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 kfree(kvm->arch.mpidr_data);
209 kvm_destroy_vcpus(kvm);
211 kvm_unshare_hyp(kvm, kvm + 1);
213 kvm_arm_teardown_hypercalls(kvm);
216 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
220 case KVM_CAP_IRQCHIP:
223 case KVM_CAP_IOEVENTFD:
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_final_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_final_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();
320 case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
330 long kvm_arch_dev_ioctl(struct file *filp,
331 unsigned int ioctl, unsigned long arg)
336 struct kvm *kvm_arch_alloc_vm(void)
338 size_t sz = sizeof(struct kvm);
341 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
343 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
346 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
348 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
351 if (id >= kvm->max_vcpus)
357 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
361 spin_lock_init(&vcpu->arch.mp_state_lock);
363 #ifdef CONFIG_LOCKDEP
364 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
365 mutex_lock(&vcpu->mutex);
366 mutex_lock(&vcpu->kvm->arch.config_lock);
367 mutex_unlock(&vcpu->kvm->arch.config_lock);
368 mutex_unlock(&vcpu->mutex);
371 /* Force users to call KVM_ARM_VCPU_INIT */
372 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
374 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
377 * Default value for the FP state, will be overloaded at load
378 * time if we support FP (pretty likely)
380 vcpu->arch.fp_state = FP_STATE_FREE;
382 /* Set up the timer */
383 kvm_timer_vcpu_init(vcpu);
385 kvm_pmu_vcpu_init(vcpu);
387 kvm_arm_reset_debug_ptr(vcpu);
389 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
391 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
393 err = kvm_vgic_vcpu_init(vcpu);
397 return kvm_share_hyp(vcpu, vcpu + 1);
400 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
404 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
406 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
407 static_branch_dec(&userspace_irqchip_in_use);
409 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
410 kvm_timer_vcpu_terminate(vcpu);
411 kvm_pmu_vcpu_destroy(vcpu);
412 kvm_vgic_vcpu_destroy(vcpu);
413 kvm_arm_vcpu_destroy(vcpu);
416 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
421 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
426 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
428 struct kvm_s2_mmu *mmu;
431 mmu = vcpu->arch.hw_mmu;
432 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
435 * We guarantee that both TLBs and I-cache are private to each
436 * vcpu. If detecting that a vcpu from the same VM has
437 * previously run on the same physical CPU, call into the
438 * hypervisor code to nuke the relevant contexts.
440 * We might get preempted before the vCPU actually runs, but
441 * over-invalidation doesn't affect correctness.
443 if (*last_ran != vcpu->vcpu_idx) {
444 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
445 *last_ran = vcpu->vcpu_idx;
451 kvm_timer_vcpu_load(vcpu);
453 kvm_vcpu_load_vhe(vcpu);
454 kvm_arch_vcpu_load_fp(vcpu);
455 kvm_vcpu_pmu_restore_guest(vcpu);
456 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
457 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
459 if (single_task_running())
460 vcpu_clear_wfx_traps(vcpu);
462 vcpu_set_wfx_traps(vcpu);
464 if (vcpu_has_ptrauth(vcpu))
465 vcpu_ptrauth_disable(vcpu);
466 kvm_arch_vcpu_load_debug_state_flags(vcpu);
468 if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
469 vcpu_set_on_unsupported_cpu(vcpu);
472 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
474 kvm_arch_vcpu_put_debug_state_flags(vcpu);
475 kvm_arch_vcpu_put_fp(vcpu);
477 kvm_vcpu_put_vhe(vcpu);
478 kvm_timer_vcpu_put(vcpu);
480 kvm_vcpu_pmu_restore_host(vcpu);
481 kvm_arm_vmid_clear_active();
483 vcpu_clear_on_unsupported_cpu(vcpu);
487 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
489 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
490 kvm_make_request(KVM_REQ_SLEEP, vcpu);
494 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
496 spin_lock(&vcpu->arch.mp_state_lock);
497 __kvm_arm_vcpu_power_off(vcpu);
498 spin_unlock(&vcpu->arch.mp_state_lock);
501 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
503 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
506 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
508 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
509 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
513 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
515 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
518 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
519 struct kvm_mp_state *mp_state)
521 *mp_state = READ_ONCE(vcpu->arch.mp_state);
526 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
527 struct kvm_mp_state *mp_state)
531 spin_lock(&vcpu->arch.mp_state_lock);
533 switch (mp_state->mp_state) {
534 case KVM_MP_STATE_RUNNABLE:
535 WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
537 case KVM_MP_STATE_STOPPED:
538 __kvm_arm_vcpu_power_off(vcpu);
540 case KVM_MP_STATE_SUSPENDED:
541 kvm_arm_vcpu_suspend(vcpu);
547 spin_unlock(&vcpu->arch.mp_state_lock);
553 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
554 * @v: The VCPU pointer
556 * If the guest CPU is not waiting for interrupts or an interrupt line is
557 * asserted, the CPU is by definition runnable.
559 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
561 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
562 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
563 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
566 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
568 return vcpu_mode_priv(vcpu);
571 #ifdef CONFIG_GUEST_PERF_EVENTS
572 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
574 return *vcpu_pc(vcpu);
578 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
580 return vcpu_get_flag(vcpu, VCPU_INITIALIZED);
583 static void kvm_init_mpidr_data(struct kvm *kvm)
585 struct kvm_mpidr_data *data = NULL;
586 unsigned long c, mask, nr_entries;
587 u64 aff_set = 0, aff_clr = ~0UL;
588 struct kvm_vcpu *vcpu;
590 mutex_lock(&kvm->arch.config_lock);
592 if (kvm->arch.mpidr_data || atomic_read(&kvm->online_vcpus) == 1)
595 kvm_for_each_vcpu(c, vcpu, kvm) {
596 u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
602 * A significant bit can be either 0 or 1, and will only appear in
603 * aff_set. Use aff_clr to weed out the useless stuff.
