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/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kvm.h>
20 #include <linux/kvm_irqfd.h>
21 #include <linux/irqbypass.h>
22 #include <linux/sched/stat.h>
23 #include <linux/psci.h>
24 #include <trace/events/kvm.h>
26 #define CREATE_TRACE_POINTS
27 #include "trace_arm.h"
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_pkvm.h>
40 #include <asm/kvm_emulate.h>
41 #include <asm/sections.h>
43 #include <kvm/arm_hypercalls.h>
44 #include <kvm/arm_pmu.h>
45 #include <kvm/arm_psci.h>
47 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
49 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
51 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
56 static bool vgic_present, kvm_arm_initialised;
58 static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized);
59 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
61 bool is_kvm_arm_initialised(void)
63 return kvm_arm_initialised;
66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
68 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
71 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
72 struct kvm_enable_cap *cap)
81 case KVM_CAP_ARM_NISV_TO_USER:
83 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
87 mutex_lock(&kvm->lock);
88 if (!system_supports_mte() || kvm->created_vcpus) {
92 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
94 mutex_unlock(&kvm->lock);
96 case KVM_CAP_ARM_SYSTEM_SUSPEND:
98 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
100 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
101 new_cap = cap->args[0];
103 mutex_lock(&kvm->slots_lock);
105 * To keep things simple, allow changing the chunk
106 * size only when no memory slots have been created.
108 if (!kvm_are_all_memslots_empty(kvm)) {
110 } else if (new_cap && !kvm_is_block_size_supported(new_cap)) {
114 kvm->arch.mmu.split_page_chunk_size = new_cap;
116 mutex_unlock(&kvm->slots_lock);
126 static int kvm_arm_default_max_vcpus(void)
128 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
132 * kvm_arch_init_vm - initializes a VM data structure
133 * @kvm: pointer to the KVM struct
135 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
139 mutex_init(&kvm->arch.config_lock);
141 #ifdef CONFIG_LOCKDEP
142 /* Clue in lockdep that the config_lock must be taken inside kvm->lock */
143 mutex_lock(&kvm->lock);
144 mutex_lock(&kvm->arch.config_lock);
145 mutex_unlock(&kvm->arch.config_lock);
146 mutex_unlock(&kvm->lock);
149 ret = kvm_share_hyp(kvm, kvm + 1);
153 ret = pkvm_init_host_vm(kvm);
155 goto err_unshare_kvm;
157 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
159 goto err_unshare_kvm;
161 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
163 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
165 goto err_free_cpumask;
167 kvm_vgic_early_init(kvm);
169 kvm_timer_init_vm(kvm);
171 /* The maximum number of VCPUs is limited by the host's GIC model */
172 kvm->max_vcpus = kvm_arm_default_max_vcpus();
174 kvm_arm_init_hypercalls(kvm);
176 bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
181 free_cpumask_var(kvm->arch.supported_cpus);
183 kvm_unshare_hyp(kvm, kvm + 1);
187 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
189 return VM_FAULT_SIGBUS;
194 * kvm_arch_destroy_vm - destroy the VM data structure
195 * @kvm: pointer to the KVM struct
197 void kvm_arch_destroy_vm(struct kvm *kvm)
199 bitmap_free(kvm->arch.pmu_filter);
200 free_cpumask_var(kvm->arch.supported_cpus);
202 kvm_vgic_destroy(kvm);
204 if (is_protected_kvm_enabled())
205 pkvm_destroy_hyp_vm(kvm);
207 kvm_destroy_vcpus(kvm);
209 kvm_unshare_hyp(kvm, kvm + 1);
211 kvm_arm_teardown_hypercalls(kvm);
214 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
218 case KVM_CAP_IRQCHIP:
221 case KVM_CAP_IOEVENTFD:
222 case KVM_CAP_DEVICE_CTRL:
223 case KVM_CAP_USER_MEMORY:
224 case KVM_CAP_SYNC_MMU:
225 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
226 case KVM_CAP_ONE_REG:
227 case KVM_CAP_ARM_PSCI:
228 case KVM_CAP_ARM_PSCI_0_2:
229 case KVM_CAP_READONLY_MEM:
230 case KVM_CAP_MP_STATE:
231 case KVM_CAP_IMMEDIATE_EXIT:
232 case KVM_CAP_VCPU_EVENTS:
233 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
234 case KVM_CAP_ARM_NISV_TO_USER:
235 case KVM_CAP_ARM_INJECT_EXT_DABT:
236 case KVM_CAP_SET_GUEST_DEBUG:
237 case KVM_CAP_VCPU_ATTRIBUTES:
238 case KVM_CAP_PTP_KVM:
239 case KVM_CAP_ARM_SYSTEM_SUSPEND:
240 case KVM_CAP_IRQFD_RESAMPLE:
241 case KVM_CAP_COUNTER_OFFSET:
244 case KVM_CAP_SET_GUEST_DEBUG2:
245 return KVM_GUESTDBG_VALID_MASK;
246 case KVM_CAP_ARM_SET_DEVICE_ADDR:
249 case KVM_CAP_NR_VCPUS:
251 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
252 * architectures, as it does not always bound it to
253 * KVM_CAP_MAX_VCPUS. It should not matter much because
254 * this is just an advisory value.
