v->arch.power_off && !v->arch.pause);
455 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
457 return vcpu_mode_priv(vcpu);
460 /* Just ensure a guest exit from a particular CPU */
461 static void exit_vm_noop(void *info)
465 void force_vm_exit(const cpumask_t *mask)
468 smp_call_function_many(mask, exit_vm_noop, NULL, true);
473 * need_new_vmid_gen - check that the VMID is still valid
474 * @vmid: The VMID to check
476 * return true if there is a new generation of VMIDs being used
478 * The hardware supports a limited set of values with the value zero reserved
479 * for the host, so we check if an assigned value belongs to a previous
480 * generation, which which requires us to assign a new value. If we're the
481 * first to use a VMID for the new generation, we must flush necessary caches
482 * and TLBs on all CPUs.
484 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
486 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
487 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
488 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
492 * update_vmid - Update the vmid with a valid VMID for the current generation
493 * @kvm: The guest that struct vmid belongs to
494 * @vmid: The stage-2 VMID information struct
496 static void update_vmid(struct kvm_vmid *vmid)
498 if (!need_new_vmid_gen(vmid))
501 spin_lock(&kvm_vmid_lock);
504 * We need to re-check the vmid_gen here to ensure that if another vcpu
505 * already allocated a valid vmid for this vm, then this vcpu should
508 if (!need_new_vmid_gen(vmid)) {
509 spin_unlock(&kvm_vmid_lock);
513 /* First user of a new VMID generation? */
514 if (unlikely(kvm_next_vmid == 0)) {
515 atomic64_inc(&kvm_vmid_gen);
519 * On SMP we know no other CPUs can use this CPU's or each
520 * other's VMID after force_vm_exit returns since the
521 * kvm_vmid_lock blocks them from reentry to the guest.
523 force_vm_exit(cpu_all_mask);
525 * Now broadcast TLB + ICACHE invalidation over the inner
526 * shareable domain to make sure all data structures are
529 kvm_call_hyp(__kvm_flush_vm_context);
532 vmid->vmid = kvm_next_vmid;
534 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
537 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
539 spin_unlock(&kvm_vmid_lock);
542 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
544 struct kvm *kvm = vcpu->kvm;
547 if (likely(vcpu->arch.has_run_once))
550 if (!kvm_arm_vcpu_is_finalized(vcpu))
553 vcpu->arch.has_run_once = true;
555 if (likely(irqchip_in_kernel(kvm))) {
557 * Map the VGIC hardware resources before running a vcpu the
558 * first time on this VM.
560 if (unlikely(!vgic_ready(kvm))) {
561 ret = kvm_vgic_map_resources(kvm);
567 * Tell the rest of the code that there are userspace irqchip
570 static_branch_inc(&userspace_irqchip_in_use);
573 ret = kvm_timer_enable(vcpu);
577 ret = kvm_arm_pmu_v3_enable(vcpu);
582 bool kvm_arch_intc_initialized(struct kvm *kvm)
584 return vgic_initialized(kvm);
587 void kvm_arm_halt_guest(struct kvm *kvm)
590 struct kvm_vcpu *vcpu;
592 kvm_for_each_vcpu(i, vcpu, kvm)
593 vcpu->arch.pause = true;
594 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
597 void kvm_arm_resume_guest(struct kvm *kvm)
600 struct kvm_vcpu *vcpu;
602 kvm_for_each_vcpu(i, vcpu, kvm) {
603 vcpu->arch.pause = false;
604 swake_up_one(kvm_arch_vcpu_wq(vcpu));
608 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
610 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
612 swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
613 (!vcpu->arch.pause)));
615 if (vcpu->arch.power_off || vcpu->arch.pause) {
616 /* Awaken to handle a signal, request we sleep again later. */
617 kvm_make_request(KVM_REQ_SLEEP, vcpu);
621 * Make sure we will observe a potential reset request if we've
622 * observed a change to the power state. Pairs with the smp_wmb() in
623 * kvm_psci_vcpu_on().
