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_nested.h>
40 #include <asm/kvm_pkvm.h>
41 #include <asm/kvm_emulate.h>
42 #include <asm/sections.h>
44 #include <kvm/arm_hypercalls.h>
45 #include <kvm/arm_pmu.h>
46 #include <kvm/arm_psci.h>
48 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
52 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
55 DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
57 static bool vgic_present, kvm_arm_initialised;
59 static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized);
60 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
62 bool is_kvm_arm_initialised(void)
64 return kvm_arm_initialised;
67 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
69 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
72 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
73 struct kvm_enable_cap *cap)
82 case KVM_CAP_ARM_NISV_TO_USER:
84 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
88 mutex_lock(&kvm->lock);
89 if (!system_supports_mte() || kvm->created_vcpus) {
93 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
95 mutex_unlock(&kvm->lock);
97 case KVM_CAP_ARM_SYSTEM_SUSPEND:
99 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
101 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
102 new_cap = cap->args[0];
104 mutex_lock(&kvm->slots_lock);
106 * To keep things simple, allow changing the chunk
107 * size only when no memory slots have been created.
109 if (!kvm_are_all_memslots_empty(kvm)) {
111 } else if (new_cap && !kvm_is_block_size_supported(new_cap)) {
115 kvm->arch.mmu.split_page_chunk_size = new_cap;
117 mutex_unlock(&kvm->slots_lock);
127 static int kvm_arm_default_max_vcpus(void)
129 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
133 * kvm_arch_init_vm - initializes a VM data structure
134 * @kvm: pointer to the KVM struct
136 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
140 mutex_init(&kvm->arch.config_lock);
142 #ifdef CONFIG_LOCKDEP
143 /* Clue in lockdep that the config_lock must be taken inside kvm->lock */
144 mutex_lock(&kvm->lock);
145 mutex_lock(&kvm->arch.config_lock);
146 mutex_unlock(&kvm->arch.config_lock);
147 mutex_unlock(&kvm->lock);
150 ret = kvm_share_hyp(kvm, kvm + 1);
154 ret = pkvm_init_host_vm(kvm);
156 goto err_unshare_kvm;
158 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
160 goto err_unshare_kvm;
162 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
164 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
166 goto err_free_cpumask;
168 kvm_vgic_early_init(kvm);
170 kvm_timer_init_vm(kvm);
172 /* The maximum number of VCPUs is limited by the host's GIC model */
173 kvm->max_vcpus = kvm_arm_default_max_vcpus();
175 kvm_arm_init_hypercalls(kvm);
177 bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
182 free_cpumask_var(kvm->arch.supported_cpus);
184 kvm_unshare_hyp(kvm, kvm + 1);
188 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
190 return VM_FAULT_SIGBUS;
195 * kvm_arch_destroy_vm - destroy the VM data structure
196 * @kvm: pointer to the KVM struct
198 void kvm_arch_destroy_vm(struct kvm *kvm)
200 bitmap_free(kvm->arch.pmu_filter);
201 free_cpumask_var(kvm->arch.supported_cpus);
203 kvm_vgic_destroy(kvm);
205 if (is_protected_kvm_enabled())
206 pkvm_destroy_hyp_vm(kvm);
208 kvm_destroy_vcpus(kvm);
210 kvm_unshare_hyp(kvm, kvm + 1);
212 kvm_arm_teardown_hypercalls(kvm);
215 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
219 case KVM_CAP_IRQCHIP:
222 case KVM_CAP_IOEVENTFD:
223 case KVM_CAP_DEVICE_CTRL:
224 case KVM_CAP_USER_MEMORY:
225 case KVM_CAP_SYNC_MMU:
226 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
227 case KVM_CAP_ONE_REG:
228 case KVM_CAP_ARM_PSCI:
229 case KVM_CAP_ARM_PSCI_0_2:
230 case KVM_CAP_READONLY_MEM:
231 case KVM_CAP_MP_STATE:
232 case KVM_CAP_IMMEDIATE_EXIT:
233 case KVM_CAP_VCPU_EVENTS:
234 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
235 case KVM_CAP_ARM_NISV_TO_USER:
236 case KVM_CAP_ARM_INJECT_EXT_DABT:
237 case KVM_CAP_SET_GUEST_DEBUG:
238 case KVM_CAP_VCPU_ATTRIBUTES:
239 case KVM_CAP_PTP_KVM:
240 case KVM_CAP_ARM_SYSTEM_SUSPEND:
241 case KVM_CAP_IRQFD_RESAMPLE:
242 case KVM_CAP_COUNTER_OFFSET:
245 case KVM_CAP_SET_GUEST_DEBUG2:
246 return KVM_GUESTDBG_VALID_MASK;
247 case KVM_CAP_ARM_SET_DEVICE_ADDR:
250 case KVM_CAP_NR_VCPUS:
252 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
253 * architectures, as it does not always bound it to
254 * KVM_CAP_MAX_VCPUS. It should not matter much because
255 * this is just an advisory value.
