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_DEVICE_CTRL:
225 case KVM_CAP_USER_MEMORY:
226 case KVM_CAP_SYNC_MMU:
227 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
228 case KVM_CAP_ONE_REG:
229 case KVM_CAP_ARM_PSCI:
230 case KVM_CAP_ARM_PSCI_0_2:
231 case KVM_CAP_READONLY_MEM:
232 case KVM_CAP_MP_STATE:
233 case KVM_CAP_IMMEDIATE_EXIT:
234 case KVM_CAP_VCPU_EVENTS:
235 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
236 case KVM_CAP_ARM_NISV_TO_USER:
237 case KVM_CAP_ARM_INJECT_EXT_DABT:
238 case KVM_CAP_SET_GUEST_DEBUG:
239 case KVM_CAP_VCPU_ATTRIBUTES:
240 case KVM_CAP_PTP_KVM:
241 case KVM_CAP_ARM_SYSTEM_SUSPEND:
242 case KVM_CAP_IRQFD_RESAMPLE:
243 case KVM_CAP_COUNTER_OFFSET:
246 case KVM_CAP_SET_GUEST_DEBUG2:
247 return KVM_GUESTDBG_VALID_MASK;
248 case KVM_CAP_ARM_SET_DEVICE_ADDR:
251 case KVM_CAP_NR_VCPUS:
253 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
254 * architectures, as it does not always bound it to
255 * KVM_CAP_MAX_VCPUS. It should not matter much because
256 * this is just an advisory value.
258 r = min_t(unsigned int, num_online_cpus(),
259 kvm_arm_default_max_vcpus());
261 case KVM_CAP_MAX_VCPUS:
262 case KVM_CAP_MAX_VCPU_ID:
266 r = kvm_arm_default_max_vcpus();
268 case KVM_CAP_MSI_DEVID:
272 r = kvm->arch.vgic.msis_require_devid;
274 case KVM_CAP_ARM_USER_IRQ:
276 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
277 * (bump this number if adding more devices)
281 case KVM_CAP_ARM_MTE:
282 r = system_supports_mte();
284 case KVM_CAP_STEAL_TIME:
285 r = kvm_arm_pvtime_supported();
287 case KVM_CAP_ARM_EL1_32BIT:
288 r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
290 case KVM_CAP_GUEST_DEBUG_HW_BPS:
293 case KVM_CAP_GUEST_DEBUG_HW_WPS:
296 case KVM_CAP_ARM_PMU_V3:
297 r = kvm_arm_support_pmu_v3();
299 case KVM_CAP_ARM_INJECT_SERROR_ESR:
300 r = cpus_have_final_cap(ARM64_HAS_RAS_EXTN);
302 case KVM_CAP_ARM_VM_IPA_SIZE:
303 r = get_kvm_ipa_limit();
305 case KVM_CAP_ARM_SVE:
306 r = system_supports_sve();
308 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
309 case KVM_CAP_ARM_PTRAUTH_GENERIC:
310 r = system_has_full_ptr_auth();
312 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
314 r = kvm->arch.mmu.split_page_chunk_size;
316 r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
318 case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
319 r = kvm_supported_block_sizes();
321 case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
331 long kvm_arch_dev_ioctl(struct file *filp,
332 unsigned int ioctl, unsigned long arg)
337 struct kvm *kvm_arch_alloc_vm(void)
339 size_t sz = sizeof(struct kvm);
342 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
344 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
347 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
349 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
352 if (id >= kvm->max_vcpus)
358 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
362 spin_lock_init(&vcpu->arch.mp_state_lock);
364 #ifdef CONFIG_LOCKDEP
365 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
366 mutex_lock(&vcpu->mutex);
367 mutex_lock(&vcpu->kvm->arch.config_lock);
368 mutex_unlock(&vcpu->kvm->arch.config_lock);
369 mutex_unlock(&vcpu->mutex);
372 /* Force users to call KVM_ARM_VCPU_INIT */
373 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
375 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
378 * Default value for the FP state, will be overloaded at load
379 * time if we support FP (pretty likely)
381 vcpu->arch.fp_state = FP_STATE_FREE;
383 /* Set up the timer */
384 kvm_timer_vcpu_init(vcpu);
386 kvm_pmu_vcpu_init(vcpu);
388 kvm_arm_reset_debug_ptr(vcpu);
390 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
392 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
394 err = kvm_vgic_vcpu_init(vcpu);
398 return kvm_share_hyp(vcpu, vcpu + 1);
401 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
405 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
407 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
408 static_branch_dec(&userspace_irqchip_in_use);
410 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
411 kvm_timer_vcpu_terminate(vcpu);
412 kvm_pmu_vcpu_destroy(vcpu);
413 kvm_vgic_vcpu_destroy(vcpu);
414 kvm_arm_vcpu_destroy(vcpu);
417 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
422 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
427 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
429 struct kvm_s2_mmu *mmu;
432 mmu = vcpu->arch.hw_mmu;
433 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
436 * We guarantee that both TLBs and I-cache are private to each
437 * vcpu. If detecting that a vcpu from the same VM has
438 * previously run on the same physical CPU, call into the
439 * hypervisor code to nuke the relevant contexts.
441 * We might get preempted before the vCPU actually runs, but
442 * over-invalidation doesn't affect correctness.