628 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
630 return vcpu->arch.target >= 0;
633 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
635 if (kvm_request_pending(vcpu)) {
636 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
637 vcpu_req_sleep(vcpu);
639 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
640 kvm_reset_vcpu(vcpu);
643 * Clear IRQ_PENDING requests that were made to guarantee
644 * that a VCPU sees new virtual interrupts.
646 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
651 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
652 * @vcpu: The VCPU pointer
653 * @run: The kvm_run structure pointer used for userspace state exchange
655 * This function is called through the VCPU_RUN ioctl called from user space. It
656 * will execute VM code in a loop until the time slice for the process is used
657 * or some emulation is needed from user space in which case the function will
658 * return with return value 0 and with the kvm_run structure filled in with the
659 * required data for the requested emulation.
661 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
665 if (unlikely(!kvm_vcpu_initialized(vcpu)))
668 ret = kvm_vcpu_first_run_init(vcpu);
672 if (run->exit_reason == KVM_EXIT_MMIO) {
673 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
678 if (run->immediate_exit)
683 kvm_sigset_activate(vcpu);
686 run->exit_reason = KVM_EXIT_UNKNOWN;
689 * Check conditions before entering the guest
693 update_vmid(&vcpu->kvm->arch.vmid);
695 check_vcpu_requests(vcpu);
698 * Preparing the interrupts to be injected also
699 * involves poking the GIC, which must be done in a
700 * non-preemptible context.
704 kvm_pmu_flush_hwstate(vcpu);
708 kvm_vgic_flush_hwstate(vcpu);
711 * Exit if we have a signal pending so that we can deliver the
712 * signal to user space.
714 if (signal_pending(current)) {
716 run->exit_reason = KVM_EXIT_INTR;
720 * If we're using a userspace irqchip, then check if we need
721 * to tell a userspace irqchip about timer or PMU level
722 * changes and if so, exit to userspace (the actual level
723 * state gets updated in kvm_timer_update_run and
724 * kvm_pmu_update_run below).
726 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
727 if (kvm_timer_should_notify_user(vcpu) ||
728 kvm_pmu_should_notify_user(vcpu)) {
730 run->exit_reason = KVM_EXIT_INTR;
735 * Ensure we set mode to IN_GUEST_MODE after we disable
736 * interrupts and before the final VCPU requests check.
737 * See the comment in kvm_vcpu_exiting_guest_mode() and
738 * Documentation/virt/kvm/vcpu-requests.rst
740 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
742 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
743 kvm_request_pending(vcpu)) {
744 vcpu->mode = OUTSIDE_GUEST_MODE;
745 isb(); /* Ensure work in x_flush_hwstate is committed */
746 kvm_pmu_sync_hwstate(vcpu);
747 if (static_branch_unlikely(&userspace_irqchip_in_use))
748 kvm_timer_sync_hwstate(vcpu);
749 kvm_vgic_sync_hwstate(vcpu);
755 kvm_arm_setup_debug(vcpu);
757 /**************************************************************
760 trace_kvm_entry(*vcpu_pc(vcpu));
761 guest_enter_irqoff();
764 kvm_arm_vhe_guest_enter();
765 ret = kvm_vcpu_run_vhe(vcpu);
766 kvm_arm_vhe_guest_exit();
768 ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
771 vcpu->mode = OUTSIDE_GUEST_MODE;
775 *************************************************************/
777 kvm_arm_clear_debug(vcpu);
780 * We must sync the PMU state before the vgic state so
781 * that the vgic can properly sample the updated state of the
784 kvm_pmu_sync_hwstate(vcpu);
787 * Sync the vgic state before syncing the timer state because
788 * the timer code needs to know if the virtual timer
789 * interrupts are active.
791 kvm_vgic_sync_hwstate(vcpu);
794 * Sync the timer hardware state before enabling interrupts as
795 * we don't want vtimer interrupts to race with syncing the
796 * timer virtual interrupt state.