257 r = min_t(unsigned int, num_online_cpus(),
258 kvm_arm_default_max_vcpus());
260 case KVM_CAP_MAX_VCPUS:
261 case KVM_CAP_MAX_VCPU_ID:
265 r = kvm_arm_default_max_vcpus();
267 case KVM_CAP_MSI_DEVID:
271 r = kvm->arch.vgic.msis_require_devid;
273 case KVM_CAP_ARM_USER_IRQ:
275 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
276 * (bump this number if adding more devices)
280 case KVM_CAP_ARM_MTE:
281 r = system_supports_mte();
283 case KVM_CAP_STEAL_TIME:
284 r = kvm_arm_pvtime_supported();
286 case KVM_CAP_ARM_EL1_32BIT:
287 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
289 case KVM_CAP_GUEST_DEBUG_HW_BPS:
292 case KVM_CAP_GUEST_DEBUG_HW_WPS:
295 case KVM_CAP_ARM_PMU_V3:
296 r = kvm_arm_support_pmu_v3();
298 case KVM_CAP_ARM_INJECT_SERROR_ESR:
299 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
301 case KVM_CAP_ARM_VM_IPA_SIZE:
302 r = get_kvm_ipa_limit();
304 case KVM_CAP_ARM_SVE:
305 r = system_supports_sve();
307 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
308 case KVM_CAP_ARM_PTRAUTH_GENERIC:
309 r = system_has_full_ptr_auth();
311 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
313 r = kvm->arch.mmu.split_page_chunk_size;
315 r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
317 case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
318 r = kvm_supported_block_sizes();
327 long kvm_arch_dev_ioctl(struct file *filp,
328 unsigned int ioctl, unsigned long arg)
333 struct kvm *kvm_arch_alloc_vm(void)
335 size_t sz = sizeof(struct kvm);
338 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
340 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
343 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
345 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
348 if (id >= kvm->max_vcpus)
354 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
358 spin_lock_init(&vcpu->arch.mp_state_lock);
360 #ifdef CONFIG_LOCKDEP
361 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
362 mutex_lock(&vcpu->mutex);
363 mutex_lock(&vcpu->kvm->arch.config_lock);
364 mutex_unlock(&vcpu->kvm->arch.config_lock);
365 mutex_unlock(&vcpu->mutex);
368 /* Force users to call KVM_ARM_VCPU_INIT */
369 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
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(cpu, 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_get_flag(vcpu, VCPU_INITIALIZED);
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_RESYNC_PMU_EL0, vcpu))
807 kvm_vcpu_pmu_restore_guest(vcpu);
809 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
810 return kvm_vcpu_suspend(vcpu);
812 if (kvm_dirty_ring_check_request(vcpu))
819 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
821 if (likely(!vcpu_mode_is_32bit(vcpu)))
824 if (vcpu_has_nv(vcpu))
827 return !kvm_supports_32bit_el0();
831 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
832 * @vcpu: The VCPU pointer
833 * @ret: Pointer to write optional return code
835 * Returns: true if the VCPU needs to return to a preemptible + interruptible
836 * and skip guest entry.
838 * This function disambiguates between two different types of exits: exits to a
839 * preemptible + interruptible kernel context and exits to userspace. For an
840 * exit to userspace, this function will write the return code to ret and return
841 * true. For an exit to preemptible + interruptible kernel context (i.e. check
842 * for pending work and re-enter), return true without writing to ret.
844 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
846 struct kvm_run *run = vcpu->run;
849 * If we're using a userspace irqchip, then check if we need
850 * to tell a userspace irqchip about timer or PMU level
851 * changes and if so, exit to userspace (the actual level
852 * state gets updated in kvm_timer_update_run and
853 * kvm_pmu_update_run below).
855 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
856 if (kvm_timer_should_notify_user(vcpu) ||
857 kvm_pmu_should_notify_user(vcpu)) {
859 run->exit_reason = KVM_EXIT_INTR;
864 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
865 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
866 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
867 run->fail_entry.cpu = smp_processor_id();
872 return kvm_request_pending(vcpu) ||
873 xfer_to_guest_mode_work_pending();
877 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
878 * the vCPU is running.
880 * This must be noinstr as instrumentation may make use of RCU, and this is not
881 * safe during the EQS.
883 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
887 guest_state_enter_irqoff();
888 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
889 guest_state_exit_irqoff();
895 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
896 * @vcpu: The VCPU pointer
898 * This function is called through the VCPU_RUN ioctl called from user space. It
899 * will execute VM code in a loop until the time slice for the process is used
900 * or some emulation is needed from user space in which case the function will
901 * return with return value 0 and with the kvm_run structure filled in with the
902 * required data for the requested emulation.
904 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
906 struct kvm_run *run = vcpu->run;
909 if (run->exit_reason == KVM_EXIT_MMIO) {
910 ret = kvm_handle_mmio_return(vcpu);
917 if (run->immediate_exit) {
922 kvm_sigset_activate(vcpu);
925 run->exit_reason = KVM_EXIT_UNKNOWN;
929 * Check conditions before entering the guest
931 ret = xfer_to_guest_mode_handle_work(vcpu);
936 ret = check_vcpu_requests(vcpu);
939 * Preparing the interrupts to be injected also
940 * involves poking the GIC, which must be done in a
941 * non-preemptible context.