444 if (*last_ran != vcpu->vcpu_idx) {
445 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
446 *last_ran = vcpu->vcpu_idx;
452 kvm_timer_vcpu_load(vcpu);
454 kvm_vcpu_load_vhe(vcpu);
455 kvm_arch_vcpu_load_fp(vcpu);
456 kvm_vcpu_pmu_restore_guest(vcpu);
457 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
458 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
460 if (single_task_running())
461 vcpu_clear_wfx_traps(vcpu);
463 vcpu_set_wfx_traps(vcpu);
465 if (vcpu_has_ptrauth(vcpu))
466 vcpu_ptrauth_disable(vcpu);
467 kvm_arch_vcpu_load_debug_state_flags(vcpu);
469 if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
470 vcpu_set_on_unsupported_cpu(vcpu);
473 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
475 kvm_arch_vcpu_put_debug_state_flags(vcpu);
476 kvm_arch_vcpu_put_fp(vcpu);
478 kvm_vcpu_put_vhe(vcpu);
479 kvm_timer_vcpu_put(vcpu);
481 kvm_vcpu_pmu_restore_host(vcpu);
482 kvm_arm_vmid_clear_active();
484 vcpu_clear_on_unsupported_cpu(vcpu);
488 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
490 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
491 kvm_make_request(KVM_REQ_SLEEP, vcpu);
495 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
497 spin_lock(&vcpu->arch.mp_state_lock);
498 __kvm_arm_vcpu_power_off(vcpu);
499 spin_unlock(&vcpu->arch.mp_state_lock);
502 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
504 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
507 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
509 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
510 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
514 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
516 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
519 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
520 struct kvm_mp_state *mp_state)
522 *mp_state = READ_ONCE(vcpu->arch.mp_state);
527 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
528 struct kvm_mp_state *mp_state)
532 spin_lock(&vcpu->arch.mp_state_lock);
534 switch (mp_state->mp_state) {
535 case KVM_MP_STATE_RUNNABLE:
536 WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
538 case KVM_MP_STATE_STOPPED:
539 __kvm_arm_vcpu_power_off(vcpu);
541 case KVM_MP_STATE_SUSPENDED:
542 kvm_arm_vcpu_suspend(vcpu);
548 spin_unlock(&vcpu->arch.mp_state_lock);
554 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
555 * @v: The VCPU pointer
557 * If the guest CPU is not waiting for interrupts or an interrupt line is
558 * asserted, the CPU is by definition runnable.
560 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
562 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
563 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
564 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
567 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
569 return vcpu_mode_priv(vcpu);
572 #ifdef CONFIG_GUEST_PERF_EVENTS
573 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
575 return *vcpu_pc(vcpu);
579 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
581 return vcpu_get_flag(vcpu, VCPU_INITIALIZED);
584 static void kvm_init_mpidr_data(struct kvm *kvm)
586 struct kvm_mpidr_data *data = NULL;
587 unsigned long c, mask, nr_entries;
588 u64 aff_set = 0, aff_clr = ~0UL;
589 struct kvm_vcpu *vcpu;
591 mutex_lock(&kvm->arch.config_lock);
593 if (kvm->arch.mpidr_data || atomic_read(&kvm->online_vcpus) == 1)
596 kvm_for_each_vcpu(c, vcpu, kvm) {
597 u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
603 * A significant bit can be either 0 or 1, and will only appear in
604 * aff_set. Use aff_clr to weed out the useless stuff.
606 mask = aff_set ^ aff_clr;
607 nr_entries = BIT_ULL(hweight_long(mask));
610 * Don't let userspace fool us. If we need more than a single page
611 * to describe the compressed MPIDR array, just fall back to the
612 * iterative method. Single vcpu VMs do not need this either.
614 if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
615 data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
621 data->mpidr_mask = mask;
623 kvm_for_each_vcpu(c, vcpu, kvm) {
624 u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
625 u16 index = kvm_mpidr_index(data, aff);
627 data->cmpidr_to_idx[index] = c;
630 kvm->arch.mpidr_data = data;
632 mutex_unlock(&kvm->arch.config_lock);
636 * Handle both the initialisation that is being done when the vcpu is
637 * run for the first time, as well as the updates that must be
638 * performed each time we get a new thread dealing with this vcpu.
640 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
642 struct kvm *kvm = vcpu->kvm;
645 if (!kvm_vcpu_initialized(vcpu))
648 if (!kvm_arm_vcpu_is_finalized(vcpu))
651 ret = kvm_arch_vcpu_run_map_fp(vcpu);
655 if (likely(vcpu_has_run_once(vcpu)))
658 kvm_init_mpidr_data(kvm);
660 kvm_arm_vcpu_init_debug(vcpu);
662 if (likely(irqchip_in_kernel(kvm))) {
664 * Map the VGIC hardware resources before running a vcpu the
665 * first time on this VM.
667 ret = kvm_vgic_map_resources(kvm);
672 ret = kvm_timer_enable(vcpu);
676 ret = kvm_arm_pmu_v3_enable(vcpu);
680 if (is_protected_kvm_enabled()) {
681 ret = pkvm_create_hyp_vm(kvm);
686 if (!irqchip_in_kernel(kvm)) {
688 * Tell the rest of the code that there are userspace irqchip
691 static_branch_inc(&userspace_irqchip_in_use);
695 * Initialize traps for protected VMs.
696 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
697 * the code is in place for first run initialization at EL2.
699 if (kvm_vm_is_protected(kvm))
700 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
702 mutex_lock(&kvm->arch.config_lock);
703 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
704 mutex_unlock(&kvm->arch.config_lock);
709 bool kvm_arch_intc_initialized(struct kvm *kvm)
711 return vgic_initialized(kvm);
714 void kvm_arm_halt_guest(struct kvm *kvm)
717 struct kvm_vcpu *vcpu;
719 kvm_for_each_vcpu(i, vcpu, kvm)
720 vcpu->arch.pause = true;
721 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
724 void kvm_arm_resume_guest(struct kvm *kvm)
727 struct kvm_vcpu *vcpu;
729 kvm_for_each_vcpu(i, vcpu, kvm) {
730 vcpu->arch.pause = false;
731 __kvm_vcpu_wake_up(vcpu);
735 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
737 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
739 rcuwait_wait_event(wait,
740 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
743 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
744 /* Awaken to handle a signal, request we sleep again later. */
745 kvm_make_request(KVM_REQ_SLEEP, vcpu);
749 * Make sure we will observe a potential reset request if we've
750 * observed a change to the power state. Pairs with the smp_wmb() in
751 * kvm_psci_vcpu_on().