605 mask = aff_set ^ aff_clr;
606 nr_entries = BIT_ULL(hweight_long(mask));
609 * Don't let userspace fool us. If we need more than a single page
610 * to describe the compressed MPIDR array, just fall back to the
611 * iterative method. Single vcpu VMs do not need this either.
613 if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
614 data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
620 data->mpidr_mask = mask;
622 kvm_for_each_vcpu(c, vcpu, kvm) {
623 u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
624 u16 index = kvm_mpidr_index(data, aff);
626 data->cmpidr_to_idx[index] = c;
629 kvm->arch.mpidr_data = data;
631 mutex_unlock(&kvm->arch.config_lock);
635 * Handle both the initialisation that is being done when the vcpu is
636 * run for the first time, as well as the updates that must be
637 * performed each time we get a new thread dealing with this vcpu.
639 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
641 struct kvm *kvm = vcpu->kvm;
644 if (!kvm_vcpu_initialized(vcpu))
647 if (!kvm_arm_vcpu_is_finalized(vcpu))
650 ret = kvm_arch_vcpu_run_map_fp(vcpu);
654 if (likely(vcpu_has_run_once(vcpu)))
657 kvm_init_mpidr_data(kvm);
659 kvm_arm_vcpu_init_debug(vcpu);
661 if (likely(irqchip_in_kernel(kvm))) {
663 * Map the VGIC hardware resources before running a vcpu the
664 * first time on this VM.
666 ret = kvm_vgic_map_resources(kvm);
671 if (vcpu_has_nv(vcpu)) {
672 ret = kvm_init_nv_sysregs(vcpu->kvm);
677 ret = kvm_timer_enable(vcpu);
681 ret = kvm_arm_pmu_v3_enable(vcpu);
685 if (is_protected_kvm_enabled()) {
686 ret = pkvm_create_hyp_vm(kvm);
691 if (!irqchip_in_kernel(kvm)) {
693 * Tell the rest of the code that there are userspace irqchip
696 static_branch_inc(&userspace_irqchip_in_use);
700 * Initialize traps for protected VMs.
701 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
702 * the code is in place for first run initialization at EL2.
704 if (kvm_vm_is_protected(kvm))
705 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
707 mutex_lock(&kvm->arch.config_lock);
708 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
709 mutex_unlock(&kvm->arch.config_lock);
714 bool kvm_arch_intc_initialized(struct kvm *kvm)
716 return vgic_initialized(kvm);
719 void kvm_arm_halt_guest(struct kvm *kvm)
722 struct kvm_vcpu *vcpu;
724 kvm_for_each_vcpu(i, vcpu, kvm)
725 vcpu->arch.pause = true;
726 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
729 void kvm_arm_resume_guest(struct kvm *kvm)
732 struct kvm_vcpu *vcpu;
734 kvm_for_each_vcpu(i, vcpu, kvm) {
735 vcpu->arch.pause = false;
736 __kvm_vcpu_wake_up(vcpu);
740 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
742 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
744 rcuwait_wait_event(wait,
745 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
748 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
749 /* Awaken to handle a signal, request we sleep again later. */
750 kvm_make_request(KVM_REQ_SLEEP, vcpu);
754 * Make sure we will observe a potential reset request if we've
755 * observed a change to the power state. Pairs with the smp_wmb() in
756 * kvm_psci_vcpu_on().
762 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
763 * @vcpu: The VCPU pointer
765 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
766 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
767 * on when a wake event arrives, e.g. there may already be a pending wake event.
769 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
772 * Sync back the state of the GIC CPU interface so that we have
773 * the latest PMR and group enables. This ensures that
774 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
775 * we have pending interrupts, e.g. when determining if the
778 * For the same reason, we want to tell GICv4 that we need
779 * doorbells to be signalled, should an interrupt become pending.
782 kvm_vgic_vmcr_sync(vcpu);
783 vcpu_set_flag(vcpu, IN_WFI);
788 vcpu_clear_flag(vcpu, IN_WFIT);
791 vcpu_clear_flag(vcpu, IN_WFI);
796 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
798 if (!kvm_arm_vcpu_suspended(vcpu))
804 * The suspend state is sticky; we do not leave it until userspace
805 * explicitly marks the vCPU as runnable. Request that we suspend again
808 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
811 * Check to make sure the vCPU is actually runnable. If so, exit to
812 * userspace informing it of the wakeup condition.
814 if (kvm_arch_vcpu_runnable(vcpu)) {
815 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
816 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
817 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
822 * Otherwise, we were unblocked to process a different event, such as a
823 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
830 * check_vcpu_requests - check and handle pending vCPU requests
831 * @vcpu: the VCPU pointer
833 * Return: 1 if we should enter the guest
834 * 0 if we should exit to userspace
835 * < 0 if we should exit to userspace, where the return value indicates
838 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
840 if (kvm_request_pending(vcpu)) {
841 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
842 kvm_vcpu_sleep(vcpu);
844 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
845 kvm_reset_vcpu(vcpu);
848 * Clear IRQ_PENDING requests that were made to guarantee
849 * that a VCPU sees new virtual interrupts.
851 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
853 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
854 kvm_update_stolen_time(vcpu);
856 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
857 /* The distributor enable bits were changed */
864 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
865 kvm_vcpu_reload_pmu(vcpu);
867 if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
868 kvm_vcpu_pmu_restore_guest(vcpu);
870 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
871 return kvm_vcpu_suspend(vcpu);
873 if (kvm_dirty_ring_check_request(vcpu))
880 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
882 if (likely(!vcpu_mode_is_32bit(vcpu)))
885 if (vcpu_has_nv(vcpu))
888 return !kvm_supports_32bit_el0();
892 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
893 * @vcpu: The VCPU pointer
894 * @ret: Pointer to write optional return code
896 * Returns: true if the VCPU needs to return to a preemptible + interruptible
897 * and skip guest entry.