256 r = min_t(unsigned int, num_online_cpus(),
257 kvm_arm_default_max_vcpus());
259 case KVM_CAP_MAX_VCPUS:
260 case KVM_CAP_MAX_VCPU_ID:
264 r = kvm_arm_default_max_vcpus();
266 case KVM_CAP_MSI_DEVID:
270 r = kvm->arch.vgic.msis_require_devid;
272 case KVM_CAP_ARM_USER_IRQ:
274 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
275 * (bump this number if adding more devices)
279 case KVM_CAP_ARM_MTE:
280 r = system_supports_mte();
282 case KVM_CAP_STEAL_TIME:
283 r = kvm_arm_pvtime_supported();
285 case KVM_CAP_ARM_EL1_32BIT:
286 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
288 case KVM_CAP_GUEST_DEBUG_HW_BPS:
291 case KVM_CAP_GUEST_DEBUG_HW_WPS:
294 case KVM_CAP_ARM_PMU_V3:
295 r = kvm_arm_support_pmu_v3();
297 case KVM_CAP_ARM_INJECT_SERROR_ESR:
298 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
300 case KVM_CAP_ARM_VM_IPA_SIZE:
301 r = get_kvm_ipa_limit();
303 case KVM_CAP_ARM_SVE:
304 r = system_supports_sve();
306 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
307 case KVM_CAP_ARM_PTRAUTH_GENERIC:
308 r = system_has_full_ptr_auth();
310 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
312 r = kvm->arch.mmu.split_page_chunk_size;
314 r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
316 case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
317 r = kvm_supported_block_sizes();
326 long kvm_arch_dev_ioctl(struct file *filp,
327 unsigned int ioctl, unsigned long arg)
332 struct kvm *kvm_arch_alloc_vm(void)
334 size_t sz = sizeof(struct kvm);
337 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
339 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
342 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
344 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
347 if (id >= kvm->max_vcpus)
353 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
357 spin_lock_init(&vcpu->arch.mp_state_lock);
359 #ifdef CONFIG_LOCKDEP
360 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
361 mutex_lock(&vcpu->mutex);
362 mutex_lock(&vcpu->kvm->arch.config_lock);
363 mutex_unlock(&vcpu->kvm->arch.config_lock);
364 mutex_unlock(&vcpu->mutex);
367 /* Force users to call KVM_ARM_VCPU_INIT */
368 vcpu->arch.target = -1;
369 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
371 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
374 * Default value for the FP state, will be overloaded at load
375 * time if we support FP (pretty likely)
377 vcpu->arch.fp_state = FP_STATE_FREE;
379 /* Set up the timer */
380 kvm_timer_vcpu_init(vcpu);
382 kvm_pmu_vcpu_init(vcpu);
384 kvm_arm_reset_debug_ptr(vcpu);
386 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
388 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
390 err = kvm_vgic_vcpu_init(vcpu);
394 return kvm_share_hyp(vcpu, vcpu + 1);
397 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
401 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
403 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
404 static_branch_dec(&userspace_irqchip_in_use);
406 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
407 kvm_timer_vcpu_terminate(vcpu);
408 kvm_pmu_vcpu_destroy(vcpu);
410 kvm_arm_vcpu_destroy(vcpu);
413 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
418 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
423 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
425 struct kvm_s2_mmu *mmu;
428 mmu = vcpu->arch.hw_mmu;
429 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
432 * We guarantee that both TLBs and I-cache are private to each
433 * vcpu. If detecting that a vcpu from the same VM has
434 * previously run on the same physical CPU, call into the
435 * hypervisor code to nuke the relevant contexts.
437 * We might get preempted before the vCPU actually runs, but
438 * over-invalidation doesn't affect correctness.
440 if (*last_ran != vcpu->vcpu_id) {
441 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
442 *last_ran = vcpu->vcpu_id;
448 kvm_timer_vcpu_load(vcpu);
450 kvm_vcpu_load_sysregs_vhe(vcpu);
451 kvm_arch_vcpu_load_fp(vcpu);
452 kvm_vcpu_pmu_restore_guest(vcpu);
453 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
454 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
456 if (single_task_running())
457 vcpu_clear_wfx_traps(vcpu);
459 vcpu_set_wfx_traps(vcpu);
461 if (vcpu_has_ptrauth(vcpu))
462 vcpu_ptrauth_disable(vcpu);
463 kvm_arch_vcpu_load_debug_state_flags(vcpu);
465 if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
466 vcpu_set_on_unsupported_cpu(vcpu);
469 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
471 kvm_arch_vcpu_put_debug_state_flags(vcpu);
472 kvm_arch_vcpu_put_fp(vcpu);
474 kvm_vcpu_put_sysregs_vhe(vcpu);
475 kvm_timer_vcpu_put(vcpu);
477 kvm_vcpu_pmu_restore_host(vcpu);
478 kvm_arm_vmid_clear_active();
480 vcpu_clear_on_unsupported_cpu(vcpu);
484 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
486 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
487 kvm_make_request(KVM_REQ_SLEEP, vcpu);
491 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
493 spin_lock(&vcpu->arch.mp_state_lock);
494 __kvm_arm_vcpu_power_off(vcpu);
495 spin_unlock(&vcpu->arch.mp_state_lock);
498 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
500 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
503 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
505 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
506 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
510 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
512 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
515 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
516 struct kvm_mp_state *mp_state)
518 *mp_state = READ_ONCE(vcpu->arch.mp_state);
523 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
524 struct kvm_mp_state *mp_state)
528 spin_lock(&vcpu->arch.mp_state_lock);
530 switch (mp_state->mp_state) {
531 case KVM_MP_STATE_RUNNABLE:
532 WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
534 case KVM_MP_STATE_STOPPED:
535 __kvm_arm_vcpu_power_off(vcpu);
537 case KVM_MP_STATE_SUSPENDED:
538 kvm_arm_vcpu_suspend(vcpu);
544 spin_unlock(&vcpu->arch.mp_state_lock);
550 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
551 * @v: The VCPU pointer
553 * If the guest CPU is not waiting for interrupts or an interrupt line is
554 * asserted, the CPU is by definition runnable.
556 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
558 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
559 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
560 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
563 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
565 return vcpu_mode_priv(vcpu);
568 #ifdef CONFIG_GUEST_PERF_EVENTS
569 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
571 return *vcpu_pc(vcpu);
575 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
577 return vcpu->arch.target >= 0;
581 * Handle both the initialisation that is being done when the vcpu is
582 * run for the first time, as well as the updates that must be
583 * performed each time we get a new thread dealing with this vcpu.
585 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
587 struct kvm *kvm = vcpu->kvm;
590 if (!kvm_vcpu_initialized(vcpu))
593 if (!kvm_arm_vcpu_is_finalized(vcpu))
596 ret = kvm_arch_vcpu_run_map_fp(vcpu);
600 if (likely(vcpu_has_run_once(vcpu)))
603 kvm_arm_vcpu_init_debug(vcpu);
605 if (likely(irqchip_in_kernel(kvm))) {
607 * Map the VGIC hardware resources before running a vcpu the
608 * first time on this VM.