946 * The VMID allocator only tracks active VMIDs per
947 * physical CPU, and therefore the VMID allocated may not be
948 * preserved on VMID roll-over if the task was preempted,
949 * making a thread's VMID inactive. So we need to call
950 * kvm_arm_vmid_update() in non-premptible context.
952 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
954 kvm_pmu_flush_hwstate(vcpu);
958 kvm_vgic_flush_hwstate(vcpu);
960 kvm_pmu_update_vcpu_events(vcpu);
963 * Ensure we set mode to IN_GUEST_MODE after we disable
964 * interrupts and before the final VCPU requests check.
965 * See the comment in kvm_vcpu_exiting_guest_mode() and
966 * Documentation/virt/kvm/vcpu-requests.rst
968 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
970 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
971 vcpu->mode = OUTSIDE_GUEST_MODE;
972 isb(); /* Ensure work in x_flush_hwstate is committed */
973 kvm_pmu_sync_hwstate(vcpu);
974 if (static_branch_unlikely(&userspace_irqchip_in_use))
975 kvm_timer_sync_user(vcpu);
976 kvm_vgic_sync_hwstate(vcpu);
982 kvm_arm_setup_debug(vcpu);
983 kvm_arch_vcpu_ctxflush_fp(vcpu);
985 /**************************************************************
988 trace_kvm_entry(*vcpu_pc(vcpu));
989 guest_timing_enter_irqoff();
991 ret = kvm_arm_vcpu_enter_exit(vcpu);
993 vcpu->mode = OUTSIDE_GUEST_MODE;
997 *************************************************************/
999 kvm_arm_clear_debug(vcpu);
1002 * We must sync the PMU state before the vgic state so
1003 * that the vgic can properly sample the updated state of the
1006 kvm_pmu_sync_hwstate(vcpu);
1009 * Sync the vgic state before syncing the timer state because
1010 * the timer code needs to know if the virtual timer
1011 * interrupts are active.
1013 kvm_vgic_sync_hwstate(vcpu);
1016 * Sync the timer hardware state before enabling interrupts as
1017 * we don't want vtimer interrupts to race with syncing the
1018 * timer virtual interrupt state.
1020 if (static_branch_unlikely(&userspace_irqchip_in_use))
1021 kvm_timer_sync_user(vcpu);
1023 kvm_arch_vcpu_ctxsync_fp(vcpu);
1026 * We must ensure that any pending interrupts are taken before
1027 * we exit guest timing so that timer ticks are accounted as
1028 * guest time. Transiently unmask interrupts so that any
1029 * pending interrupts are taken.
1031 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1032 * context synchronization event) is necessary to ensure that
1033 * pending interrupts are taken.
1035 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1038 local_irq_disable();
1041 guest_timing_exit_irqoff();
1045 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1047 /* Exit types that need handling before we can be preempted */
1048 handle_exit_early(vcpu, ret);
1053 * The ARMv8 architecture doesn't give the hypervisor
1054 * a mechanism to prevent a guest from dropping to AArch32 EL0
1055 * if implemented by the CPU. If we spot the guest in such
1056 * state and that we decided it wasn't supposed to do so (like
1057 * with the asymmetric AArch32 case), return to userspace with
1060 if (vcpu_mode_is_bad_32bit(vcpu)) {
1062 * As we have caught the guest red-handed, decide that
1063 * it isn't fit for purpose anymore by making the vcpu
1064 * invalid. The VMM can try and fix it by issuing a
1065 * KVM_ARM_VCPU_INIT if it really wants to.
1067 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1068 ret = ARM_EXCEPTION_IL;
1071 ret = handle_exit(vcpu, ret);
1074 /* Tell userspace about in-kernel device output levels */
1075 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1076 kvm_timer_update_run(vcpu);
1077 kvm_pmu_update_run(vcpu);
1080 kvm_sigset_deactivate(vcpu);
1084 * In the unlikely event that we are returning to userspace
1085 * with pending exceptions or PC adjustment, commit these
1086 * adjustments in order to give userspace a consistent view of
1087 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1088 * being preempt-safe on VHE.
1090 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1091 vcpu_get_flag(vcpu, INCREMENT_PC)))
1092 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1098 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1104 if (number == KVM_ARM_IRQ_CPU_IRQ)
1105 bit_index = __ffs(HCR_VI);
1106 else /* KVM_ARM_IRQ_CPU_FIQ */
1107 bit_index = __ffs(HCR_VF);
1109 hcr = vcpu_hcr(vcpu);
1111 set = test_and_set_bit(bit_index, hcr);
1113 set = test_and_clear_bit(bit_index, hcr);
1116 * If we didn't change anything, no need to wake up or kick other CPUs
1122 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1123 * trigger a world-switch round on the running physical CPU to set the
1124 * virtual IRQ/FIQ fields in the HCR appropriately.