757 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
758 * @vcpu: The VCPU pointer
760 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
761 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
762 * on when a wake event arrives, e.g. there may already be a pending wake event.
764 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
767 * Sync back the state of the GIC CPU interface so that we have
768 * the latest PMR and group enables. This ensures that
769 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
770 * we have pending interrupts, e.g. when determining if the
773 * For the same reason, we want to tell GICv4 that we need
774 * doorbells to be signalled, should an interrupt become pending.
777 kvm_vgic_vmcr_sync(vcpu);
778 vcpu_set_flag(vcpu, IN_WFI);
783 vcpu_clear_flag(vcpu, IN_WFIT);
786 vcpu_clear_flag(vcpu, IN_WFI);
791 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
793 if (!kvm_arm_vcpu_suspended(vcpu))
799 * The suspend state is sticky; we do not leave it until userspace
800 * explicitly marks the vCPU as runnable. Request that we suspend again
803 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
806 * Check to make sure the vCPU is actually runnable. If so, exit to
807 * userspace informing it of the wakeup condition.
809 if (kvm_arch_vcpu_runnable(vcpu)) {
810 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
811 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
812 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
817 * Otherwise, we were unblocked to process a different event, such as a
818 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
825 * check_vcpu_requests - check and handle pending vCPU requests
826 * @vcpu: the VCPU pointer
828 * Return: 1 if we should enter the guest
829 * 0 if we should exit to userspace
830 * < 0 if we should exit to userspace, where the return value indicates
833 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
835 if (kvm_request_pending(vcpu)) {
836 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
837 kvm_vcpu_sleep(vcpu);
839 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
840 kvm_reset_vcpu(vcpu);
843 * Clear IRQ_PENDING requests that were made to guarantee
844 * that a VCPU sees new virtual interrupts.
846 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
848 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
849 kvm_update_stolen_time(vcpu);
851 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
852 /* The distributor enable bits were changed */
859 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
860 kvm_vcpu_reload_pmu(vcpu);
862 if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
863 kvm_vcpu_pmu_restore_guest(vcpu);
865 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
866 return kvm_vcpu_suspend(vcpu);
868 if (kvm_dirty_ring_check_request(vcpu))
875 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
877 if (likely(!vcpu_mode_is_32bit(vcpu)))
880 if (vcpu_has_nv(vcpu))
883 return !kvm_supports_32bit_el0();
887 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
888 * @vcpu: The VCPU pointer
889 * @ret: Pointer to write optional return code
891 * Returns: true if the VCPU needs to return to a preemptible + interruptible
892 * and skip guest entry.
894 * This function disambiguates between two different types of exits: exits to a
895 * preemptible + interruptible kernel context and exits to userspace. For an
896 * exit to userspace, this function will write the return code to ret and return
897 * true. For an exit to preemptible + interruptible kernel context (i.e. check
898 * for pending work and re-enter), return true without writing to ret.
900 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
902 struct kvm_run *run = vcpu->run;
905 * If we're using a userspace irqchip, then check if we need
906 * to tell a userspace irqchip about timer or PMU level
907 * changes and if so, exit to userspace (the actual level
908 * state gets updated in kvm_timer_update_run and
909 * kvm_pmu_update_run below).
911 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
912 if (kvm_timer_should_notify_user(vcpu) ||
913 kvm_pmu_should_notify_user(vcpu)) {
915 run->exit_reason = KVM_EXIT_INTR;
920 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
921 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
922 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
923 run->fail_entry.cpu = smp_processor_id();
928 return kvm_request_pending(vcpu) ||
929 xfer_to_guest_mode_work_pending();
933 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
934 * the vCPU is running.
936 * This must be noinstr as instrumentation may make use of RCU, and this is not
937 * safe during the EQS.
939 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
943 guest_state_enter_irqoff();
944 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
945 guest_state_exit_irqoff();
951 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
952 * @vcpu: The VCPU pointer
954 * This function is called through the VCPU_RUN ioctl called from user space. It
955 * will execute VM code in a loop until the time slice for the process is used
956 * or some emulation is needed from user space in which case the function will
957 * return with return value 0 and with the kvm_run structure filled in with the
958 * required data for the requested emulation.
960 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
962 struct kvm_run *run = vcpu->run;
965 if (run->exit_reason == KVM_EXIT_MMIO) {
966 ret = kvm_handle_mmio_return(vcpu);
973 if (run->immediate_exit) {
978 kvm_sigset_activate(vcpu);
981 run->exit_reason = KVM_EXIT_UNKNOWN;
985 * Check conditions before entering the guest
987 ret = xfer_to_guest_mode_handle_work(vcpu);
992 ret = check_vcpu_requests(vcpu);
995 * Preparing the interrupts to be injected also
996 * involves poking the GIC, which must be done in a
997 * non-preemptible context.
1002 * The VMID allocator only tracks active VMIDs per
1003 * physical CPU, and therefore the VMID allocated may not be
1004 * preserved on VMID roll-over if the task was preempted,
1005 * making a thread's VMID inactive. So we need to call
1006 * kvm_arm_vmid_update() in non-premptible context.
1008 if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1010 __load_stage2(vcpu->arch.hw_mmu,
1011 vcpu->arch.hw_mmu->arch);
1013 kvm_pmu_flush_hwstate(vcpu);
1015 local_irq_disable();
1017 kvm_vgic_flush_hwstate(vcpu);
1019 kvm_pmu_update_vcpu_events(vcpu);
1022 * Ensure we set mode to IN_GUEST_MODE after we disable
1023 * interrupts and before the final VCPU requests check.