899 * This function disambiguates between two different types of exits: exits to a
900 * preemptible + interruptible kernel context and exits to userspace. For an
901 * exit to userspace, this function will write the return code to ret and return
902 * true. For an exit to preemptible + interruptible kernel context (i.e. check
903 * for pending work and re-enter), return true without writing to ret.
905 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
907 struct kvm_run *run = vcpu->run;
910 * If we're using a userspace irqchip, then check if we need
911 * to tell a userspace irqchip about timer or PMU level
912 * changes and if so, exit to userspace (the actual level
913 * state gets updated in kvm_timer_update_run and
914 * kvm_pmu_update_run below).
916 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
917 if (kvm_timer_should_notify_user(vcpu) ||
918 kvm_pmu_should_notify_user(vcpu)) {
920 run->exit_reason = KVM_EXIT_INTR;
925 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
926 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
927 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
928 run->fail_entry.cpu = smp_processor_id();
933 return kvm_request_pending(vcpu) ||
934 xfer_to_guest_mode_work_pending();
938 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
939 * the vCPU is running.
941 * This must be noinstr as instrumentation may make use of RCU, and this is not
942 * safe during the EQS.
944 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
948 guest_state_enter_irqoff();
949 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
950 guest_state_exit_irqoff();
956 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
957 * @vcpu: The VCPU pointer
959 * This function is called through the VCPU_RUN ioctl called from user space. It
960 * will execute VM code in a loop until the time slice for the process is used
961 * or some emulation is needed from user space in which case the function will
962 * return with return value 0 and with the kvm_run structure filled in with the
963 * required data for the requested emulation.
965 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
967 struct kvm_run *run = vcpu->run;
970 if (run->exit_reason == KVM_EXIT_MMIO) {
971 ret = kvm_handle_mmio_return(vcpu);
978 if (run->immediate_exit) {
983 kvm_sigset_activate(vcpu);
986 run->exit_reason = KVM_EXIT_UNKNOWN;
990 * Check conditions before entering the guest
992 ret = xfer_to_guest_mode_handle_work(vcpu);
997 ret = check_vcpu_requests(vcpu);
1000 * Preparing the interrupts to be injected also
1001 * involves poking the GIC, which must be done in a
1002 * non-preemptible context.
1007 * The VMID allocator only tracks active VMIDs per
1008 * physical CPU, and therefore the VMID allocated may not be
1009 * preserved on VMID roll-over if the task was preempted,
1010 * making a thread's VMID inactive. So we need to call
1011 * kvm_arm_vmid_update() in non-premptible context.
1013 if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1015 __load_stage2(vcpu->arch.hw_mmu,
1016 vcpu->arch.hw_mmu->arch);
1018 kvm_pmu_flush_hwstate(vcpu);
1020 local_irq_disable();
1022 kvm_vgic_flush_hwstate(vcpu);
1024 kvm_pmu_update_vcpu_events(vcpu);
1027 * Ensure we set mode to IN_GUEST_MODE after we disable
1028 * interrupts and before the final VCPU requests check.
1029 * See the comment in kvm_vcpu_exiting_guest_mode() and
1030 * Documentation/virt/kvm/vcpu-requests.rst
1032 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1034 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1035 vcpu->mode = OUTSIDE_GUEST_MODE;
1036 isb(); /* Ensure work in x_flush_hwstate is committed */
1037 kvm_pmu_sync_hwstate(vcpu);
1038 if (static_branch_unlikely(&userspace_irqchip_in_use))
1039 kvm_timer_sync_user(vcpu);
1040 kvm_vgic_sync_hwstate(vcpu);
1046 kvm_arm_setup_debug(vcpu);
1047 kvm_arch_vcpu_ctxflush_fp(vcpu);
1049 /**************************************************************
1052 trace_kvm_entry(*vcpu_pc(vcpu));
1053 guest_timing_enter_irqoff();
1055 ret = kvm_arm_vcpu_enter_exit(vcpu);
1057 vcpu->mode = OUTSIDE_GUEST_MODE;
1061 *************************************************************/
1063 kvm_arm_clear_debug(vcpu);
1066 * We must sync the PMU state before the vgic state so
1067 * that the vgic can properly sample the updated state of the
1070 kvm_pmu_sync_hwstate(vcpu);
1073 * Sync the vgic state before syncing the timer state because
1074 * the timer code needs to know if the virtual timer
1075 * interrupts are active.
1077 kvm_vgic_sync_hwstate(vcpu);
1080 * Sync the timer hardware state before enabling interrupts as
1081 * we don't want vtimer interrupts to race with syncing the
1082 * timer virtual interrupt state.
1084 if (static_branch_unlikely(&userspace_irqchip_in_use))
1085 kvm_timer_sync_user(vcpu);
1087 kvm_arch_vcpu_ctxsync_fp(vcpu);
1090 * We must ensure that any pending interrupts are taken before
1091 * we exit guest timing so that timer ticks are accounted as
1092 * guest time. Transiently unmask interrupts so that any
1093 * pending interrupts are taken.
1095 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1096 * context synchronization event) is necessary to ensure that
1097 * pending interrupts are taken.