1126 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1127 kvm_vcpu_kick(vcpu);
1132 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1135 u32 irq = irq_level->irq;
1136 unsigned int irq_type, vcpu_idx, irq_num;
1137 int nrcpus = atomic_read(&kvm->online_vcpus);
1138 struct kvm_vcpu *vcpu = NULL;
1139 bool level = irq_level->level;
1141 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1142 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1143 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1144 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1146 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1149 case KVM_ARM_IRQ_TYPE_CPU:
1150 if (irqchip_in_kernel(kvm))
1153 if (vcpu_idx >= nrcpus)
1156 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1160 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1163 return vcpu_interrupt_line(vcpu, irq_num, level);
1164 case KVM_ARM_IRQ_TYPE_PPI:
1165 if (!irqchip_in_kernel(kvm))
1168 if (vcpu_idx >= nrcpus)
1171 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1175 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1178 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1179 case KVM_ARM_IRQ_TYPE_SPI:
1180 if (!irqchip_in_kernel(kvm))
1183 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1186 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1192 static unsigned long system_supported_vcpu_features(void)
1194 unsigned long features = KVM_VCPU_VALID_FEATURES;
1196 if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1197 clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1199 if (!kvm_arm_support_pmu_v3())
1200 clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1202 if (!system_supports_sve())
1203 clear_bit(KVM_ARM_VCPU_SVE, &features);
1205 if (!system_has_full_ptr_auth()) {
1206 clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1207 clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1210 if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1211 clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1216 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1217 const struct kvm_vcpu_init *init)
1219 unsigned long features = init->features[0];
1222 if (features & ~KVM_VCPU_VALID_FEATURES)
1225 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1226 if (init->features[i])
1230 if (features & ~system_supported_vcpu_features())
1234 * For now make sure that both address/generic pointer authentication
1235 * features are requested by the userspace together.
1237 if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1238 test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1241 /* Disallow NV+SVE for the time being */
1242 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features) &&
1243 test_bit(KVM_ARM_VCPU_SVE, &features))
1246 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1249 /* MTE is incompatible with AArch32 */
1250 if (kvm_has_mte(vcpu->kvm))
1253 /* NV is incompatible with AArch32 */
1254 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1260 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1261 const struct kvm_vcpu_init *init)
1263 unsigned long features = init->features[0];
1265 return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1266 KVM_VCPU_MAX_FEATURES);
1269 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1270 const struct kvm_vcpu_init *init)
1272 unsigned long features = init->features[0];
1273 struct kvm *kvm = vcpu->kvm;
1276 mutex_lock(&kvm->arch.config_lock);
1278 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1279 kvm_vcpu_init_changed(vcpu, init))
1282 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1284 /* Now we know what it is, we can reset it. */
1285 kvm_reset_vcpu(vcpu);
1287 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1288 vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1291 mutex_unlock(&kvm->arch.config_lock);
1295 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1296 const struct kvm_vcpu_init *init)
1300 if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1301 init->target != kvm_target_cpu())
1304 ret = kvm_vcpu_init_check_features(vcpu, init);
1308 if (!kvm_vcpu_initialized(vcpu))
1309 return __kvm_vcpu_set_target(vcpu, init);
1311 if (kvm_vcpu_init_changed(vcpu, init))
1314 kvm_reset_vcpu(vcpu);
1318 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1319 struct kvm_vcpu_init *init)
1321 bool power_off = false;
1325 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1326 * reflecting it in the finalized feature set, thus limiting its scope
1327 * to a single KVM_ARM_VCPU_INIT call.
1329 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1330 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1334 ret = kvm_vcpu_set_target(vcpu, init);
1339 * Ensure a rebooted VM will fault in RAM pages and detect if the
1340 * guest MMU is turned off and flush the caches as needed.
1342 * S2FWB enforces all memory accesses to RAM being cacheable,
1343 * ensuring that the data side is always coherent. We still
1344 * need to invalidate the I-cache though, as FWB does *not*
1345 * imply CTR_EL0.DIC.
1347 if (vcpu_has_run_once(vcpu)) {
1348 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1349 stage2_unmap_vm(vcpu->kvm);
1351 icache_inval_all_pou();
1354 vcpu_reset_hcr(vcpu);
1355 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1358 * Handle the "start in power-off" case.