1024 * See the comment in kvm_vcpu_exiting_guest_mode() and
1025 * Documentation/virt/kvm/vcpu-requests.rst
1027 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1029 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1030 vcpu->mode = OUTSIDE_GUEST_MODE;
1031 isb(); /* Ensure work in x_flush_hwstate is committed */
1032 kvm_pmu_sync_hwstate(vcpu);
1033 if (static_branch_unlikely(&userspace_irqchip_in_use))
1034 kvm_timer_sync_user(vcpu);
1035 kvm_vgic_sync_hwstate(vcpu);
1041 kvm_arm_setup_debug(vcpu);
1042 kvm_arch_vcpu_ctxflush_fp(vcpu);
1044 /**************************************************************
1047 trace_kvm_entry(*vcpu_pc(vcpu));
1048 guest_timing_enter_irqoff();
1050 ret = kvm_arm_vcpu_enter_exit(vcpu);
1052 vcpu->mode = OUTSIDE_GUEST_MODE;
1056 *************************************************************/
1058 kvm_arm_clear_debug(vcpu);
1061 * We must sync the PMU state before the vgic state so
1062 * that the vgic can properly sample the updated state of the
1065 kvm_pmu_sync_hwstate(vcpu);
1068 * Sync the vgic state before syncing the timer state because
1069 * the timer code needs to know if the virtual timer
1070 * interrupts are active.
1072 kvm_vgic_sync_hwstate(vcpu);
1075 * Sync the timer hardware state before enabling interrupts as
1076 * we don't want vtimer interrupts to race with syncing the
1077 * timer virtual interrupt state.
1079 if (static_branch_unlikely(&userspace_irqchip_in_use))
1080 kvm_timer_sync_user(vcpu);
1082 kvm_arch_vcpu_ctxsync_fp(vcpu);
1085 * We must ensure that any pending interrupts are taken before
1086 * we exit guest timing so that timer ticks are accounted as
1087 * guest time. Transiently unmask interrupts so that any
1088 * pending interrupts are taken.
1090 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1091 * context synchronization event) is necessary to ensure that
1092 * pending interrupts are taken.
1094 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1097 local_irq_disable();
1100 guest_timing_exit_irqoff();
1104 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1106 /* Exit types that need handling before we can be preempted */
1107 handle_exit_early(vcpu, ret);
1112 * The ARMv8 architecture doesn't give the hypervisor
1113 * a mechanism to prevent a guest from dropping to AArch32 EL0
1114 * if implemented by the CPU. If we spot the guest in such
1115 * state and that we decided it wasn't supposed to do so (like
1116 * with the asymmetric AArch32 case), return to userspace with
1119 if (vcpu_mode_is_bad_32bit(vcpu)) {
1121 * As we have caught the guest red-handed, decide that
1122 * it isn't fit for purpose anymore by making the vcpu
1123 * invalid. The VMM can try and fix it by issuing a
1124 * KVM_ARM_VCPU_INIT if it really wants to.
1126 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1127 ret = ARM_EXCEPTION_IL;
1130 ret = handle_exit(vcpu, ret);
1133 /* Tell userspace about in-kernel device output levels */
1134 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1135 kvm_timer_update_run(vcpu);
1136 kvm_pmu_update_run(vcpu);
1139 kvm_sigset_deactivate(vcpu);
1143 * In the unlikely event that we are returning to userspace
1144 * with pending exceptions or PC adjustment, commit these
1145 * adjustments in order to give userspace a consistent view of
1146 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1147 * being preempt-safe on VHE.
1149 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1150 vcpu_get_flag(vcpu, INCREMENT_PC)))
1151 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1157 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1163 if (number == KVM_ARM_IRQ_CPU_IRQ)
1164 bit_index = __ffs(HCR_VI);
1165 else /* KVM_ARM_IRQ_CPU_FIQ */
1166 bit_index = __ffs(HCR_VF);
1168 hcr = vcpu_hcr(vcpu);
1170 set = test_and_set_bit(bit_index, hcr);
1172 set = test_and_clear_bit(bit_index, hcr);
1175 * If we didn't change anything, no need to wake up or kick other CPUs
1181 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1182 * trigger a world-switch round on the running physical CPU to set the
1183 * virtual IRQ/FIQ fields in the HCR appropriately.
1185 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1186 kvm_vcpu_kick(vcpu);
1191 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1194 u32 irq = irq_level->irq;
1195 unsigned int irq_type, vcpu_id, irq_num;
1196 struct kvm_vcpu *vcpu = NULL;
1197 bool level = irq_level->level;
1199 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1200 vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1201 vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1202 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1204 trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1207 case KVM_ARM_IRQ_TYPE_CPU:
1208 if (irqchip_in_kernel(kvm))
1211 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1215 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1218 return vcpu_interrupt_line(vcpu, irq_num, level);
1219 case KVM_ARM_IRQ_TYPE_PPI:
1220 if (!irqchip_in_kernel(kvm))
1223 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1227 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1230 return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1231 case KVM_ARM_IRQ_TYPE_SPI:
1232 if (!irqchip_in_kernel(kvm))
1235 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1238 return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1244 static unsigned long system_supported_vcpu_features(void)
1246 unsigned long features = KVM_VCPU_VALID_FEATURES;
1248 if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1249 clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1251 if (!kvm_arm_support_pmu_v3())
1252 clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1254 if (!system_supports_sve())
1255 clear_bit(KVM_ARM_VCPU_SVE, &features);
1257 if (!system_has_full_ptr_auth()) {
1258 clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1259 clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1262 if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1263 clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1268 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1269 const struct kvm_vcpu_init *init)
1271 unsigned long features = init->features[0];
1274 if (features & ~KVM_VCPU_VALID_FEATURES)
1277 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1278 if (init->features[i])
1282 if (features & ~system_supported_vcpu_features())
1286 * For now make sure that both address/generic pointer authentication
1287 * features are requested by the userspace together.