1099 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1102 local_irq_disable();
1105 guest_timing_exit_irqoff();
1109 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1111 /* Exit types that need handling before we can be preempted */
1112 handle_exit_early(vcpu, ret);
1117 * The ARMv8 architecture doesn't give the hypervisor
1118 * a mechanism to prevent a guest from dropping to AArch32 EL0
1119 * if implemented by the CPU. If we spot the guest in such
1120 * state and that we decided it wasn't supposed to do so (like
1121 * with the asymmetric AArch32 case), return to userspace with
1124 if (vcpu_mode_is_bad_32bit(vcpu)) {
1126 * As we have caught the guest red-handed, decide that
1127 * it isn't fit for purpose anymore by making the vcpu
1128 * invalid. The VMM can try and fix it by issuing a
1129 * KVM_ARM_VCPU_INIT if it really wants to.
1131 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1132 ret = ARM_EXCEPTION_IL;
1135 ret = handle_exit(vcpu, ret);
1138 /* Tell userspace about in-kernel device output levels */
1139 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1140 kvm_timer_update_run(vcpu);
1141 kvm_pmu_update_run(vcpu);
1144 kvm_sigset_deactivate(vcpu);
1148 * In the unlikely event that we are returning to userspace
1149 * with pending exceptions or PC adjustment, commit these
1150 * adjustments in order to give userspace a consistent view of
1151 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1152 * being preempt-safe on VHE.
1154 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1155 vcpu_get_flag(vcpu, INCREMENT_PC)))
1156 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1162 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1168 if (number == KVM_ARM_IRQ_CPU_IRQ)
1169 bit_index = __ffs(HCR_VI);
1170 else /* KVM_ARM_IRQ_CPU_FIQ */
1171 bit_index = __ffs(HCR_VF);
1173 hcr = vcpu_hcr(vcpu);
1175 set = test_and_set_bit(bit_index, hcr);
1177 set = test_and_clear_bit(bit_index, hcr);
1180 * If we didn't change anything, no need to wake up or kick other CPUs
1186 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1187 * trigger a world-switch round on the running physical CPU to set the
1188 * virtual IRQ/FIQ fields in the HCR appropriately.
1190 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1191 kvm_vcpu_kick(vcpu);
1196 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1199 u32 irq = irq_level->irq;
1200 unsigned int irq_type, vcpu_id, irq_num;
1201 struct kvm_vcpu *vcpu = NULL;
1202 bool level = irq_level->level;
1204 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1205 vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1206 vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1207 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1209 trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1212 case KVM_ARM_IRQ_TYPE_CPU:
1213 if (irqchip_in_kernel(kvm))
1216 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1220 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1223 return vcpu_interrupt_line(vcpu, irq_num, level);
1224 case KVM_ARM_IRQ_TYPE_PPI:
1225 if (!irqchip_in_kernel(kvm))
1228 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1232 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1235 return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1236 case KVM_ARM_IRQ_TYPE_SPI:
1237 if (!irqchip_in_kernel(kvm))
1240 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1243 return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1249 static unsigned long system_supported_vcpu_features(void)
1251 unsigned long features = KVM_VCPU_VALID_FEATURES;
1253 if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1254 clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1256 if (!kvm_arm_support_pmu_v3())
1257 clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1259 if (!system_supports_sve())
1260 clear_bit(KVM_ARM_VCPU_SVE, &features);
1262 if (!system_has_full_ptr_auth()) {
1263 clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1264 clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1267 if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1268 clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1273 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1274 const struct kvm_vcpu_init *init)
1276 unsigned long features = init->features[0];
1279 if (features & ~KVM_VCPU_VALID_FEATURES)
1282 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1283 if (init->features[i])
1287 if (features & ~system_supported_vcpu_features())
1291 * For now make sure that both address/generic pointer authentication
1292 * features are requested by the userspace together.
1294 if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1295 test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1298 /* Disallow NV+SVE for the time being */
1299 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features) &&
1300 test_bit(KVM_ARM_VCPU_SVE, &features))
1303 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1306 /* MTE is incompatible with AArch32 */
1307 if (kvm_has_mte(vcpu->kvm))
1310 /* NV is incompatible with AArch32 */
1311 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1317 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1318 const struct kvm_vcpu_init *init)
1320 unsigned long features = init->features[0];
1322 return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1323 KVM_VCPU_MAX_FEATURES);
1326 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1328 struct kvm *kvm = vcpu->kvm;
1332 * When the vCPU has a PMU, but no PMU is set for the guest
1333 * yet, set the default one.
1335 if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1336 ret = kvm_arm_set_default_pmu(kvm);
1341 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1342 const struct kvm_vcpu_init *init)
1344 unsigned long features = init->features[0];
1345 struct kvm *kvm = vcpu->kvm;
1348 mutex_lock(&kvm->arch.config_lock);
1350 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1351 kvm_vcpu_init_changed(vcpu, init))
1354 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1356 ret = kvm_setup_vcpu(vcpu);
1360 /* Now we know what it is, we can reset it. */
1361 kvm_reset_vcpu(vcpu);
1363 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1364 vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1367 mutex_unlock(&kvm->arch.config_lock);
1371 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1372 const struct kvm_vcpu_init *init)
1376 if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1377 init->target != kvm_target_cpu())
1380 ret = kvm_vcpu_init_check_features(vcpu, init);
1384 if (!kvm_vcpu_initialized(vcpu))
1385 return __kvm_vcpu_set_target(vcpu, init);
1387 if (kvm_vcpu_init_changed(vcpu, init))
1390 kvm_reset_vcpu(vcpu);
1394 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1395 struct kvm_vcpu_init *init)
1397 bool power_off = false;
1401 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1402 * reflecting it in the finalized feature set, thus limiting its scope
1403 * to a single KVM_ARM_VCPU_INIT call.
1405 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1406 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1410 ret = kvm_vcpu_set_target(vcpu, init);
1415 * Ensure a rebooted VM will fault in RAM pages and detect if the
1416 * guest MMU is turned off and flush the caches as needed.