1360 spin_lock(&vcpu->arch.mp_state_lock);
1363 __kvm_arm_vcpu_power_off(vcpu);
1365 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1367 spin_unlock(&vcpu->arch.mp_state_lock);
1372 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1373 struct kvm_device_attr *attr)
1377 switch (attr->group) {
1379 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1386 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1387 struct kvm_device_attr *attr)
1391 switch (attr->group) {
1393 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1400 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1401 struct kvm_device_attr *attr)
1405 switch (attr->group) {
1407 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1414 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1415 struct kvm_vcpu_events *events)
1417 memset(events, 0, sizeof(*events));
1419 return __kvm_arm_vcpu_get_events(vcpu, events);
1422 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1423 struct kvm_vcpu_events *events)
1427 /* check whether the reserved field is zero */
1428 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1429 if (events->reserved[i])
1432 /* check whether the pad field is zero */
1433 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1434 if (events->exception.pad[i])
1437 return __kvm_arm_vcpu_set_events(vcpu, events);
1440 long kvm_arch_vcpu_ioctl(struct file *filp,
1441 unsigned int ioctl, unsigned long arg)
1443 struct kvm_vcpu *vcpu = filp->private_data;
1444 void __user *argp = (void __user *)arg;
1445 struct kvm_device_attr attr;
1449 case KVM_ARM_VCPU_INIT: {
1450 struct kvm_vcpu_init init;
1453 if (copy_from_user(&init, argp, sizeof(init)))
1456 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1459 case KVM_SET_ONE_REG:
1460 case KVM_GET_ONE_REG: {
1461 struct kvm_one_reg reg;
1464 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1468 if (copy_from_user(®, argp, sizeof(reg)))
1472 * We could owe a reset due to PSCI. Handle the pending reset
1473 * here to ensure userspace register accesses are ordered after
1476 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1477 kvm_reset_vcpu(vcpu);
1479 if (ioctl == KVM_SET_ONE_REG)
1480 r = kvm_arm_set_reg(vcpu, ®);
1482 r = kvm_arm_get_reg(vcpu, ®);
1485 case KVM_GET_REG_LIST: {
1486 struct kvm_reg_list __user *user_list = argp;
1487 struct kvm_reg_list reg_list;
1491 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1495 if (!kvm_arm_vcpu_is_finalized(vcpu))
1499 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1502 reg_list.n = kvm_arm_num_regs(vcpu);
1503 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1508 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1511 case KVM_SET_DEVICE_ATTR: {
1513 if (copy_from_user(&attr, argp, sizeof(attr)))
1515 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1518 case KVM_GET_DEVICE_ATTR: {
1520 if (copy_from_user(&attr, argp, sizeof(attr)))
1522 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1525 case KVM_HAS_DEVICE_ATTR: {
1527 if (copy_from_user(&attr, argp, sizeof(attr)))
1529 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1532 case KVM_GET_VCPU_EVENTS: {
1533 struct kvm_vcpu_events events;
1535 if (kvm_arm_vcpu_get_events(vcpu, &events))
1538 if (copy_to_user(argp, &events, sizeof(events)))
1543 case KVM_SET_VCPU_EVENTS: {
1544 struct kvm_vcpu_events events;
1546 if (copy_from_user(&events, argp, sizeof(events)))
1549 return kvm_arm_vcpu_set_events(vcpu, &events);
1551 case KVM_ARM_VCPU_FINALIZE: {
1554 if (!kvm_vcpu_initialized(vcpu))
1557 if (get_user(what, (const int __user *)argp))
1560 return kvm_arm_vcpu_finalize(vcpu, what);
1569 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1574 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1575 struct kvm_arm_device_addr *dev_addr)
1577 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1578 case KVM_ARM_DEVICE_VGIC_V2:
1581 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1587 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1589 switch (attr->group) {
1590 case KVM_ARM_VM_SMCCC_CTRL:
1591 return kvm_vm_smccc_has_attr(kvm, attr);
1597 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1599 switch (attr->group) {
1600 case KVM_ARM_VM_SMCCC_CTRL:
1601 return kvm_vm_smccc_set_attr(kvm, attr);
1607 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1609 struct kvm *kvm = filp->private_data;
1610 void __user *argp = (void __user *)arg;
1611 struct kvm_device_attr attr;
1614 case KVM_CREATE_IRQCHIP: {
1618 mutex_lock(&kvm->lock);
1619 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1620 mutex_unlock(&kvm->lock);
1623 case KVM_ARM_SET_DEVICE_ADDR: {
1624 struct kvm_arm_device_addr dev_addr;
1626 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1628 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1630 case KVM_ARM_PREFERRED_TARGET: {
1631 struct kvm_vcpu_init init = {
1632 .target = KVM_ARM_TARGET_GENERIC_V8,
1635 if (copy_to_user(argp, &init, sizeof(init)))
1640 case KVM_ARM_MTE_COPY_TAGS: {
1641 struct kvm_arm_copy_mte_tags copy_tags;
1643 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1645 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1647 case KVM_ARM_SET_COUNTER_OFFSET: {
1648 struct kvm_arm_counter_offset offset;
1650 if (copy_from_user(&offset, argp, sizeof(offset)))
1652 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1654 case KVM_HAS_DEVICE_ATTR: {
1655 if (copy_from_user(&attr, argp, sizeof(attr)))
1658 return kvm_vm_has_attr(kvm, &attr);
1660 case KVM_SET_DEVICE_ATTR: {
1661 if (copy_from_user(&attr, argp, sizeof(attr)))
1664 return kvm_vm_set_attr(kvm, &attr);
1671 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1672 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1674 struct kvm_vcpu *tmp_vcpu;
1676 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1677 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1678 mutex_unlock(&tmp_vcpu->mutex);
1682 void unlock_all_vcpus(struct kvm *kvm)
1684 lockdep_assert_held(&kvm->lock);
1686 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1689 /* Returns true if all vcpus were locked, false otherwise */
1690 bool lock_all_vcpus(struct kvm *kvm)
1692 struct kvm_vcpu *tmp_vcpu;
1695 lockdep_assert_held(&kvm->lock);
1698 * Any time a vcpu is in an ioctl (including running), the
1699 * core KVM code tries to grab the vcpu->mutex.