1289 if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1290 test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1293 /* Disallow NV+SVE for the time being */
1294 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features) &&
1295 test_bit(KVM_ARM_VCPU_SVE, &features))
1298 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1301 /* MTE is incompatible with AArch32 */
1302 if (kvm_has_mte(vcpu->kvm))
1305 /* NV is incompatible with AArch32 */
1306 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1312 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1313 const struct kvm_vcpu_init *init)
1315 unsigned long features = init->features[0];
1317 return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1318 KVM_VCPU_MAX_FEATURES);
1321 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1323 struct kvm *kvm = vcpu->kvm;
1327 * When the vCPU has a PMU, but no PMU is set for the guest
1328 * yet, set the default one.
1330 if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1331 ret = kvm_arm_set_default_pmu(kvm);
1336 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1337 const struct kvm_vcpu_init *init)
1339 unsigned long features = init->features[0];
1340 struct kvm *kvm = vcpu->kvm;
1343 mutex_lock(&kvm->arch.config_lock);
1345 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1346 kvm_vcpu_init_changed(vcpu, init))
1349 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1351 ret = kvm_setup_vcpu(vcpu);
1355 /* Now we know what it is, we can reset it. */
1356 kvm_reset_vcpu(vcpu);
1358 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1359 vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1362 mutex_unlock(&kvm->arch.config_lock);
1366 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1367 const struct kvm_vcpu_init *init)
1371 if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1372 init->target != kvm_target_cpu())
1375 ret = kvm_vcpu_init_check_features(vcpu, init);
1379 if (!kvm_vcpu_initialized(vcpu))
1380 return __kvm_vcpu_set_target(vcpu, init);
1382 if (kvm_vcpu_init_changed(vcpu, init))
1385 kvm_reset_vcpu(vcpu);
1389 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1390 struct kvm_vcpu_init *init)
1392 bool power_off = false;
1396 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1397 * reflecting it in the finalized feature set, thus limiting its scope
1398 * to a single KVM_ARM_VCPU_INIT call.
1400 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1401 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1405 ret = kvm_vcpu_set_target(vcpu, init);
1410 * Ensure a rebooted VM will fault in RAM pages and detect if the
1411 * guest MMU is turned off and flush the caches as needed.
1413 * S2FWB enforces all memory accesses to RAM being cacheable,
1414 * ensuring that the data side is always coherent. We still
1415 * need to invalidate the I-cache though, as FWB does *not*
1416 * imply CTR_EL0.DIC.
1418 if (vcpu_has_run_once(vcpu)) {
1419 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1420 stage2_unmap_vm(vcpu->kvm);
1422 icache_inval_all_pou();
1425 vcpu_reset_hcr(vcpu);
1426 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1429 * Handle the "start in power-off" case.
1431 spin_lock(&vcpu->arch.mp_state_lock);
1434 __kvm_arm_vcpu_power_off(vcpu);
1436 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1438 spin_unlock(&vcpu->arch.mp_state_lock);
1443 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1444 struct kvm_device_attr *attr)
1448 switch (attr->group) {
1450 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1457 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1458 struct kvm_device_attr *attr)
1462 switch (attr->group) {
1464 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1471 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1472 struct kvm_device_attr *attr)
1476 switch (attr->group) {
1478 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1485 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1486 struct kvm_vcpu_events *events)
1488 memset(events, 0, sizeof(*events));
1490 return __kvm_arm_vcpu_get_events(vcpu, events);
1493 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1494 struct kvm_vcpu_events *events)
1498 /* check whether the reserved field is zero */
1499 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1500 if (events->reserved[i])
1503 /* check whether the pad field is zero */
1504 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1505 if (events->exception.pad[i])
1508 return __kvm_arm_vcpu_set_events(vcpu, events);
1511 long kvm_arch_vcpu_ioctl(struct file *filp,
1512 unsigned int ioctl, unsigned long arg)
1514 struct kvm_vcpu *vcpu = filp->private_data;
1515 void __user *argp = (void __user *)arg;
1516 struct kvm_device_attr attr;
1520 case KVM_ARM_VCPU_INIT: {
1521 struct kvm_vcpu_init init;
1524 if (copy_from_user(&init, argp, sizeof(init)))
1527 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1530 case KVM_SET_ONE_REG:
1531 case KVM_GET_ONE_REG: {
1532 struct kvm_one_reg reg;
1535 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1539 if (copy_from_user(®, argp, sizeof(reg)))
1543 * We could owe a reset due to PSCI. Handle the pending reset
1544 * here to ensure userspace register accesses are ordered after
1547 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1548 kvm_reset_vcpu(vcpu);
1550 if (ioctl == KVM_SET_ONE_REG)
1551 r = kvm_arm_set_reg(vcpu, ®);
1553 r = kvm_arm_get_reg(vcpu, ®);
1556 case KVM_GET_REG_LIST: {
1557 struct kvm_reg_list __user *user_list = argp;
1558 struct kvm_reg_list reg_list;
1562 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1566 if (!kvm_arm_vcpu_is_finalized(vcpu))
1570 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1573 reg_list.n = kvm_arm_num_regs(vcpu);
1574 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1579 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1582 case KVM_SET_DEVICE_ATTR: {
1584 if (copy_from_user(&attr, argp, sizeof(attr)))
1586 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1589 case KVM_GET_DEVICE_ATTR: {
1591 if (copy_from_user(&attr, argp, sizeof(attr)))
1593 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1596 case KVM_HAS_DEVICE_ATTR: {
1598 if (copy_from_user(&attr, argp, sizeof(attr)))
1600 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1603 case KVM_GET_VCPU_EVENTS: {
1604 struct kvm_vcpu_events events;