610 ret = kvm_vgic_map_resources(kvm);
615 ret = kvm_timer_enable(vcpu);
619 ret = kvm_arm_pmu_v3_enable(vcpu);
623 if (is_protected_kvm_enabled()) {
624 ret = pkvm_create_hyp_vm(kvm);
629 if (!irqchip_in_kernel(kvm)) {
631 * Tell the rest of the code that there are userspace irqchip
634 static_branch_inc(&userspace_irqchip_in_use);
638 * Initialize traps for protected VMs.
639 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
640 * the code is in place for first run initialization at EL2.
642 if (kvm_vm_is_protected(kvm))
643 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
645 mutex_lock(&kvm->arch.config_lock);
646 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
647 mutex_unlock(&kvm->arch.config_lock);
652 bool kvm_arch_intc_initialized(struct kvm *kvm)
654 return vgic_initialized(kvm);
657 void kvm_arm_halt_guest(struct kvm *kvm)
660 struct kvm_vcpu *vcpu;
662 kvm_for_each_vcpu(i, vcpu, kvm)
663 vcpu->arch.pause = true;
664 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
667 void kvm_arm_resume_guest(struct kvm *kvm)
670 struct kvm_vcpu *vcpu;
672 kvm_for_each_vcpu(i, vcpu, kvm) {
673 vcpu->arch.pause = false;
674 __kvm_vcpu_wake_up(vcpu);
678 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
680 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
682 rcuwait_wait_event(wait,
683 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
686 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
687 /* Awaken to handle a signal, request we sleep again later. */
688 kvm_make_request(KVM_REQ_SLEEP, vcpu);
692 * Make sure we will observe a potential reset request if we've
693 * observed a change to the power state. Pairs with the smp_wmb() in
694 * kvm_psci_vcpu_on().
700 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
701 * @vcpu: The VCPU pointer
703 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
704 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
705 * on when a wake event arrives, e.g. there may already be a pending wake event.
707 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
710 * Sync back the state of the GIC CPU interface so that we have
711 * the latest PMR and group enables. This ensures that
712 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
713 * we have pending interrupts, e.g. when determining if the
716 * For the same reason, we want to tell GICv4 that we need
717 * doorbells to be signalled, should an interrupt become pending.
720 kvm_vgic_vmcr_sync(vcpu);
721 vcpu_set_flag(vcpu, IN_WFI);
726 vcpu_clear_flag(vcpu, IN_WFIT);
729 vcpu_clear_flag(vcpu, IN_WFI);
734 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
736 if (!kvm_arm_vcpu_suspended(vcpu))
742 * The suspend state is sticky; we do not leave it until userspace
743 * explicitly marks the vCPU as runnable. Request that we suspend again
746 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
749 * Check to make sure the vCPU is actually runnable. If so, exit to
750 * userspace informing it of the wakeup condition.
752 if (kvm_arch_vcpu_runnable(vcpu)) {
753 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
754 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
755 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
760 * Otherwise, we were unblocked to process a different event, such as a
761 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
768 * check_vcpu_requests - check and handle pending vCPU requests
769 * @vcpu: the VCPU pointer
771 * Return: 1 if we should enter the guest
772 * 0 if we should exit to userspace
773 * < 0 if we should exit to userspace, where the return value indicates
776 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
778 if (kvm_request_pending(vcpu)) {
779 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
780 kvm_vcpu_sleep(vcpu);
782 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
783 kvm_reset_vcpu(vcpu);
786 * Clear IRQ_PENDING requests that were made to guarantee
787 * that a VCPU sees new virtual interrupts.
789 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
791 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
792 kvm_update_stolen_time(vcpu);
794 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
795 /* The distributor enable bits were changed */
802 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
803 kvm_pmu_handle_pmcr(vcpu,
804 __vcpu_sys_reg(vcpu, PMCR_EL0));
806 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
807 return kvm_vcpu_suspend(vcpu);
809 if (kvm_dirty_ring_check_request(vcpu))
816 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
818 if (likely(!vcpu_mode_is_32bit(vcpu)))
821 return !kvm_supports_32bit_el0();
825 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
826 * @vcpu: The VCPU pointer
827 * @ret: Pointer to write optional return code
829 * Returns: true if the VCPU needs to return to a preemptible + interruptible
830 * and skip guest entry.
832 * This function disambiguates between two different types of exits: exits to a
833 * preemptible + interruptible kernel context and exits to userspace. For an
834 * exit to userspace, this function will write the return code to ret and return
835 * true. For an exit to preemptible + interruptible kernel context (i.e. check
836 * for pending work and re-enter), return true without writing to ret.
838 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
840 struct kvm_run *run = vcpu->run;
843 * If we're using a userspace irqchip, then check if we need
844 * to tell a userspace irqchip about timer or PMU level
845 * changes and if so, exit to userspace (the actual level
846 * state gets updated in kvm_timer_update_run and
847 * kvm_pmu_update_run below).
849 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
850 if (kvm_timer_should_notify_user(vcpu) ||
851 kvm_pmu_should_notify_user(vcpu)) {
853 run->exit_reason = KVM_EXIT_INTR;
858 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
859 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
860 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
861 run->fail_entry.cpu = smp_processor_id();
866 return kvm_request_pending(vcpu) ||
867 xfer_to_guest_mode_work_pending();
871 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
872 * the vCPU is running.
874 * This must be noinstr as instrumentation may make use of RCU, and this is not
875 * safe during the EQS.
877 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
881 guest_state_enter_irqoff();
882 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
883 guest_state_exit_irqoff();
889 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
890 * @vcpu: The VCPU pointer
892 * This function is called through the VCPU_RUN ioctl called from user space. It
893 * will execute VM code in a loop until the time slice for the process is used
894 * or some emulation is needed from user space in which case the function will
895 * return with return value 0 and with the kvm_run structure filled in with the
896 * required data for the requested emulation.
898 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
900 struct kvm_run *run = vcpu->run;
903 if (run->exit_reason == KVM_EXIT_MMIO) {
904 ret = kvm_handle_mmio_return(vcpu);
911 if (run->immediate_exit) {
916 kvm_sigset_activate(vcpu);
919 run->exit_reason = KVM_EXIT_UNKNOWN;
923 * Check conditions before entering the guest
925 ret = xfer_to_guest_mode_handle_work(vcpu);
930 ret = check_vcpu_requests(vcpu);
933 * Preparing the interrupts to be injected also
934 * involves poking the GIC, which must be done in a
935 * non-preemptible context.