798 if (static_branch_unlikely(&userspace_irqchip_in_use))
799 kvm_timer_sync_hwstate(vcpu);
801 kvm_arch_vcpu_ctxsync_fp(vcpu);
804 * We may have taken a host interrupt in HYP mode (ie
805 * while executing the guest). This interrupt is still
806 * pending, as we haven't serviced it yet!
808 * We're now back in SVC mode, with interrupts
809 * disabled. Enabling the interrupts now will have
810 * the effect of taking the interrupt again, in SVC
816 * We do local_irq_enable() before calling guest_exit() so
817 * that if a timer interrupt hits while running the guest we
818 * account that tick as being spent in the guest. We enable
819 * preemption after calling guest_exit() so that if we get
820 * preempted we make sure ticks after that is not counted as
824 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
826 /* Exit types that need handling before we can be preempted */
827 handle_exit_early(vcpu, run, ret);
831 ret = handle_exit(vcpu, run, ret);
834 /* Tell userspace about in-kernel device output levels */
835 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
836 kvm_timer_update_run(vcpu);
837 kvm_pmu_update_run(vcpu);
840 kvm_sigset_deactivate(vcpu);
846 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
852 if (number == KVM_ARM_IRQ_CPU_IRQ)
853 bit_index = __ffs(HCR_VI);
854 else /* KVM_ARM_IRQ_CPU_FIQ */
855 bit_index = __ffs(HCR_VF);
857 hcr = vcpu_hcr(vcpu);
859 set = test_and_set_bit(bit_index, hcr);
861 set = test_and_clear_bit(bit_index, hcr);
864 * If we didn't change anything, no need to wake up or kick other CPUs
870 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
871 * trigger a world-switch round on the running physical CPU to set the
872 * virtual IRQ/FIQ fields in the HCR appropriately.
874 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
880 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
883 u32 irq = irq_level->irq;
884 unsigned int irq_type, vcpu_idx, irq_num;
885 int nrcpus = atomic_read(&kvm->online_vcpus);
886 struct kvm_vcpu *vcpu = NULL;
887 bool level = irq_level->level;
889 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
890 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
891 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
893 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
896 case KVM_ARM_IRQ_TYPE_CPU:
897 if (irqchip_in_kernel(kvm))
900 if (vcpu_idx >= nrcpus)
903 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
907 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
910 return vcpu_interrupt_line(vcpu, irq_num, level);
911 case KVM_ARM_IRQ_TYPE_PPI:
912 if (!irqchip_in_kernel(kvm))
915 if (vcpu_idx >= nrcpus)
918 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
922 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
925 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
926 case KVM_ARM_IRQ_TYPE_SPI:
927 if (!irqchip_in_kernel(kvm))
930 if (irq_num < VGIC_NR_PRIVATE_IRQS)
933 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
939 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
940 const struct kvm_vcpu_init *init)
943 int phys_target = kvm_target_cpu();
945 if (init->target != phys_target)
949 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
950 * use the same target.
952 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
955 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
956 for (i = 0; i < sizeof(init->features) * 8; i++) {
957 bool set = (init->features[i / 32] & (1 << (i % 32)));
959 if (set && i >= KVM_VCPU_MAX_FEATURES)
963 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
964 * use the same feature set.
966 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
967 test_bit(i, vcpu->arch.features) != set)
971 set_bit(i, vcpu->arch.features);
974 vcpu->arch.target = phys_target;
976 /* Now we know what it is, we can reset it. */
977 ret = kvm_reset_vcpu(vcpu);
979 vcpu->arch.target = -1;
980 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
986 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
987 struct kvm_vcpu_init *init)
991 ret = kvm_vcpu_set_target(vcpu, init);
996 * Ensure a rebooted VM will fault in RAM pages and detect if the
997 * guest MMU is turned off and flush the caches as needed.
999 if (vcpu->arch.has_run_once)
1000 stage2_unmap_vm(vcpu->kvm);
1002 vcpu_reset_hcr(vcpu);
1005 * Handle the "start in power-off" case.