1418 * S2FWB enforces all memory accesses to RAM being cacheable,
1419 * ensuring that the data side is always coherent. We still
1420 * need to invalidate the I-cache though, as FWB does *not*
1421 * imply CTR_EL0.DIC.
1423 if (vcpu_has_run_once(vcpu)) {
1424 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1425 stage2_unmap_vm(vcpu->kvm);
1427 icache_inval_all_pou();
1430 vcpu_reset_hcr(vcpu);
1431 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1434 * Handle the "start in power-off" case.
1436 spin_lock(&vcpu->arch.mp_state_lock);
1439 __kvm_arm_vcpu_power_off(vcpu);
1441 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1443 spin_unlock(&vcpu->arch.mp_state_lock);
1448 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1449 struct kvm_device_attr *attr)
1453 switch (attr->group) {
1455 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1462 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1463 struct kvm_device_attr *attr)
1467 switch (attr->group) {
1469 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1476 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1477 struct kvm_device_attr *attr)
1481 switch (attr->group) {
1483 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1490 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1491 struct kvm_vcpu_events *events)
1493 memset(events, 0, sizeof(*events));
1495 return __kvm_arm_vcpu_get_events(vcpu, events);
1498 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1499 struct kvm_vcpu_events *events)
1503 /* check whether the reserved field is zero */
1504 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1505 if (events->reserved[i])
1508 /* check whether the pad field is zero */
1509 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1510 if (events->exception.pad[i])
1513 return __kvm_arm_vcpu_set_events(vcpu, events);
1516 long kvm_arch_vcpu_ioctl(struct file *filp,
1517 unsigned int ioctl, unsigned long arg)
1519 struct kvm_vcpu *vcpu = filp->private_data;
1520 void __user *argp = (void __user *)arg;
1521 struct kvm_device_attr attr;
1525 case KVM_ARM_VCPU_INIT: {
1526 struct kvm_vcpu_init init;
1529 if (copy_from_user(&init, argp, sizeof(init)))
1532 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1535 case KVM_SET_ONE_REG:
1536 case KVM_GET_ONE_REG: {
1537 struct kvm_one_reg reg;
1540 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1544 if (copy_from_user(®, argp, sizeof(reg)))
1548 * We could owe a reset due to PSCI. Handle the pending reset
1549 * here to ensure userspace register accesses are ordered after
1552 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1553 kvm_reset_vcpu(vcpu);
1555 if (ioctl == KVM_SET_ONE_REG)
1556 r = kvm_arm_set_reg(vcpu, ®);
1558 r = kvm_arm_get_reg(vcpu, ®);
1561 case KVM_GET_REG_LIST: {
1562 struct kvm_reg_list __user *user_list = argp;
1563 struct kvm_reg_list reg_list;
1567 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1571 if (!kvm_arm_vcpu_is_finalized(vcpu))
1575 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1578 reg_list.n = kvm_arm_num_regs(vcpu);
1579 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1584 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1587 case KVM_SET_DEVICE_ATTR: {
1589 if (copy_from_user(&attr, argp, sizeof(attr)))
1591 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1594 case KVM_GET_DEVICE_ATTR: {
1596 if (copy_from_user(&attr, argp, sizeof(attr)))
1598 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1601 case KVM_HAS_DEVICE_ATTR: {
1603 if (copy_from_user(&attr, argp, sizeof(attr)))
1605 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1608 case KVM_GET_VCPU_EVENTS: {
1609 struct kvm_vcpu_events events;
1611 if (kvm_arm_vcpu_get_events(vcpu, &events))
1614 if (copy_to_user(argp, &events, sizeof(events)))
1619 case KVM_SET_VCPU_EVENTS: {
1620 struct kvm_vcpu_events events;
1622 if (copy_from_user(&events, argp, sizeof(events)))
1625 return kvm_arm_vcpu_set_events(vcpu, &events);
1627 case KVM_ARM_VCPU_FINALIZE: {
1630 if (!kvm_vcpu_initialized(vcpu))
1633 if (get_user(what, (const int __user *)argp))
1636 return kvm_arm_vcpu_finalize(vcpu, what);
1645 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1650 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1651 struct kvm_arm_device_addr *dev_addr)
1653 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1654 case KVM_ARM_DEVICE_VGIC_V2:
1657 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1663 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1665 switch (attr->group) {
1666 case KVM_ARM_VM_SMCCC_CTRL:
1667 return kvm_vm_smccc_has_attr(kvm, attr);
1673 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1675 switch (attr->group) {
1676 case KVM_ARM_VM_SMCCC_CTRL:
1677 return kvm_vm_smccc_set_attr(kvm, attr);
1683 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1685 struct kvm *kvm = filp->private_data;
1686 void __user *argp = (void __user *)arg;
1687 struct kvm_device_attr attr;
1690 case KVM_CREATE_IRQCHIP: {
1694 mutex_lock(&kvm->lock);
1695 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1696 mutex_unlock(&kvm->lock);
1699 case KVM_ARM_SET_DEVICE_ADDR: {
1700 struct kvm_arm_device_addr dev_addr;
1702 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1704 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1706 case KVM_ARM_PREFERRED_TARGET: {
1707 struct kvm_vcpu_init init = {
1708 .target = KVM_ARM_TARGET_GENERIC_V8,
1711 if (copy_to_user(argp, &init, sizeof(init)))
1716 case KVM_ARM_MTE_COPY_TAGS: {
1717 struct kvm_arm_copy_mte_tags copy_tags;
1719 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1721 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1723 case KVM_ARM_SET_COUNTER_OFFSET: {
1724 struct kvm_arm_counter_offset offset;
1726 if (copy_from_user(&offset, argp, sizeof(offset)))
1728 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1730 case KVM_HAS_DEVICE_ATTR: {
1731 if (copy_from_user(&attr, argp, sizeof(attr)))
1734 return kvm_vm_has_attr(kvm, &attr);
1736 case KVM_SET_DEVICE_ATTR: {
1737 if (copy_from_user(&attr, argp, sizeof(attr)))
1740 return kvm_vm_set_attr(kvm, &attr);
1742 case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1743 struct reg_mask_range range;
1745 if (copy_from_user(&range, argp, sizeof(range)))
1747 return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1754 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1755 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1757 struct kvm_vcpu *tmp_vcpu;
1759 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1760 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1761 mutex_unlock(&tmp_vcpu->mutex);
1765 void unlock_all_vcpus(struct kvm *kvm)
1767 lockdep_assert_held(&kvm->lock);
1769 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1772 /* Returns true if all vcpus were locked, false otherwise */
1773 bool lock_all_vcpus(struct kvm *kvm)
1775 struct kvm_vcpu *tmp_vcpu;
1778 lockdep_assert_held(&kvm->lock);
1781 * Any time a vcpu is in an ioctl (including running), the
1782 * core KVM code tries to grab the vcpu->mutex.