940 * The VMID allocator only tracks active VMIDs per
941 * physical CPU, and therefore the VMID allocated may not be
942 * preserved on VMID roll-over if the task was preempted,
943 * making a thread's VMID inactive. So we need to call
944 * kvm_arm_vmid_update() in non-premptible context.
946 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
948 kvm_pmu_flush_hwstate(vcpu);
952 kvm_vgic_flush_hwstate(vcpu);
954 kvm_pmu_update_vcpu_events(vcpu);
957 * Ensure we set mode to IN_GUEST_MODE after we disable
958 * interrupts and before the final VCPU requests check.
959 * See the comment in kvm_vcpu_exiting_guest_mode() and
960 * Documentation/virt/kvm/vcpu-requests.rst
962 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
964 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
965 vcpu->mode = OUTSIDE_GUEST_MODE;
966 isb(); /* Ensure work in x_flush_hwstate is committed */
967 kvm_pmu_sync_hwstate(vcpu);
968 if (static_branch_unlikely(&userspace_irqchip_in_use))
969 kvm_timer_sync_user(vcpu);
970 kvm_vgic_sync_hwstate(vcpu);
976 kvm_arm_setup_debug(vcpu);
977 kvm_arch_vcpu_ctxflush_fp(vcpu);
979 /**************************************************************
982 trace_kvm_entry(*vcpu_pc(vcpu));
983 guest_timing_enter_irqoff();
985 ret = kvm_arm_vcpu_enter_exit(vcpu);
987 vcpu->mode = OUTSIDE_GUEST_MODE;
991 *************************************************************/
993 kvm_arm_clear_debug(vcpu);
996 * We must sync the PMU state before the vgic state so
997 * that the vgic can properly sample the updated state of the
1000 kvm_pmu_sync_hwstate(vcpu);
1003 * Sync the vgic state before syncing the timer state because
1004 * the timer code needs to know if the virtual timer
1005 * interrupts are active.
1007 kvm_vgic_sync_hwstate(vcpu);
1010 * Sync the timer hardware state before enabling interrupts as
1011 * we don't want vtimer interrupts to race with syncing the
1012 * timer virtual interrupt state.
1014 if (static_branch_unlikely(&userspace_irqchip_in_use))
1015 kvm_timer_sync_user(vcpu);
1017 kvm_arch_vcpu_ctxsync_fp(vcpu);
1020 * We must ensure that any pending interrupts are taken before
1021 * we exit guest timing so that timer ticks are accounted as
1022 * guest time. Transiently unmask interrupts so that any
1023 * pending interrupts are taken.
1025 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1026 * context synchronization event) is necessary to ensure that
1027 * pending interrupts are taken.
1029 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1032 local_irq_disable();
1035 guest_timing_exit_irqoff();
1039 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1041 /* Exit types that need handling before we can be preempted */
1042 handle_exit_early(vcpu, ret);
1047 * The ARMv8 architecture doesn't give the hypervisor
1048 * a mechanism to prevent a guest from dropping to AArch32 EL0
1049 * if implemented by the CPU. If we spot the guest in such
1050 * state and that we decided it wasn't supposed to do so (like
1051 * with the asymmetric AArch32 case), return to userspace with
1054 if (vcpu_mode_is_bad_32bit(vcpu)) {
1056 * As we have caught the guest red-handed, decide that
1057 * it isn't fit for purpose anymore by making the vcpu
1058 * invalid. The VMM can try and fix it by issuing a
1059 * KVM_ARM_VCPU_INIT if it really wants to.
1061 vcpu->arch.target = -1;
1062 ret = ARM_EXCEPTION_IL;
1065 ret = handle_exit(vcpu, ret);
1068 /* Tell userspace about in-kernel device output levels */
1069 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1070 kvm_timer_update_run(vcpu);
1071 kvm_pmu_update_run(vcpu);
1074 kvm_sigset_deactivate(vcpu);
1078 * In the unlikely event that we are returning to userspace
1079 * with pending exceptions or PC adjustment, commit these
1080 * adjustments in order to give userspace a consistent view of
1081 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1082 * being preempt-safe on VHE.
1084 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1085 vcpu_get_flag(vcpu, INCREMENT_PC)))
1086 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1092 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1098 if (number == KVM_ARM_IRQ_CPU_IRQ)
1099 bit_index = __ffs(HCR_VI);
1100 else /* KVM_ARM_IRQ_CPU_FIQ */
1101 bit_index = __ffs(HCR_VF);
1103 hcr = vcpu_hcr(vcpu);
1105 set = test_and_set_bit(bit_index, hcr);
1107 set = test_and_clear_bit(bit_index, hcr);
1110 * If we didn't change anything, no need to wake up or kick other CPUs
1116 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1117 * trigger a world-switch round on the running physical CPU to set the
1118 * virtual IRQ/FIQ fields in the HCR appropriately.