1606 if (kvm_arm_vcpu_get_events(vcpu, &events))
1609 if (copy_to_user(argp, &events, sizeof(events)))
1614 case KVM_SET_VCPU_EVENTS: {
1615 struct kvm_vcpu_events events;
1617 if (copy_from_user(&events, argp, sizeof(events)))
1620 return kvm_arm_vcpu_set_events(vcpu, &events);
1622 case KVM_ARM_VCPU_FINALIZE: {
1625 if (!kvm_vcpu_initialized(vcpu))
1628 if (get_user(what, (const int __user *)argp))
1631 return kvm_arm_vcpu_finalize(vcpu, what);
1640 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1645 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1646 struct kvm_arm_device_addr *dev_addr)
1648 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1649 case KVM_ARM_DEVICE_VGIC_V2:
1652 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1658 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1660 switch (attr->group) {
1661 case KVM_ARM_VM_SMCCC_CTRL:
1662 return kvm_vm_smccc_has_attr(kvm, attr);
1668 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1670 switch (attr->group) {
1671 case KVM_ARM_VM_SMCCC_CTRL:
1672 return kvm_vm_smccc_set_attr(kvm, attr);
1678 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1680 struct kvm *kvm = filp->private_data;
1681 void __user *argp = (void __user *)arg;
1682 struct kvm_device_attr attr;
1685 case KVM_CREATE_IRQCHIP: {
1689 mutex_lock(&kvm->lock);
1690 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1691 mutex_unlock(&kvm->lock);
1694 case KVM_ARM_SET_DEVICE_ADDR: {
1695 struct kvm_arm_device_addr dev_addr;
1697 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1699 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1701 case KVM_ARM_PREFERRED_TARGET: {
1702 struct kvm_vcpu_init init = {
1703 .target = KVM_ARM_TARGET_GENERIC_V8,
1706 if (copy_to_user(argp, &init, sizeof(init)))
1711 case KVM_ARM_MTE_COPY_TAGS: {
1712 struct kvm_arm_copy_mte_tags copy_tags;
1714 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1716 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1718 case KVM_ARM_SET_COUNTER_OFFSET: {
1719 struct kvm_arm_counter_offset offset;
1721 if (copy_from_user(&offset, argp, sizeof(offset)))
1723 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1725 case KVM_HAS_DEVICE_ATTR: {
1726 if (copy_from_user(&attr, argp, sizeof(attr)))
1729 return kvm_vm_has_attr(kvm, &attr);
1731 case KVM_SET_DEVICE_ATTR: {
1732 if (copy_from_user(&attr, argp, sizeof(attr)))
1735 return kvm_vm_set_attr(kvm, &attr);
1737 case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1738 struct reg_mask_range range;
1740 if (copy_from_user(&range, argp, sizeof(range)))
1742 return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1749 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1750 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1752 struct kvm_vcpu *tmp_vcpu;
1754 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1755 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1756 mutex_unlock(&tmp_vcpu->mutex);
1760 void unlock_all_vcpus(struct kvm *kvm)
1762 lockdep_assert_held(&kvm->lock);
1764 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1767 /* Returns true if all vcpus were locked, false otherwise */
1768 bool lock_all_vcpus(struct kvm *kvm)
1770 struct kvm_vcpu *tmp_vcpu;
1773 lockdep_assert_held(&kvm->lock);
1776 * Any time a vcpu is in an ioctl (including running), the
1777 * core KVM code tries to grab the vcpu->mutex.
1779 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1780 * other VCPUs can fiddle with the state while we access it.
1782 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1783 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1784 unlock_vcpus(kvm, c - 1);
1792 static unsigned long nvhe_percpu_size(void)
1794 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1795 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1798 static unsigned long nvhe_percpu_order(void)
1800 unsigned long size = nvhe_percpu_size();
1802 return size ? get_order(size) : 0;
1805 /* A lookup table holding the hypervisor VA for each vector slot */
1806 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1808 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1810 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1813 static int kvm_init_vector_slots(void)
1818 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1819 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1821 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1822 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1824 if (kvm_system_needs_idmapped_vectors() &&
1825 !is_protected_kvm_enabled()) {
1826 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1827 __BP_HARDEN_HYP_VECS_SZ, &base);
1832 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1833 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1837 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1839 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1843 * Calculate the raw per-cpu offset without a translation from the
1844 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1845 * so that we can use adr_l to access per-cpu variables in EL2.
1846 * Also drop the KASAN tag which gets in the way...
1848 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1849 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1851 params->mair_el2 = read_sysreg(mair_el1);
1853 tcr = read_sysreg(tcr_el1);
1854 if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1855 tcr |= TCR_EPD1_MASK;
1857 tcr &= TCR_EL2_MASK;
1858 tcr |= TCR_EL2_RES1;
1860 tcr &= ~TCR_T0SZ_MASK;
1861 tcr |= TCR_T0SZ(hyp_va_bits);
1862 params->tcr_el2 = tcr;
1864 params->pgd_pa = kvm_mmu_get_httbr();
1865 if (is_protected_kvm_enabled())
1866 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1868 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1869 if (cpus_have_final_cap(ARM64_KVM_HVHE))
1870 params->hcr_el2 |= HCR_E2H;
1871 params->vttbr = params->vtcr = 0;
1874 * Flush the init params from the data cache because the struct will
1875 * be read while the MMU is off.
1877 kvm_flush_dcache_to_poc(params, sizeof(*params));
1880 static void hyp_install_host_vector(void)
1882 struct kvm_nvhe_init_params *params;
1883 struct arm_smccc_res res;
1885 /* Switch from the HYP stub to our own HYP init vector */
1886 __hyp_set_vectors(kvm_get_idmap_vector());
1889 * Call initialization code, and switch to the full blown HYP code.