1784 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1785 * other VCPUs can fiddle with the state while we access it.
1787 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1788 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1789 unlock_vcpus(kvm, c - 1);
1797 static unsigned long nvhe_percpu_size(void)
1799 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1800 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1803 static unsigned long nvhe_percpu_order(void)
1805 unsigned long size = nvhe_percpu_size();
1807 return size ? get_order(size) : 0;
1810 /* A lookup table holding the hypervisor VA for each vector slot */
1811 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1813 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1815 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1818 static int kvm_init_vector_slots(void)
1823 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1824 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1826 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1827 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1829 if (kvm_system_needs_idmapped_vectors() &&
1830 !is_protected_kvm_enabled()) {
1831 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1832 __BP_HARDEN_HYP_VECS_SZ, &base);
1837 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1838 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1842 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1844 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1845 u64 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1849 * Calculate the raw per-cpu offset without a translation from the
1850 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1851 * so that we can use adr_l to access per-cpu variables in EL2.
1852 * Also drop the KASAN tag which gets in the way...
1854 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1855 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1857 params->mair_el2 = read_sysreg(mair_el1);
1859 tcr = read_sysreg(tcr_el1);
1860 if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1861 tcr |= TCR_EPD1_MASK;
1863 tcr &= TCR_EL2_MASK;
1864 tcr |= TCR_EL2_RES1;
1866 tcr &= ~TCR_T0SZ_MASK;
1867 tcr |= TCR_T0SZ(hyp_va_bits);
1868 tcr &= ~TCR_EL2_PS_MASK;
1869 tcr |= FIELD_PREP(TCR_EL2_PS_MASK, kvm_get_parange(mmfr0));
1870 if (kvm_lpa2_is_enabled())
1872 params->tcr_el2 = tcr;
1874 params->pgd_pa = kvm_mmu_get_httbr();
1875 if (is_protected_kvm_enabled())
1876 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1878 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1879 if (cpus_have_final_cap(ARM64_KVM_HVHE))
1880 params->hcr_el2 |= HCR_E2H;
1881 params->vttbr = params->vtcr = 0;
1884 * Flush the init params from the data cache because the struct will
1885 * be read while the MMU is off.
1887 kvm_flush_dcache_to_poc(params, sizeof(*params));
1890 static void hyp_install_host_vector(void)
1892 struct kvm_nvhe_init_params *params;
1893 struct arm_smccc_res res;
1895 /* Switch from the HYP stub to our own HYP init vector */
1896 __hyp_set_vectors(kvm_get_idmap_vector());
1899 * Call initialization code, and switch to the full blown HYP code.
1900 * If the cpucaps haven't been finalized yet, something has gone very
1901 * wrong, and hyp will crash and burn when it uses any
1902 * cpus_have_*_cap() wrapper.
1904 BUG_ON(!system_capabilities_finalized());
1905 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1906 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1907 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1910 static void cpu_init_hyp_mode(void)
1912 hyp_install_host_vector();
1915 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1918 if (this_cpu_has_cap(ARM64_SSBS) &&
1919 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1920 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1924 static void cpu_hyp_reset(void)
1926 if (!is_kernel_in_hyp_mode())
1927 __hyp_reset_vectors();
1931 * EL2 vectors can be mapped and rerouted in a number of ways,
1932 * depending on the kernel configuration and CPU present:
1934 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1935 * placed in one of the vector slots, which is executed before jumping
1936 * to the real vectors.
1938 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1939 * containing the hardening sequence is mapped next to the idmap page,
1940 * and executed before jumping to the real vectors.
1942 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1943 * empty slot is selected, mapped next to the idmap page, and
1944 * executed before jumping to the real vectors.
1946 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1947 * VHE, as we don't have hypervisor-specific mappings. If the system
1948 * is VHE and yet selects this capability, it will be ignored.