1120 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1121 kvm_vcpu_kick(vcpu);
1126 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1129 u32 irq = irq_level->irq;
1130 unsigned int irq_type, vcpu_idx, irq_num;
1131 int nrcpus = atomic_read(&kvm->online_vcpus);
1132 struct kvm_vcpu *vcpu = NULL;
1133 bool level = irq_level->level;
1135 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1136 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1137 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1138 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1140 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1143 case KVM_ARM_IRQ_TYPE_CPU:
1144 if (irqchip_in_kernel(kvm))
1147 if (vcpu_idx >= nrcpus)
1150 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1154 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1157 return vcpu_interrupt_line(vcpu, irq_num, level);
1158 case KVM_ARM_IRQ_TYPE_PPI:
1159 if (!irqchip_in_kernel(kvm))
1162 if (vcpu_idx >= nrcpus)
1165 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1169 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1172 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1173 case KVM_ARM_IRQ_TYPE_SPI:
1174 if (!irqchip_in_kernel(kvm))
1177 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1180 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1186 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1187 const struct kvm_vcpu_init *init)
1189 unsigned long features = init->features[0];
1192 if (features & ~KVM_VCPU_VALID_FEATURES)
1195 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1196 if (init->features[i])
1200 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1203 if (!cpus_have_const_cap(ARM64_HAS_32BIT_EL1))
1206 /* MTE is incompatible with AArch32 */
1207 if (kvm_has_mte(vcpu->kvm))
1210 /* NV is incompatible with AArch32 */
1211 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1217 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1218 const struct kvm_vcpu_init *init)
1220 unsigned long features = init->features[0];
1222 return !bitmap_equal(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES) ||
1223 vcpu->arch.target != init->target;
1226 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1227 const struct kvm_vcpu_init *init)
1229 unsigned long features = init->features[0];
1230 struct kvm *kvm = vcpu->kvm;
1233 mutex_lock(&kvm->arch.config_lock);
1235 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1236 !bitmap_equal(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES))
1239 vcpu->arch.target = init->target;
1240 bitmap_copy(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES);
1242 /* Now we know what it is, we can reset it. */
1243 ret = kvm_reset_vcpu(vcpu);
1245 vcpu->arch.target = -1;
1246 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1250 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1251 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1254 mutex_unlock(&kvm->arch.config_lock);
1258 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1259 const struct kvm_vcpu_init *init)
1263 if (init->target != kvm_target_cpu())
1266 ret = kvm_vcpu_init_check_features(vcpu, init);
1270 if (vcpu->arch.target == -1)
1271 return __kvm_vcpu_set_target(vcpu, init);
1273 if (kvm_vcpu_init_changed(vcpu, init))
1276 return kvm_reset_vcpu(vcpu);
1279 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1280 struct kvm_vcpu_init *init)
1282 bool power_off = false;
1286 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1287 * reflecting it in the finalized feature set, thus limiting its scope
1288 * to a single KVM_ARM_VCPU_INIT call.
1290 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1291 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1295 ret = kvm_vcpu_set_target(vcpu, init);
1300 * Ensure a rebooted VM will fault in RAM pages and detect if the
1301 * guest MMU is turned off and flush the caches as needed.
1303 * S2FWB enforces all memory accesses to RAM being cacheable,
1304 * ensuring that the data side is always coherent. We still
1305 * need to invalidate the I-cache though, as FWB does *not*
1306 * imply CTR_EL0.DIC.
1308 if (vcpu_has_run_once(vcpu)) {
1309 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1310 stage2_unmap_vm(vcpu->kvm);
1312 icache_inval_all_pou();
1315 vcpu_reset_hcr(vcpu);
1316 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1319 * Handle the "start in power-off" case.
1321 spin_lock(&vcpu->arch.mp_state_lock);
1324 __kvm_arm_vcpu_power_off(vcpu);
1326 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1328 spin_unlock(&vcpu->arch.mp_state_lock);
1333 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1334 struct kvm_device_attr *attr)
1338 switch (attr->group) {
1340 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1347 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1348 struct kvm_device_attr *attr)
1352 switch (attr->group) {
1354 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1361 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1362 struct kvm_device_attr *attr)
1366 switch (attr->group) {
1368 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1375 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1376 struct kvm_vcpu_events *events)
1378 memset(events, 0, sizeof(*events));
1380 return __kvm_arm_vcpu_get_events(vcpu, events);
1383 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1384 struct kvm_vcpu_events *events)
1388 /* check whether the reserved field is zero */
1389 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1390 if (events->reserved[i])
1393 /* check whether the pad field is zero */
1394 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1395 if (events->exception.pad[i])
1398 return __kvm_arm_vcpu_set_events(vcpu, events);
1401 long kvm_arch_vcpu_ioctl(struct file *filp,
1402 unsigned int ioctl, unsigned long arg)
1404 struct kvm_vcpu *vcpu = filp->private_data;
1405 void __user *argp = (void __user *)arg;
1406 struct kvm_device_attr attr;
1410 case KVM_ARM_VCPU_INIT: {
1411 struct kvm_vcpu_init init;
1414 if (copy_from_user(&init, argp, sizeof(init)))
1417 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1420 case KVM_SET_ONE_REG:
1421 case KVM_GET_ONE_REG: {
1422 struct kvm_one_reg reg;
1425 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1429 if (copy_from_user(®, argp, sizeof(reg)))
1433 * We could owe a reset due to PSCI. Handle the pending reset
1434 * here to ensure userspace register accesses are ordered after
1437 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1438 kvm_reset_vcpu(vcpu);
1440 if (ioctl == KVM_SET_ONE_REG)
1441 r = kvm_arm_set_reg(vcpu, ®);
1443 r = kvm_arm_get_reg(vcpu, ®);
1446 case KVM_GET_REG_LIST: {
1447 struct kvm_reg_list __user *user_list = argp;
1448 struct kvm_reg_list reg_list;
1452 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1456 if (!kvm_arm_vcpu_is_finalized(vcpu))
1460 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1463 reg_list.n = kvm_arm_num_regs(vcpu);
1464 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1469 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1472 case KVM_SET_DEVICE_ATTR: {
1474 if (copy_from_user(&attr, argp, sizeof(attr)))
1476 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1479 case KVM_GET_DEVICE_ATTR: {
1481 if (copy_from_user(&attr, argp, sizeof(attr)))
1483 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1486 case KVM_HAS_DEVICE_ATTR: {
1488 if (copy_from_user(&attr, argp, sizeof(attr)))
1490 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1493 case KVM_GET_VCPU_EVENTS: {
1494 struct kvm_vcpu_events events;
1496 if (kvm_arm_vcpu_get_events(vcpu, &events))
1499 if (copy_to_user(argp, &events, sizeof(events)))
1504 case KVM_SET_VCPU_EVENTS: {