1890 * If the cpucaps haven't been finalized yet, something has gone very
1891 * wrong, and hyp will crash and burn when it uses any
1892 * cpus_have_*_cap() wrapper.
1894 BUG_ON(!system_capabilities_finalized());
1895 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1896 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1897 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1900 static void cpu_init_hyp_mode(void)
1902 hyp_install_host_vector();
1905 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1908 if (this_cpu_has_cap(ARM64_SSBS) &&
1909 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1910 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1914 static void cpu_hyp_reset(void)
1916 if (!is_kernel_in_hyp_mode())
1917 __hyp_reset_vectors();
1921 * EL2 vectors can be mapped and rerouted in a number of ways,
1922 * depending on the kernel configuration and CPU present:
1924 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1925 * placed in one of the vector slots, which is executed before jumping
1926 * to the real vectors.
1928 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1929 * containing the hardening sequence is mapped next to the idmap page,
1930 * and executed before jumping to the real vectors.
1932 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1933 * empty slot is selected, mapped next to the idmap page, and
1934 * executed before jumping to the real vectors.
1936 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1937 * VHE, as we don't have hypervisor-specific mappings. If the system
1938 * is VHE and yet selects this capability, it will be ignored.
1940 static void cpu_set_hyp_vector(void)
1942 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1943 void *vector = hyp_spectre_vector_selector[data->slot];
1945 if (!is_protected_kvm_enabled())
1946 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1948 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1951 static void cpu_hyp_init_context(void)
1953 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1955 if (!is_kernel_in_hyp_mode())
1956 cpu_init_hyp_mode();
1959 static void cpu_hyp_init_features(void)
1961 cpu_set_hyp_vector();
1962 kvm_arm_init_debug();
1964 if (is_kernel_in_hyp_mode())
1965 kvm_timer_init_vhe();
1968 kvm_vgic_init_cpu_hardware();
1971 static void cpu_hyp_reinit(void)
1974 cpu_hyp_init_context();
1975 cpu_hyp_init_features();
1978 static void cpu_hyp_init(void *discard)
1980 if (!__this_cpu_read(kvm_hyp_initialized)) {
1982 __this_cpu_write(kvm_hyp_initialized, 1);
1986 static void cpu_hyp_uninit(void *discard)
1988 if (__this_cpu_read(kvm_hyp_initialized)) {
1990 __this_cpu_write(kvm_hyp_initialized, 0);
1994 int kvm_arch_hardware_enable(void)
1997 * Most calls to this function are made with migration
1998 * disabled, but not with preemption disabled. The former is
1999 * enough to ensure correctness, but most of the helpers
2000 * expect the later and will throw a tantrum otherwise.
2014 void kvm_arch_hardware_disable(void)
2016 kvm_timer_cpu_down();
2017 kvm_vgic_cpu_down();
2019 if (!is_protected_kvm_enabled())
2020 cpu_hyp_uninit(NULL);
2023 #ifdef CONFIG_CPU_PM
2024 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2029 * kvm_hyp_initialized is left with its old value over
2030 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2035 if (__this_cpu_read(kvm_hyp_initialized))
2037 * don't update kvm_hyp_initialized here
2038 * so that the hyp will be re-enabled
2039 * when we resume. See below.
2044 case CPU_PM_ENTER_FAILED:
2046 if (__this_cpu_read(kvm_hyp_initialized))
2047 /* The hyp was enabled before suspend. */
2057 static struct notifier_block hyp_init_cpu_pm_nb = {
2058 .notifier_call = hyp_init_cpu_pm_notifier,
2061 static void __init hyp_cpu_pm_init(void)
2063 if (!is_protected_kvm_enabled())
2064 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2066 static void __init hyp_cpu_pm_exit(void)
2068 if (!is_protected_kvm_enabled())
2069 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2072 static inline void __init hyp_cpu_pm_init(void)
2075 static inline void __init hyp_cpu_pm_exit(void)
2080 static void __init init_cpu_logical_map(void)
2085 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2086 * Only copy the set of online CPUs whose features have been checked
2087 * against the finalized system capabilities. The hypervisor will not
2088 * allow any other CPUs from the `possible` set to boot.
2090 for_each_online_cpu(cpu)
2091 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2094 #define init_psci_0_1_impl_state(config, what) \
2095 config.psci_0_1_ ## what ## _implemented = psci_ops.what
2097 static bool __init init_psci_relay(void)
2100 * If PSCI has not been initialized, protected KVM cannot install
2101 * itself on newly booted CPUs.
2103 if (!psci_ops.get_version) {
2104 kvm_err("Cannot initialize protected mode without PSCI\n");
2108 kvm_host_psci_config.version = psci_ops.get_version();
2109 kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2111 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2112 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2113 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2114 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2115 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2116 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2121 static int __init init_subsystems(void)
2126 * Enable hardware so that subsystem initialisation can access EL2.
2128 on_each_cpu(cpu_hyp_init, NULL, 1);
2131 * Register CPU lower-power notifier
2136 * Init HYP view of VGIC
2138 err = kvm_vgic_hyp_init();
2141 vgic_present = true;
2145 vgic_present = false;
2153 * Init HYP architected timer support
2155 err = kvm_timer_hyp_init(vgic_present);
2159 kvm_register_perf_callbacks(NULL);
2165 if (err || !is_protected_kvm_enabled())
2166 on_each_cpu(cpu_hyp_uninit, NULL, 1);
2171 static void __init teardown_subsystems(void)
2173 kvm_unregister_perf_callbacks();
2177 static void __init teardown_hyp_mode(void)
2182 for_each_possible_cpu(cpu) {
2183 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2184 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2188 static int __init do_pkvm_init(u32 hyp_va_bits)
2190 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2194 cpu_hyp_init_context();
2195 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2196 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2198 cpu_hyp_init_features();
2201 * The stub hypercalls are now disabled, so set our local flag to
2202 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2204 __this_cpu_write(kvm_hyp_initialized, 1);
2210 static u64 get_hyp_id_aa64pfr0_el1(void)
2213 * Track whether the system isn't affected by spectre/meltdown in the
2214 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2215 * Although this is per-CPU, we make it global for simplicity, e.g., not
2216 * to have to worry about vcpu migration.