1007 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1008 vcpu_power_off(vcpu);
1010 vcpu->arch.power_off = false;
1015 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1016 struct kvm_device_attr *attr)
1020 switch (attr->group) {
1022 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1029 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1030 struct kvm_device_attr *attr)
1034 switch (attr->group) {
1036 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1043 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1044 struct kvm_device_attr *attr)
1048 switch (attr->group) {
1050 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1057 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1058 struct kvm_vcpu_events *events)
1060 memset(events, 0, sizeof(*events));
1062 return __kvm_arm_vcpu_get_events(vcpu, events);
1065 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1066 struct kvm_vcpu_events *events)
1070 /* check whether the reserved field is zero */
1071 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1072 if (events->reserved[i])
1075 /* check whether the pad field is zero */
1076 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1077 if (events->exception.pad[i])
1080 return __kvm_arm_vcpu_set_events(vcpu, events);
1083 long kvm_arch_vcpu_ioctl(struct file *filp,
1084 unsigned int ioctl, unsigned long arg)
1086 struct kvm_vcpu *vcpu = filp->private_data;
1087 void __user *argp = (void __user *)arg;
1088 struct kvm_device_attr attr;
1092 case KVM_ARM_VCPU_INIT: {
1093 struct kvm_vcpu_init init;
1096 if (copy_from_user(&init, argp, sizeof(init)))
1099 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1102 case KVM_SET_ONE_REG:
1103 case KVM_GET_ONE_REG: {
1104 struct kvm_one_reg reg;
1107 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1111 if (copy_from_user(®, argp, sizeof(reg)))
1114 if (ioctl == KVM_SET_ONE_REG)
1115 r = kvm_arm_set_reg(vcpu, ®);
1117 r = kvm_arm_get_reg(vcpu, ®);
1120 case KVM_GET_REG_LIST: {
1121 struct kvm_reg_list __user *user_list = argp;
1122 struct kvm_reg_list reg_list;
1126 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1130 if (!kvm_arm_vcpu_is_finalized(vcpu))
1134 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1137 reg_list.n = kvm_arm_num_regs(vcpu);
1138 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1143 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1146 case KVM_SET_DEVICE_ATTR: {
1148 if (copy_from_user(&attr, argp, sizeof(attr)))
1150 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1153 case KVM_GET_DEVICE_ATTR: {
1155 if (copy_from_user(&attr, argp, sizeof(attr)))
1157 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1160 case KVM_HAS_DEVICE_ATTR: {
1162 if (copy_from_user(&attr, argp, sizeof(attr)))
1164 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1167 case KVM_GET_VCPU_EVENTS: {
1168 struct kvm_vcpu_events events;
1170 if (kvm_arm_vcpu_get_events(vcpu, &events))
1173 if (copy_to_user(argp, &events, sizeof(events)))
1178 case KVM_SET_VCPU_EVENTS: {
1179 struct kvm_vcpu_events events;
1181 if (copy_from_user(&events, argp, sizeof(events)))
1184 return kvm_arm_vcpu_set_events(vcpu, &events);
1186 case KVM_ARM_VCPU_FINALIZE: {
1189 if (!kvm_vcpu_initialized(vcpu))
1192 if (get_user(what, (const int __user *)argp))
1195 return kvm_arm_vcpu_finalize(vcpu, what);
1205 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1206 * @kvm: kvm instance
1207 * @log: slot id and address to which we copy the log
1209 * Steps 1-4 below provide general overview of dirty page logging. See
1210 * kvm_get_dirty_log_protect() function description for additional details.
1212 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1213 * always flush the TLB (step 4) even if previous step failed and the dirty
1214 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1215 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1216 * writes will be marked dirty for next log read.
1218 * 1. Take a snapshot of the bit and clear it if needed.
1219 * 2. Write protect the corresponding page.
1220 * 3. Copy the snapshot to the userspace.
1221 * 4. Flush TLB's if needed.