1505 struct kvm_vcpu_events events;
1507 if (copy_from_user(&events, argp, sizeof(events)))
1510 return kvm_arm_vcpu_set_events(vcpu, &events);
1512 case KVM_ARM_VCPU_FINALIZE: {
1515 if (!kvm_vcpu_initialized(vcpu))
1518 if (get_user(what, (const int __user *)argp))
1521 return kvm_arm_vcpu_finalize(vcpu, what);
1530 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1535 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1536 const struct kvm_memory_slot *memslot)
1538 kvm_flush_remote_tlbs(kvm);
1541 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1542 struct kvm_arm_device_addr *dev_addr)
1544 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1545 case KVM_ARM_DEVICE_VGIC_V2:
1548 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1554 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1556 switch (attr->group) {
1557 case KVM_ARM_VM_SMCCC_CTRL:
1558 return kvm_vm_smccc_has_attr(kvm, attr);
1564 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1566 switch (attr->group) {
1567 case KVM_ARM_VM_SMCCC_CTRL:
1568 return kvm_vm_smccc_set_attr(kvm, attr);
1574 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1576 struct kvm *kvm = filp->private_data;
1577 void __user *argp = (void __user *)arg;
1578 struct kvm_device_attr attr;
1581 case KVM_CREATE_IRQCHIP: {
1585 mutex_lock(&kvm->lock);
1586 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1587 mutex_unlock(&kvm->lock);
1590 case KVM_ARM_SET_DEVICE_ADDR: {
1591 struct kvm_arm_device_addr dev_addr;
1593 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1595 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1597 case KVM_ARM_PREFERRED_TARGET: {
1598 struct kvm_vcpu_init init;
1600 kvm_vcpu_preferred_target(&init);
1602 if (copy_to_user(argp, &init, sizeof(init)))
1607 case KVM_ARM_MTE_COPY_TAGS: {
1608 struct kvm_arm_copy_mte_tags copy_tags;
1610 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1612 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1614 case KVM_ARM_SET_COUNTER_OFFSET: {
1615 struct kvm_arm_counter_offset offset;
1617 if (copy_from_user(&offset, argp, sizeof(offset)))
1619 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1621 case KVM_HAS_DEVICE_ATTR: {
1622 if (copy_from_user(&attr, argp, sizeof(attr)))
1625 return kvm_vm_has_attr(kvm, &attr);
1627 case KVM_SET_DEVICE_ATTR: {
1628 if (copy_from_user(&attr, argp, sizeof(attr)))
1631 return kvm_vm_set_attr(kvm, &attr);
1638 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1639 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1641 struct kvm_vcpu *tmp_vcpu;
1643 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1644 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1645 mutex_unlock(&tmp_vcpu->mutex);
1649 void unlock_all_vcpus(struct kvm *kvm)
1651 lockdep_assert_held(&kvm->lock);
1653 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1656 /* Returns true if all vcpus were locked, false otherwise */
1657 bool lock_all_vcpus(struct kvm *kvm)
1659 struct kvm_vcpu *tmp_vcpu;
1662 lockdep_assert_held(&kvm->lock);
1665 * Any time a vcpu is in an ioctl (including running), the
1666 * core KVM code tries to grab the vcpu->mutex.
1668 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1669 * other VCPUs can fiddle with the state while we access it.
1671 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1672 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1673 unlock_vcpus(kvm, c - 1);
1681 static unsigned long nvhe_percpu_size(void)
1683 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1684 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1687 static unsigned long nvhe_percpu_order(void)
1689 unsigned long size = nvhe_percpu_size();
1691 return size ? get_order(size) : 0;
1694 /* A lookup table holding the hypervisor VA for each vector slot */
1695 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1697 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1699 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1702 static int kvm_init_vector_slots(void)
1707 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1708 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1710 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1711 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1713 if (kvm_system_needs_idmapped_vectors() &&
1714 !is_protected_kvm_enabled()) {
1715 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1716 __BP_HARDEN_HYP_VECS_SZ, &base);
1721 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1722 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1726 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1728 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1732 * Calculate the raw per-cpu offset without a translation from the
1733 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1734 * so that we can use adr_l to access per-cpu variables in EL2.
1735 * Also drop the KASAN tag which gets in the way...
1737 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1738 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1740 params->mair_el2 = read_sysreg(mair_el1);
1742 tcr = read_sysreg(tcr_el1);
1743 if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1744 tcr |= TCR_EPD1_MASK;
1746 tcr &= TCR_EL2_MASK;
1747 tcr |= TCR_EL2_RES1;
1749 tcr &= ~TCR_T0SZ_MASK;
1750 tcr |= TCR_T0SZ(hyp_va_bits);
1751 params->tcr_el2 = tcr;
1753 params->pgd_pa = kvm_mmu_get_httbr();
1754 if (is_protected_kvm_enabled())
1755 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1757 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1758 if (cpus_have_final_cap(ARM64_KVM_HVHE))
1759 params->hcr_el2 |= HCR_E2H;
1760 params->vttbr = params->vtcr = 0;
1763 * Flush the init params from the data cache because the struct will
1764 * be read while the MMU is off.
1766 kvm_flush_dcache_to_poc(params, sizeof(*params));
1769 static void hyp_install_host_vector(void)
1771 struct kvm_nvhe_init_params *params;
1772 struct arm_smccc_res res;
1774 /* Switch from the HYP stub to our own HYP init vector */
1775 __hyp_set_vectors(kvm_get_idmap_vector());
1778 * Call initialization code, and switch to the full blown HYP code.
1779 * If the cpucaps haven't been finalized yet, something has gone very
1780 * wrong, and hyp will crash and burn when it uses any
1781 * cpus_have_const_cap() wrapper.
1783 BUG_ON(!system_capabilities_finalized());
1784 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1785 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1786 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1789 static void cpu_init_hyp_mode(void)
1791 hyp_install_host_vector();
1794 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1797 if (this_cpu_has_cap(ARM64_SSBS) &&
1798 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1799 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1803 static void cpu_hyp_reset(void)
1805 if (!is_kernel_in_hyp_mode())
1806 __hyp_reset_vectors();
1810 * EL2 vectors can be mapped and rerouted in a number of ways,
1811 * depending on the kernel configuration and CPU present:
1813 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1814 * placed in one of the vector slots, which is executed before jumping
1815 * to the real vectors.
1817 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1818 * containing the hardening sequence is mapped next to the idmap page,
1819 * and executed before jumping to the real vectors.
1821 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1822 * empty slot is selected, mapped next to the idmap page, and
1823 * executed before jumping to the real vectors.
1825 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1826 * VHE, as we don't have hypervisor-specific mappings. If the system
1827 * is VHE and yet selects this capability, it will be ignored.