1950 static void cpu_set_hyp_vector(void)
1952 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1953 void *vector = hyp_spectre_vector_selector[data->slot];
1955 if (!is_protected_kvm_enabled())
1956 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1958 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1961 static void cpu_hyp_init_context(void)
1963 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1965 if (!is_kernel_in_hyp_mode())
1966 cpu_init_hyp_mode();
1969 static void cpu_hyp_init_features(void)
1971 cpu_set_hyp_vector();
1972 kvm_arm_init_debug();
1974 if (is_kernel_in_hyp_mode())
1975 kvm_timer_init_vhe();
1978 kvm_vgic_init_cpu_hardware();
1981 static void cpu_hyp_reinit(void)
1984 cpu_hyp_init_context();
1985 cpu_hyp_init_features();
1988 static void cpu_hyp_init(void *discard)
1990 if (!__this_cpu_read(kvm_hyp_initialized)) {
1992 __this_cpu_write(kvm_hyp_initialized, 1);
1996 static void cpu_hyp_uninit(void *discard)
1998 if (__this_cpu_read(kvm_hyp_initialized)) {
2000 __this_cpu_write(kvm_hyp_initialized, 0);
2004 int kvm_arch_hardware_enable(void)
2007 * Most calls to this function are made with migration
2008 * disabled, but not with preemption disabled. The former is
2009 * enough to ensure correctness, but most of the helpers
2010 * expect the later and will throw a tantrum otherwise.
2024 void kvm_arch_hardware_disable(void)
2026 kvm_timer_cpu_down();
2027 kvm_vgic_cpu_down();
2029 if (!is_protected_kvm_enabled())
2030 cpu_hyp_uninit(NULL);
2033 #ifdef CONFIG_CPU_PM
2034 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2039 * kvm_hyp_initialized is left with its old value over
2040 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2045 if (__this_cpu_read(kvm_hyp_initialized))
2047 * don't update kvm_hyp_initialized here
2048 * so that the hyp will be re-enabled
2049 * when we resume. See below.
2054 case CPU_PM_ENTER_FAILED:
2056 if (__this_cpu_read(kvm_hyp_initialized))
2057 /* The hyp was enabled before suspend. */
2067 static struct notifier_block hyp_init_cpu_pm_nb = {
2068 .notifier_call = hyp_init_cpu_pm_notifier,
2071 static void __init hyp_cpu_pm_init(void)
2073 if (!is_protected_kvm_enabled())
2074 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2076 static void __init hyp_cpu_pm_exit(void)
2078 if (!is_protected_kvm_enabled())
2079 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2082 static inline void __init hyp_cpu_pm_init(void)
2085 static inline void __init hyp_cpu_pm_exit(void)
2090 static void __init init_cpu_logical_map(void)
2095 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2096 * Only copy the set of online CPUs whose features have been checked
2097 * against the finalized system capabilities. The hypervisor will not
2098 * allow any other CPUs from the `possible` set to boot.
2100 for_each_online_cpu(cpu)
2101 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2104 #define init_psci_0_1_impl_state(config, what) \
2105 config.psci_0_1_ ## what ## _implemented = psci_ops.what
2107 static bool __init init_psci_relay(void)
2110 * If PSCI has not been initialized, protected KVM cannot install
2111 * itself on newly booted CPUs.
2113 if (!psci_ops.get_version) {
2114 kvm_err("Cannot initialize protected mode without PSCI\n");
2118 kvm_host_psci_config.version = psci_ops.get_version();
2119 kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2121 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2122 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2123 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2124 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2125 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2126 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2131 static int __init init_subsystems(void)
2136 * Enable hardware so that subsystem initialisation can access EL2.
2138 on_each_cpu(cpu_hyp_init, NULL, 1);
2141 * Register CPU lower-power notifier
2146 * Init HYP view of VGIC
2148 err = kvm_vgic_hyp_init();
2151 vgic_present = true;
2155 vgic_present = false;
2163 * Init HYP architected timer support
2165 err = kvm_timer_hyp_init(vgic_present);
2169 kvm_register_perf_callbacks(NULL);
2175 if (err || !is_protected_kvm_enabled())
2176 on_each_cpu(cpu_hyp_uninit, NULL, 1);
2181 static void __init teardown_subsystems(void)
2183 kvm_unregister_perf_callbacks();
2187 static void __init teardown_hyp_mode(void)
2192 for_each_possible_cpu(cpu) {
2193 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2194 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2198 static int __init do_pkvm_init(u32 hyp_va_bits)
2200 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2204 cpu_hyp_init_context();
2205 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2206 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2208 cpu_hyp_init_features();
2211 * The stub hypercalls are now disabled, so set our local flag to
2212 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2214 __this_cpu_write(kvm_hyp_initialized, 1);
2220 static u64 get_hyp_id_aa64pfr0_el1(void)
2223 * Track whether the system isn't affected by spectre/meltdown in the
2224 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2225 * Although this is per-CPU, we make it global for simplicity, e.g., not
2226 * to have to worry about vcpu migration.
2228 * Unlike for non-protected VMs, userspace cannot override this for
2231 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2233 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2234 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2236 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2237 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2238 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2239 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2244 static void kvm_hyp_init_symbols(void)
2246 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2247 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2248 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2249 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2250 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2251 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2252 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2253 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2254 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2255 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2256 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2259 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2261 void *addr = phys_to_virt(hyp_mem_base);
2264 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2268 ret = do_pkvm_init(hyp_va_bits);
2277 static void pkvm_hyp_init_ptrauth(void)
2279 struct kvm_cpu_context *hyp_ctxt;
2282 for_each_possible_cpu(cpu) {
2283 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2284 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2285 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2286 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2287 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2288 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2289 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2290 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2291 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2292 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2293 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2297 /* Inits Hyp-mode on all online CPUs */
2298 static int __init init_hyp_mode(void)
2305 * The protected Hyp-mode cannot be initialized if the memory pool
2306 * allocation has failed.