1701 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1702 * other VCPUs can fiddle with the state while we access it.
1704 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1705 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1706 unlock_vcpus(kvm, c - 1);
1714 static unsigned long nvhe_percpu_size(void)
1716 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1717 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1720 static unsigned long nvhe_percpu_order(void)
1722 unsigned long size = nvhe_percpu_size();
1724 return size ? get_order(size) : 0;
1727 /* A lookup table holding the hypervisor VA for each vector slot */
1728 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1730 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1732 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1735 static int kvm_init_vector_slots(void)
1740 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1741 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1743 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1744 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1746 if (kvm_system_needs_idmapped_vectors() &&
1747 !is_protected_kvm_enabled()) {
1748 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1749 __BP_HARDEN_HYP_VECS_SZ, &base);
1754 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1755 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1759 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1761 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1765 * Calculate the raw per-cpu offset without a translation from the
1766 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1767 * so that we can use adr_l to access per-cpu variables in EL2.
1768 * Also drop the KASAN tag which gets in the way...
1770 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1771 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1773 params->mair_el2 = read_sysreg(mair_el1);
1775 tcr = read_sysreg(tcr_el1);
1776 if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1777 tcr |= TCR_EPD1_MASK;
1779 tcr &= TCR_EL2_MASK;
1780 tcr |= TCR_EL2_RES1;
1782 tcr &= ~TCR_T0SZ_MASK;
1783 tcr |= TCR_T0SZ(hyp_va_bits);
1784 params->tcr_el2 = tcr;
1786 params->pgd_pa = kvm_mmu_get_httbr();
1787 if (is_protected_kvm_enabled())
1788 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1790 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1791 if (cpus_have_final_cap(ARM64_KVM_HVHE))
1792 params->hcr_el2 |= HCR_E2H;
1793 params->vttbr = params->vtcr = 0;
1796 * Flush the init params from the data cache because the struct will
1797 * be read while the MMU is off.
1799 kvm_flush_dcache_to_poc(params, sizeof(*params));
1802 static void hyp_install_host_vector(void)
1804 struct kvm_nvhe_init_params *params;
1805 struct arm_smccc_res res;
1807 /* Switch from the HYP stub to our own HYP init vector */
1808 __hyp_set_vectors(kvm_get_idmap_vector());
1811 * Call initialization code, and switch to the full blown HYP code.
1812 * If the cpucaps haven't been finalized yet, something has gone very
1813 * wrong, and hyp will crash and burn when it uses any
1814 * cpus_have_const_cap() wrapper.
1816 BUG_ON(!system_capabilities_finalized());
1817 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1818 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1819 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1822 static void cpu_init_hyp_mode(void)
1824 hyp_install_host_vector();
1827 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1830 if (this_cpu_has_cap(ARM64_SSBS) &&
1831 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1832 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1836 static void cpu_hyp_reset(void)
1838 if (!is_kernel_in_hyp_mode())
1839 __hyp_reset_vectors();
1843 * EL2 vectors can be mapped and rerouted in a number of ways,
1844 * depending on the kernel configuration and CPU present:
1846 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1847 * placed in one of the vector slots, which is executed before jumping
1848 * to the real vectors.
1850 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1851 * containing the hardening sequence is mapped next to the idmap page,
1852 * and executed before jumping to the real vectors.
1854 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1855 * empty slot is selected, mapped next to the idmap page, and
1856 * executed before jumping to the real vectors.
1858 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1859 * VHE, as we don't have hypervisor-specific mappings. If the system
1860 * is VHE and yet selects this capability, it will be ignored.
1862 static void cpu_set_hyp_vector(void)
1864 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1865 void *vector = hyp_spectre_vector_selector[data->slot];
1867 if (!is_protected_kvm_enabled())
1868 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1870 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1873 static void cpu_hyp_init_context(void)
1875 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1877 if (!is_kernel_in_hyp_mode())
1878 cpu_init_hyp_mode();
1881 static void cpu_hyp_init_features(void)
1883 cpu_set_hyp_vector();
1884 kvm_arm_init_debug();
1886 if (is_kernel_in_hyp_mode())
1887 kvm_timer_init_vhe();
1890 kvm_vgic_init_cpu_hardware();
1893 static void cpu_hyp_reinit(void)
1896 cpu_hyp_init_context();
1897 cpu_hyp_init_features();
1900 static void cpu_hyp_init(void *discard)
1902 if (!__this_cpu_read(kvm_hyp_initialized)) {
1904 __this_cpu_write(kvm_hyp_initialized, 1);
1908 static void cpu_hyp_uninit(void *discard)
1910 if (__this_cpu_read(kvm_hyp_initialized)) {
1912 __this_cpu_write(kvm_hyp_initialized, 0);
1916 int kvm_arch_hardware_enable(void)
1919 * Most calls to this function are made with migration
1920 * disabled, but not with preemption disabled. The former is
1921 * enough to ensure correctness, but most of the helpers
1922 * expect the later and will throw a tantrum otherwise.