1829 static void cpu_set_hyp_vector(void)
1831 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1832 void *vector = hyp_spectre_vector_selector[data->slot];
1834 if (!is_protected_kvm_enabled())
1835 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1837 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1840 static void cpu_hyp_init_context(void)
1842 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1844 if (!is_kernel_in_hyp_mode())
1845 cpu_init_hyp_mode();
1848 static void cpu_hyp_init_features(void)
1850 cpu_set_hyp_vector();
1851 kvm_arm_init_debug();
1853 if (is_kernel_in_hyp_mode())
1854 kvm_timer_init_vhe();
1857 kvm_vgic_init_cpu_hardware();
1860 static void cpu_hyp_reinit(void)
1863 cpu_hyp_init_context();
1864 cpu_hyp_init_features();
1867 static void cpu_hyp_init(void *discard)
1869 if (!__this_cpu_read(kvm_hyp_initialized)) {
1871 __this_cpu_write(kvm_hyp_initialized, 1);
1875 static void cpu_hyp_uninit(void *discard)
1877 if (__this_cpu_read(kvm_hyp_initialized)) {
1879 __this_cpu_write(kvm_hyp_initialized, 0);
1883 int kvm_arch_hardware_enable(void)
1886 * Most calls to this function are made with migration
1887 * disabled, but not with preemption disabled. The former is
1888 * enough to ensure correctness, but most of the helpers
1889 * expect the later and will throw a tantrum otherwise.
1903 void kvm_arch_hardware_disable(void)
1905 kvm_timer_cpu_down();
1906 kvm_vgic_cpu_down();
1908 if (!is_protected_kvm_enabled())
1909 cpu_hyp_uninit(NULL);
1912 #ifdef CONFIG_CPU_PM
1913 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1918 * kvm_hyp_initialized is left with its old value over
1919 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1924 if (__this_cpu_read(kvm_hyp_initialized))
1926 * don't update kvm_hyp_initialized here
1927 * so that the hyp will be re-enabled
1928 * when we resume. See below.
1933 case CPU_PM_ENTER_FAILED:
1935 if (__this_cpu_read(kvm_hyp_initialized))
1936 /* The hyp was enabled before suspend. */
1946 static struct notifier_block hyp_init_cpu_pm_nb = {
1947 .notifier_call = hyp_init_cpu_pm_notifier,
1950 static void __init hyp_cpu_pm_init(void)
1952 if (!is_protected_kvm_enabled())
1953 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1955 static void __init hyp_cpu_pm_exit(void)
1957 if (!is_protected_kvm_enabled())
1958 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1961 static inline void __init hyp_cpu_pm_init(void)
1964 static inline void __init hyp_cpu_pm_exit(void)
1969 static void __init init_cpu_logical_map(void)
1974 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1975 * Only copy the set of online CPUs whose features have been checked
1976 * against the finalized system capabilities. The hypervisor will not
1977 * allow any other CPUs from the `possible` set to boot.
1979 for_each_online_cpu(cpu)
1980 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1983 #define init_psci_0_1_impl_state(config, what) \
1984 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1986 static bool __init init_psci_relay(void)
1989 * If PSCI has not been initialized, protected KVM cannot install
1990 * itself on newly booted CPUs.
1992 if (!psci_ops.get_version) {
1993 kvm_err("Cannot initialize protected mode without PSCI\n");
1997 kvm_host_psci_config.version = psci_ops.get_version();
1998 kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2000 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2001 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2002 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2003 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2004 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2005 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2010 static int __init init_subsystems(void)
2015 * Enable hardware so that subsystem initialisation can access EL2.
2017 on_each_cpu(cpu_hyp_init, NULL, 1);
2020 * Register CPU lower-power notifier
2025 * Init HYP view of VGIC
2027 err = kvm_vgic_hyp_init();
2030 vgic_present = true;
2034 vgic_present = false;
2042 * Init HYP architected timer support
2044 err = kvm_timer_hyp_init(vgic_present);
2048 kvm_register_perf_callbacks(NULL);
2054 if (err || !is_protected_kvm_enabled())
2055 on_each_cpu(cpu_hyp_uninit, NULL, 1);
2060 static void __init teardown_subsystems(void)
2062 kvm_unregister_perf_callbacks();
2066 static void __init teardown_hyp_mode(void)
2071 for_each_possible_cpu(cpu) {
2072 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2073 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2077 static int __init do_pkvm_init(u32 hyp_va_bits)
2079 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2083 cpu_hyp_init_context();
2084 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2085 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2087 cpu_hyp_init_features();
2090 * The stub hypercalls are now disabled, so set our local flag to
2091 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2093 __this_cpu_write(kvm_hyp_initialized, 1);
2099 static u64 get_hyp_id_aa64pfr0_el1(void)
2102 * Track whether the system isn't affected by spectre/meltdown in the
2103 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2104 * Although this is per-CPU, we make it global for simplicity, e.g., not
2105 * to have to worry about vcpu migration.
2107 * Unlike for non-protected VMs, userspace cannot override this for
2110 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2112 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2113 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2115 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2116 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2117 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2118 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2123 static void kvm_hyp_init_symbols(void)
2125 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2126 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2127 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2128 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2129 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2130 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2131 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2132 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2133 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2134 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2135 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2138 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2140 void *addr = phys_to_virt(hyp_mem_base);
2143 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2147 ret = do_pkvm_init(hyp_va_bits);
2156 static void pkvm_hyp_init_ptrauth(void)
2158 struct kvm_cpu_context *hyp_ctxt;
2161 for_each_possible_cpu(cpu) {
2162 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2163 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2164 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2165 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2166 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2167 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2168 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2169 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2170 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2171 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2172 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2176 /* Inits Hyp-mode on all online CPUs */
2177 static int __init init_hyp_mode(void)
2184 * The protected Hyp-mode cannot be initialized if the memory pool
2185 * allocation has failed.