1223 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1228 mutex_lock(&kvm->slots_lock);
1230 r = kvm_get_dirty_log_protect(kvm, log, &flush);
1233 kvm_flush_remote_tlbs(kvm);
1235 mutex_unlock(&kvm->slots_lock);
1239 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1244 mutex_lock(&kvm->slots_lock);
1246 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1249 kvm_flush_remote_tlbs(kvm);
1251 mutex_unlock(&kvm->slots_lock);
1255 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1256 struct kvm_arm_device_addr *dev_addr)
1258 unsigned long dev_id, type;
1260 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1261 KVM_ARM_DEVICE_ID_SHIFT;
1262 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1263 KVM_ARM_DEVICE_TYPE_SHIFT;
1266 case KVM_ARM_DEVICE_VGIC_V2:
1269 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1275 long kvm_arch_vm_ioctl(struct file *filp,
1276 unsigned int ioctl, unsigned long arg)
1278 struct kvm *kvm = filp->private_data;
1279 void __user *argp = (void __user *)arg;
1282 case KVM_CREATE_IRQCHIP: {
1286 mutex_lock(&kvm->lock);
1287 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1288 mutex_unlock(&kvm->lock);
1291 case KVM_ARM_SET_DEVICE_ADDR: {
1292 struct kvm_arm_device_addr dev_addr;
1294 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1296 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1298 case KVM_ARM_PREFERRED_TARGET: {
1300 struct kvm_vcpu_init init;
1302 err = kvm_vcpu_preferred_target(&init);
1306 if (copy_to_user(argp, &init, sizeof(init)))
1316 static void cpu_init_hyp_mode(void *dummy)
1318 phys_addr_t pgd_ptr;
1319 unsigned long hyp_stack_ptr;
1320 unsigned long stack_page;
1321 unsigned long vector_ptr;
1323 /* Switch from the HYP stub to our own HYP init vector */
1324 __hyp_set_vectors(kvm_get_idmap_vector());
1326 pgd_ptr = kvm_mmu_get_httbr();
1327 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1328 hyp_stack_ptr = stack_page + PAGE_SIZE;
1329 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1331 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1332 __cpu_init_stage2();
1335 static void cpu_hyp_reset(void)
1337 if (!is_kernel_in_hyp_mode())
1338 __hyp_reset_vectors();
1341 static void cpu_hyp_reinit(void)
1343 kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1347 if (is_kernel_in_hyp_mode())
1348 kvm_timer_init_vhe();
1350 cpu_init_hyp_mode(NULL);
1352 kvm_arm_init_debug();
1355 kvm_vgic_init_cpu_hardware();
1358 static void _kvm_arch_hardware_enable(void *discard)
1360 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1362 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1366 int kvm_arch_hardware_enable(void)
1368 _kvm_arch_hardware_enable(NULL);
1372 static void _kvm_arch_hardware_disable(void *discard)
1374 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1376 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1380 void kvm_arch_hardware_disable(void)
1382 _kvm_arch_hardware_disable(NULL);
1385 #ifdef CONFIG_CPU_PM
1386 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1391 * kvm_arm_hardware_enabled is left with its old value over
1392 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1397 if (__this_cpu_read(kvm_arm_hardware_enabled))
1399 * don't update kvm_arm_hardware_enabled here
1400 * so that the hardware will be re-enabled
1401 * when we resume. See below.
1406 case CPU_PM_ENTER_FAILED:
1408 if (__this_cpu_read(kvm_arm_hardware_enabled))
1409 /* The hardware was enabled before suspend. */
1419 static struct notifier_block hyp_init_cpu_pm_nb = {
1420 .notifier_call = hyp_init_cpu_pm_notifier,
1423 static void __init hyp_cpu_pm_init(void)
1425 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1427 static void __init hyp_cpu_pm_exit(void)
1429 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1432 static inline void hyp_cpu_pm_init(void)
1435 static inline void hyp_cpu_pm_exit(void)
1440 static int init_common_resources(void)
1442 kvm_set_ipa_limit();
1447 static int init_subsystems(void)
1452 * Enable hardware so that subsystem initialisation can access EL2.