1936 void kvm_arch_hardware_disable(void)
1938 kvm_timer_cpu_down();
1939 kvm_vgic_cpu_down();
1941 if (!is_protected_kvm_enabled())
1942 cpu_hyp_uninit(NULL);
1945 #ifdef CONFIG_CPU_PM
1946 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1951 * kvm_hyp_initialized is left with its old value over
1952 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1957 if (__this_cpu_read(kvm_hyp_initialized))
1959 * don't update kvm_hyp_initialized here
1960 * so that the hyp will be re-enabled
1961 * when we resume. See below.
1966 case CPU_PM_ENTER_FAILED:
1968 if (__this_cpu_read(kvm_hyp_initialized))
1969 /* The hyp was enabled before suspend. */
1979 static struct notifier_block hyp_init_cpu_pm_nb = {
1980 .notifier_call = hyp_init_cpu_pm_notifier,
1983 static void __init hyp_cpu_pm_init(void)
1985 if (!is_protected_kvm_enabled())
1986 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1988 static void __init hyp_cpu_pm_exit(void)
1990 if (!is_protected_kvm_enabled())
1991 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1994 static inline void __init hyp_cpu_pm_init(void)
1997 static inline void __init hyp_cpu_pm_exit(void)
2002 static void __init init_cpu_logical_map(void)
2007 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2008 * Only copy the set of online CPUs whose features have been checked
2009 * against the finalized system capabilities. The hypervisor will not
2010 * allow any other CPUs from the `possible` set to boot.
2012 for_each_online_cpu(cpu)
2013 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2016 #define init_psci_0_1_impl_state(config, what) \
2017 config.psci_0_1_ ## what ## _implemented = psci_ops.what
2019 static bool __init init_psci_relay(void)
2022 * If PSCI has not been initialized, protected KVM cannot install
2023 * itself on newly booted CPUs.
2025 if (!psci_ops.get_version) {
2026 kvm_err("Cannot initialize protected mode without PSCI\n");
2030 kvm_host_psci_config.version = psci_ops.get_version();
2031 kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2033 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2034 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2035 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2036 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2037 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2038 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2043 static int __init init_subsystems(void)
2048 * Enable hardware so that subsystem initialisation can access EL2.
2050 on_each_cpu(cpu_hyp_init, NULL, 1);
2053 * Register CPU lower-power notifier
2058 * Init HYP view of VGIC
2060 err = kvm_vgic_hyp_init();
2063 vgic_present = true;
2067 vgic_present = false;
2075 * Init HYP architected timer support
2077 err = kvm_timer_hyp_init(vgic_present);
2081 kvm_register_perf_callbacks(NULL);
2087 if (err || !is_protected_kvm_enabled())
2088 on_each_cpu(cpu_hyp_uninit, NULL, 1);
2093 static void __init teardown_subsystems(void)
2095 kvm_unregister_perf_callbacks();
2099 static void __init teardown_hyp_mode(void)
2104 for_each_possible_cpu(cpu) {
2105 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2106 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2110 static int __init do_pkvm_init(u32 hyp_va_bits)
2112 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2116 cpu_hyp_init_context();
2117 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2118 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2120 cpu_hyp_init_features();
2123 * The stub hypercalls are now disabled, so set our local flag to
2124 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2126 __this_cpu_write(kvm_hyp_initialized, 1);
2132 static u64 get_hyp_id_aa64pfr0_el1(void)
2135 * Track whether the system isn't affected by spectre/meltdown in the
2136 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2137 * Although this is per-CPU, we make it global for simplicity, e.g., not
2138 * to have to worry about vcpu migration.
2140 * Unlike for non-protected VMs, userspace cannot override this for
2143 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2145 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2146 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2148 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2149 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2150 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2151 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2156 static void kvm_hyp_init_symbols(void)
2158 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2159 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2160 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2161 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2162 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2163 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2164 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2165 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2166 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2167 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2168 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2171 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2173 void *addr = phys_to_virt(hyp_mem_base);
2176 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2180 ret = do_pkvm_init(hyp_va_bits);
2189 static void pkvm_hyp_init_ptrauth(void)
2191 struct kvm_cpu_context *hyp_ctxt;
2194 for_each_possible_cpu(cpu) {
2195 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2196 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2197 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2198 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2199 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2200 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2201 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2202 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2203 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2204 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2205 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2209 /* Inits Hyp-mode on all online CPUs */
2210 static int __init init_hyp_mode(void)
2217 * The protected Hyp-mode cannot be initialized if the memory pool
2218 * allocation has failed.
2220 if (is_protected_kvm_enabled() && !hyp_mem_base)
2224 * Allocate Hyp PGD and setup Hyp identity mapping
2226 err = kvm_mmu_init(&hyp_va_bits);
2231 * Allocate stack pages for Hypervisor-mode
2233 for_each_possible_cpu(cpu) {
2234 unsigned long stack_page;
2236 stack_page = __get_free_page(GFP_KERNEL);
2242 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2246 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2248 for_each_possible_cpu(cpu) {
2252 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2258 page_addr = page_address(page);
2259 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2260 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2264 * Map the Hyp-code called directly from the host
2266 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2267 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2269 kvm_err("Cannot map world-switch code\n");
2273 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2274 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2276 kvm_err("Cannot map .hyp.rodata section\n");
2280 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2281 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2283 kvm_err("Cannot map rodata section\n");
2288 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2289 * section thanks to an assertion in the linker script. Map it RW and
2290 * the rest of .bss RO.