2187 if (is_protected_kvm_enabled() && !hyp_mem_base)
2191 * Allocate Hyp PGD and setup Hyp identity mapping
2193 err = kvm_mmu_init(&hyp_va_bits);
2198 * Allocate stack pages for Hypervisor-mode
2200 for_each_possible_cpu(cpu) {
2201 unsigned long stack_page;
2203 stack_page = __get_free_page(GFP_KERNEL);
2209 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2213 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2215 for_each_possible_cpu(cpu) {
2219 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2225 page_addr = page_address(page);
2226 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2227 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2231 * Map the Hyp-code called directly from the host
2233 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2234 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2236 kvm_err("Cannot map world-switch code\n");
2240 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2241 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2243 kvm_err("Cannot map .hyp.rodata section\n");
2247 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2248 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2250 kvm_err("Cannot map rodata section\n");
2255 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2256 * section thanks to an assertion in the linker script. Map it RW and
2257 * the rest of .bss RO.
2259 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2260 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2262 kvm_err("Cannot map hyp bss section: %d\n", err);
2266 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2267 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2269 kvm_err("Cannot map bss section\n");
2274 * Map the Hyp stack pages
2276 for_each_possible_cpu(cpu) {
2277 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2278 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2279 unsigned long hyp_addr;
2282 * Allocate a contiguous HYP private VA range for the stack
2283 * and guard page. The allocation is also aligned based on
2284 * the order of its size.
2286 err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr);
2288 kvm_err("Cannot allocate hyp stack guard page\n");
2293 * Since the stack grows downwards, map the stack to the page
2294 * at the higher address and leave the lower guard page
2297 * Any valid stack address now has the PAGE_SHIFT bit as 1
2298 * and addresses corresponding to the guard page have the
2299 * PAGE_SHIFT bit as 0 - this is used for overflow detection.
2301 err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE,
2302 __pa(stack_page), PAGE_HYP);
2304 kvm_err("Cannot map hyp stack\n");
2309 * Save the stack PA in nvhe_init_params. This will be needed
2310 * to recreate the stack mapping in protected nVHE mode.
2311 * __hyp_pa() won't do the right thing there, since the stack
2312 * has been mapped in the flexible private VA space.
2314 params->stack_pa = __pa(stack_page);
2316 params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE);
2319 for_each_possible_cpu(cpu) {
2320 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2321 char *percpu_end = percpu_begin + nvhe_percpu_size();
2323 /* Map Hyp percpu pages */
2324 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2326 kvm_err("Cannot map hyp percpu region\n");
2330 /* Prepare the CPU initialization parameters */
2331 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2334 kvm_hyp_init_symbols();
2336 if (is_protected_kvm_enabled()) {
2337 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2338 cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH))
2339 pkvm_hyp_init_ptrauth();
2341 init_cpu_logical_map();
2343 if (!init_psci_relay()) {
2348 err = kvm_hyp_init_protection(hyp_va_bits);
2350 kvm_err("Failed to init hyp memory protection\n");
2358 teardown_hyp_mode();
2359 kvm_err("error initializing Hyp mode: %d\n", err);
2363 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2365 struct kvm_vcpu *vcpu;
2368 mpidr &= MPIDR_HWID_BITMASK;
2369 kvm_for_each_vcpu(i, vcpu, kvm) {
2370 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2376 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2378 return irqchip_in_kernel(kvm);
2381 bool kvm_arch_has_irq_bypass(void)
2386 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2387 struct irq_bypass_producer *prod)
2389 struct kvm_kernel_irqfd *irqfd =
2390 container_of(cons, struct kvm_kernel_irqfd, consumer);
2392 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2395 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2396 struct irq_bypass_producer *prod)
2398 struct kvm_kernel_irqfd *irqfd =
2399 container_of(cons, struct kvm_kernel_irqfd, consumer);
2401 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2405 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2407 struct kvm_kernel_irqfd *irqfd =
2408 container_of(cons, struct kvm_kernel_irqfd, consumer);
2410 kvm_arm_halt_guest(irqfd->kvm);
2413 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2415 struct kvm_kernel_irqfd *irqfd =
2416 container_of(cons, struct kvm_kernel_irqfd, consumer);
2418 kvm_arm_resume_guest(irqfd->kvm);
2421 /* Initialize Hyp-mode and memory mappings on all CPUs */
2422 static __init int kvm_arm_init(void)
2427 if (!is_hyp_mode_available()) {
2428 kvm_info("HYP mode not available\n");
2432 if (kvm_get_mode() == KVM_MODE_NONE) {
2433 kvm_info("KVM disabled from command line\n");
2437 err = kvm_sys_reg_table_init();
2439 kvm_info("Error initializing system register tables");
2443 in_hyp_mode = is_kernel_in_hyp_mode();
2445 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2446 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2447 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2448 "Only trusted guests should be used on this system.\n");
2450 err = kvm_set_ipa_limit();
2454 err = kvm_arm_init_sve();
2458 err = kvm_arm_vmid_alloc_init();
2460 kvm_err("Failed to initialize VMID allocator.\n");
2465 err = init_hyp_mode();
2470 err = kvm_init_vector_slots();
2472 kvm_err("Cannot initialise vector slots\n");
2476 err = init_subsystems();
2480 if (is_protected_kvm_enabled()) {
2481 kvm_info("Protected nVHE mode initialized successfully\n");
2482 } else if (in_hyp_mode) {
2483 kvm_info("VHE mode initialized successfully\n");
2485 kvm_info("Hyp mode initialized successfully\n");
2489 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2490 * hypervisor protection is finalized.
2492 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2496 kvm_arm_initialised = true;
2501 teardown_subsystems();
2504 teardown_hyp_mode();
2506 kvm_arm_vmid_alloc_free();
2510 static int __init early_kvm_mode_cfg(char *arg)
2515 if (strcmp(arg, "none") == 0) {
2516 kvm_mode = KVM_MODE_NONE;
2520 if (!is_hyp_mode_available()) {
2521 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2525 if (strcmp(arg, "protected") == 0) {
2526 if (!is_kernel_in_hyp_mode())
2527 kvm_mode = KVM_MODE_PROTECTED;
2529 pr_warn_once("Protected KVM not available with VHE\n");
2534 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2535 kvm_mode = KVM_MODE_DEFAULT;
2539 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2540 kvm_mode = KVM_MODE_NV;
2546 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2548 enum kvm_mode kvm_get_mode(void)
2553 module_init(kvm_arm_init);