2308 if (is_protected_kvm_enabled() && !hyp_mem_base)
2312 * Allocate Hyp PGD and setup Hyp identity mapping
2314 err = kvm_mmu_init(&hyp_va_bits);
2319 * Allocate stack pages for Hypervisor-mode
2321 for_each_possible_cpu(cpu) {
2322 unsigned long stack_page;
2324 stack_page = __get_free_page(GFP_KERNEL);
2330 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2334 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2336 for_each_possible_cpu(cpu) {
2340 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2346 page_addr = page_address(page);
2347 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2348 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2352 * Map the Hyp-code called directly from the host
2354 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2355 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2357 kvm_err("Cannot map world-switch code\n");
2361 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2362 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2364 kvm_err("Cannot map .hyp.rodata section\n");
2368 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2369 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2371 kvm_err("Cannot map rodata section\n");
2376 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2377 * section thanks to an assertion in the linker script. Map it RW and
2378 * the rest of .bss RO.
2380 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2381 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2383 kvm_err("Cannot map hyp bss section: %d\n", err);
2387 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2388 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2390 kvm_err("Cannot map bss section\n");
2395 * Map the Hyp stack pages
2397 for_each_possible_cpu(cpu) {
2398 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2399 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2401 err = create_hyp_stack(__pa(stack_page), ¶ms->stack_hyp_va);
2403 kvm_err("Cannot map hyp stack\n");
2408 * Save the stack PA in nvhe_init_params. This will be needed
2409 * to recreate the stack mapping in protected nVHE mode.
2410 * __hyp_pa() won't do the right thing there, since the stack
2411 * has been mapped in the flexible private VA space.
2413 params->stack_pa = __pa(stack_page);
2416 for_each_possible_cpu(cpu) {
2417 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2418 char *percpu_end = percpu_begin + nvhe_percpu_size();
2420 /* Map Hyp percpu pages */
2421 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2423 kvm_err("Cannot map hyp percpu region\n");
2427 /* Prepare the CPU initialization parameters */
2428 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2431 kvm_hyp_init_symbols();
2433 if (is_protected_kvm_enabled()) {
2434 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2435 cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2436 pkvm_hyp_init_ptrauth();
2438 init_cpu_logical_map();
2440 if (!init_psci_relay()) {
2445 err = kvm_hyp_init_protection(hyp_va_bits);
2447 kvm_err("Failed to init hyp memory protection\n");
2455 teardown_hyp_mode();
2456 kvm_err("error initializing Hyp mode: %d\n", err);
2460 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2462 struct kvm_vcpu *vcpu;
2465 mpidr &= MPIDR_HWID_BITMASK;
2467 if (kvm->arch.mpidr_data) {
2468 u16 idx = kvm_mpidr_index(kvm->arch.mpidr_data, mpidr);
2470 vcpu = kvm_get_vcpu(kvm,
2471 kvm->arch.mpidr_data->cmpidr_to_idx[idx]);
2472 if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2478 kvm_for_each_vcpu(i, vcpu, kvm) {
2479 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2485 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2487 return irqchip_in_kernel(kvm);
2490 bool kvm_arch_has_irq_bypass(void)
2495 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2496 struct irq_bypass_producer *prod)
2498 struct kvm_kernel_irqfd *irqfd =
2499 container_of(cons, struct kvm_kernel_irqfd, consumer);
2501 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2504 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2505 struct irq_bypass_producer *prod)
2507 struct kvm_kernel_irqfd *irqfd =
2508 container_of(cons, struct kvm_kernel_irqfd, consumer);
2510 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2514 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2516 struct kvm_kernel_irqfd *irqfd =
2517 container_of(cons, struct kvm_kernel_irqfd, consumer);
2519 kvm_arm_halt_guest(irqfd->kvm);
2522 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2524 struct kvm_kernel_irqfd *irqfd =
2525 container_of(cons, struct kvm_kernel_irqfd, consumer);
2527 kvm_arm_resume_guest(irqfd->kvm);
2530 /* Initialize Hyp-mode and memory mappings on all CPUs */
2531 static __init int kvm_arm_init(void)
2536 if (!is_hyp_mode_available()) {
2537 kvm_info("HYP mode not available\n");
2541 if (kvm_get_mode() == KVM_MODE_NONE) {
2542 kvm_info("KVM disabled from command line\n");
2546 err = kvm_sys_reg_table_init();
2548 kvm_info("Error initializing system register tables");
2552 in_hyp_mode = is_kernel_in_hyp_mode();
2554 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2555 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2556 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2557 "Only trusted guests should be used on this system.\n");
2559 err = kvm_set_ipa_limit();
2563 err = kvm_arm_init_sve();
2567 err = kvm_arm_vmid_alloc_init();
2569 kvm_err("Failed to initialize VMID allocator.\n");
2574 err = init_hyp_mode();
2579 err = kvm_init_vector_slots();
2581 kvm_err("Cannot initialise vector slots\n");
2585 err = init_subsystems();
2589 if (is_protected_kvm_enabled()) {
2590 kvm_info("Protected nVHE mode initialized successfully\n");
2591 } else if (in_hyp_mode) {
2592 kvm_info("VHE mode initialized successfully\n");
2594 kvm_info("Hyp mode initialized successfully\n");
2598 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2599 * hypervisor protection is finalized.
2601 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2605 kvm_arm_initialised = true;
2610 teardown_subsystems();
2613 teardown_hyp_mode();
2615 kvm_arm_vmid_alloc_free();
2619 static int __init early_kvm_mode_cfg(char *arg)
2624 if (strcmp(arg, "none") == 0) {
2625 kvm_mode = KVM_MODE_NONE;
2629 if (!is_hyp_mode_available()) {
2630 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2634 if (strcmp(arg, "protected") == 0) {
2635 if (!is_kernel_in_hyp_mode())
2636 kvm_mode = KVM_MODE_PROTECTED;
2638 pr_warn_once("Protected KVM not available with VHE\n");
2643 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2644 kvm_mode = KVM_MODE_DEFAULT;
2648 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2649 kvm_mode = KVM_MODE_NV;
2655 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2657 enum kvm_mode kvm_get_mode(void)
2662 module_init(kvm_arm_init);