1454 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1457 * Register CPU lower-power notifier
1462 * Init HYP view of VGIC
1464 err = kvm_vgic_hyp_init();
1467 vgic_present = true;
1471 vgic_present = false;
1479 * Init HYP architected timer support
1481 err = kvm_timer_hyp_init(vgic_present);
1486 kvm_coproc_table_init();
1489 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1494 static void teardown_hyp_mode(void)
1499 for_each_possible_cpu(cpu)
1500 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1505 * Inits Hyp-mode on all online CPUs
1507 static int init_hyp_mode(void)
1513 * Allocate Hyp PGD and setup Hyp identity mapping
1515 err = kvm_mmu_init();
1520 * Allocate stack pages for Hypervisor-mode
1522 for_each_possible_cpu(cpu) {
1523 unsigned long stack_page;
1525 stack_page = __get_free_page(GFP_KERNEL);
1531 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1535 * Map the Hyp-code called directly from the host
1537 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1538 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1540 kvm_err("Cannot map world-switch code\n");
1544 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1545 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1547 kvm_err("Cannot map rodata section\n");
1551 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1552 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1554 kvm_err("Cannot map bss section\n");
1558 err = kvm_map_vectors();
1560 kvm_err("Cannot map vectors\n");
1565 * Map the Hyp stack pages
1567 for_each_possible_cpu(cpu) {
1568 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1569 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1573 kvm_err("Cannot map hyp stack\n");
1578 for_each_possible_cpu(cpu) {
1579 kvm_host_data_t *cpu_data;
1581 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1582 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1585 kvm_err("Cannot map host CPU state: %d\n", err);
1590 err = hyp_map_aux_data();
1592 kvm_err("Cannot map host auxiliary data: %d\n", err);
1597 teardown_hyp_mode();
1598 kvm_err("error initializing Hyp mode: %d\n", err);
1602 static void check_kvm_target_cpu(void *ret)
1604 *(int *)ret = kvm_target_cpu();
1607 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1609 struct kvm_vcpu *vcpu;
1612 mpidr &= MPIDR_HWID_BITMASK;
1613 kvm_for_each_vcpu(i, vcpu, kvm) {
1614 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1620 bool kvm_arch_has_irq_bypass(void)
1625 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1626 struct irq_bypass_producer *prod)
1628 struct kvm_kernel_irqfd *irqfd =
1629 container_of(cons, struct kvm_kernel_irqfd, consumer);
1631 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1634 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1635 struct irq_bypass_producer *prod)
1637 struct kvm_kernel_irqfd *irqfd =
1638 container_of(cons, struct kvm_kernel_irqfd, consumer);
1640 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1644 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1646 struct kvm_kernel_irqfd *irqfd =
1647 container_of(cons, struct kvm_kernel_irqfd, consumer);
1649 kvm_arm_halt_guest(irqfd->kvm);
1652 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1654 struct kvm_kernel_irqfd *irqfd =
1655 container_of(cons, struct kvm_kernel_irqfd, consumer);
1657 kvm_arm_resume_guest(irqfd->kvm);
1661 * Initialize Hyp-mode and memory mappings on all CPUs.
1663 int kvm_arch_init(void *opaque)
1669 if (!is_hyp_mode_available()) {
1670 kvm_info("HYP mode not available\n");
1674 in_hyp_mode = is_kernel_in_hyp_mode();
1676 if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1677 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1681 for_each_online_cpu(cpu) {
1682 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1684 kvm_err("Error, CPU %d not supported!\n", cpu);
1689 err = init_common_resources();
1693 err = kvm_arm_init_sve();
1698 err = init_hyp_mode();
1703 err = init_subsystems();
1708 kvm_info("VHE mode initialized successfully\n");
1710 kvm_info("Hyp mode initialized successfully\n");
1716 teardown_hyp_mode();
1721 /* NOP: Compiling as a module not supported */
1722 void kvm_arch_exit(void)
1724 kvm_perf_teardown();
1727 static int arm_init(void)
1729 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1733 module_init(arm_init);