2292 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2293 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2295 kvm_err("Cannot map hyp bss section: %d\n", err);
2299 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2300 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2302 kvm_err("Cannot map bss section\n");
2307 * Map the Hyp stack pages
2309 for_each_possible_cpu(cpu) {
2310 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2311 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2313 err = create_hyp_stack(__pa(stack_page), ¶ms->stack_hyp_va);
2315 kvm_err("Cannot map hyp stack\n");
2320 * Save the stack PA in nvhe_init_params. This will be needed
2321 * to recreate the stack mapping in protected nVHE mode.
2322 * __hyp_pa() won't do the right thing there, since the stack
2323 * has been mapped in the flexible private VA space.
2325 params->stack_pa = __pa(stack_page);
2328 for_each_possible_cpu(cpu) {
2329 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2330 char *percpu_end = percpu_begin + nvhe_percpu_size();
2332 /* Map Hyp percpu pages */
2333 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2335 kvm_err("Cannot map hyp percpu region\n");
2339 /* Prepare the CPU initialization parameters */
2340 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2343 kvm_hyp_init_symbols();
2345 if (is_protected_kvm_enabled()) {
2346 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2347 cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH))
2348 pkvm_hyp_init_ptrauth();
2350 init_cpu_logical_map();
2352 if (!init_psci_relay()) {
2357 err = kvm_hyp_init_protection(hyp_va_bits);
2359 kvm_err("Failed to init hyp memory protection\n");
2367 teardown_hyp_mode();
2368 kvm_err("error initializing Hyp mode: %d\n", err);
2372 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2374 struct kvm_vcpu *vcpu;
2377 mpidr &= MPIDR_HWID_BITMASK;
2378 kvm_for_each_vcpu(i, vcpu, kvm) {
2379 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2385 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2387 return irqchip_in_kernel(kvm);
2390 bool kvm_arch_has_irq_bypass(void)
2395 int kvm_arch_irq_bypass_add_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 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2404 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2405 struct irq_bypass_producer *prod)
2407 struct kvm_kernel_irqfd *irqfd =
2408 container_of(cons, struct kvm_kernel_irqfd, consumer);
2410 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2414 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2416 struct kvm_kernel_irqfd *irqfd =
2417 container_of(cons, struct kvm_kernel_irqfd, consumer);
2419 kvm_arm_halt_guest(irqfd->kvm);
2422 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2424 struct kvm_kernel_irqfd *irqfd =
2425 container_of(cons, struct kvm_kernel_irqfd, consumer);
2427 kvm_arm_resume_guest(irqfd->kvm);
2430 /* Initialize Hyp-mode and memory mappings on all CPUs */
2431 static __init int kvm_arm_init(void)
2436 if (!is_hyp_mode_available()) {
2437 kvm_info("HYP mode not available\n");
2441 if (kvm_get_mode() == KVM_MODE_NONE) {
2442 kvm_info("KVM disabled from command line\n");
2446 err = kvm_sys_reg_table_init();
2448 kvm_info("Error initializing system register tables");
2452 in_hyp_mode = is_kernel_in_hyp_mode();
2454 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2455 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2456 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2457 "Only trusted guests should be used on this system.\n");
2459 err = kvm_set_ipa_limit();
2463 err = kvm_arm_init_sve();
2467 err = kvm_arm_vmid_alloc_init();
2469 kvm_err("Failed to initialize VMID allocator.\n");
2474 err = init_hyp_mode();
2479 err = kvm_init_vector_slots();
2481 kvm_err("Cannot initialise vector slots\n");
2485 err = init_subsystems();
2489 if (is_protected_kvm_enabled()) {
2490 kvm_info("Protected nVHE mode initialized successfully\n");
2491 } else if (in_hyp_mode) {
2492 kvm_info("VHE mode initialized successfully\n");
2494 kvm_info("Hyp mode initialized successfully\n");
2498 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2499 * hypervisor protection is finalized.
2501 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2505 kvm_arm_initialised = true;
2510 teardown_subsystems();
2513 teardown_hyp_mode();
2515 kvm_arm_vmid_alloc_free();
2519 static int __init early_kvm_mode_cfg(char *arg)
2524 if (strcmp(arg, "none") == 0) {
2525 kvm_mode = KVM_MODE_NONE;
2529 if (!is_hyp_mode_available()) {
2530 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2534 if (strcmp(arg, "protected") == 0) {
2535 if (!is_kernel_in_hyp_mode())
2536 kvm_mode = KVM_MODE_PROTECTED;
2538 pr_warn_once("Protected KVM not available with VHE\n");
2543 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2544 kvm_mode = KVM_MODE_DEFAULT;
2548 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2549 kvm_mode = KVM_MODE_NV;
2555 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2557 enum kvm_mode kvm_get_mode(void)
2562 module_init(kvm_arm_init);