2218 * Unlike for non-protected VMs, userspace cannot override this for
2221 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2223 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2224 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2226 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2227 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2228 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2229 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2234 static void kvm_hyp_init_symbols(void)
2236 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2237 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2238 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2239 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2240 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2241 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2242 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2243 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2244 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2245 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2246 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2249 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2251 void *addr = phys_to_virt(hyp_mem_base);
2254 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2258 ret = do_pkvm_init(hyp_va_bits);
2267 static void pkvm_hyp_init_ptrauth(void)
2269 struct kvm_cpu_context *hyp_ctxt;
2272 for_each_possible_cpu(cpu) {
2273 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2274 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2275 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2276 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2277 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2278 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2279 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2280 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2281 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2282 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2283 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2287 /* Inits Hyp-mode on all online CPUs */
2288 static int __init init_hyp_mode(void)
2295 * The protected Hyp-mode cannot be initialized if the memory pool
2296 * allocation has failed.
2298 if (is_protected_kvm_enabled() && !hyp_mem_base)
2302 * Allocate Hyp PGD and setup Hyp identity mapping
2304 err = kvm_mmu_init(&hyp_va_bits);
2309 * Allocate stack pages for Hypervisor-mode
2311 for_each_possible_cpu(cpu) {
2312 unsigned long stack_page;
2314 stack_page = __get_free_page(GFP_KERNEL);
2320 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2324 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2326 for_each_possible_cpu(cpu) {
2330 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2336 page_addr = page_address(page);
2337 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2338 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2342 * Map the Hyp-code called directly from the host
2344 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2345 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2347 kvm_err("Cannot map world-switch code\n");
2351 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2352 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2354 kvm_err("Cannot map .hyp.rodata section\n");
2358 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2359 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2361 kvm_err("Cannot map rodata section\n");
2366 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2367 * section thanks to an assertion in the linker script. Map it RW and
2368 * the rest of .bss RO.
2370 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2371 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2373 kvm_err("Cannot map hyp bss section: %d\n", err);
2377 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2378 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2380 kvm_err("Cannot map bss section\n");
2385 * Map the Hyp stack pages
2387 for_each_possible_cpu(cpu) {
2388 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2389 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2391 err = create_hyp_stack(__pa(stack_page), ¶ms->stack_hyp_va);
2393 kvm_err("Cannot map hyp stack\n");
2398 * Save the stack PA in nvhe_init_params. This will be needed
2399 * to recreate the stack mapping in protected nVHE mode.
2400 * __hyp_pa() won't do the right thing there, since the stack
2401 * has been mapped in the flexible private VA space.
2403 params->stack_pa = __pa(stack_page);
2406 for_each_possible_cpu(cpu) {
2407 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2408 char *percpu_end = percpu_begin + nvhe_percpu_size();
2410 /* Map Hyp percpu pages */
2411 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2413 kvm_err("Cannot map hyp percpu region\n");
2417 /* Prepare the CPU initialization parameters */
2418 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2421 kvm_hyp_init_symbols();
2423 if (is_protected_kvm_enabled()) {
2424 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2425 cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2426 pkvm_hyp_init_ptrauth();
2428 init_cpu_logical_map();
2430 if (!init_psci_relay()) {
2435 err = kvm_hyp_init_protection(hyp_va_bits);
2437 kvm_err("Failed to init hyp memory protection\n");
2445 teardown_hyp_mode();
2446 kvm_err("error initializing Hyp mode: %d\n", err);
2450 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2452 struct kvm_vcpu *vcpu;
2455 mpidr &= MPIDR_HWID_BITMASK;
2457 if (kvm->arch.mpidr_data) {
2458 u16 idx = kvm_mpidr_index(kvm->arch.mpidr_data, mpidr);
2460 vcpu = kvm_get_vcpu(kvm,
2461 kvm->arch.mpidr_data->cmpidr_to_idx[idx]);
2462 if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2468 kvm_for_each_vcpu(i, vcpu, kvm) {
2469 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2475 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2477 return irqchip_in_kernel(kvm);
2480 bool kvm_arch_has_irq_bypass(void)
2485 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2486 struct irq_bypass_producer *prod)
2488 struct kvm_kernel_irqfd *irqfd =
2489 container_of(cons, struct kvm_kernel_irqfd, consumer);
2491 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2494 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2495 struct irq_bypass_producer *prod)
2497 struct kvm_kernel_irqfd *irqfd =
2498 container_of(cons, struct kvm_kernel_irqfd, consumer);
2500 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2504 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2506 struct kvm_kernel_irqfd *irqfd =
2507 container_of(cons, struct kvm_kernel_irqfd, consumer);
2509 kvm_arm_halt_guest(irqfd->kvm);
2512 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2514 struct kvm_kernel_irqfd *irqfd =
2515 container_of(cons, struct kvm_kernel_irqfd, consumer);
2517 kvm_arm_resume_guest(irqfd->kvm);
2520 /* Initialize Hyp-mode and memory mappings on all CPUs */
2521 static __init int kvm_arm_init(void)
2526 if (!is_hyp_mode_available()) {
2527 kvm_info("HYP mode not available\n");
2531 if (kvm_get_mode() == KVM_MODE_NONE) {
2532 kvm_info("KVM disabled from command line\n");
2536 err = kvm_sys_reg_table_init();
2538 kvm_info("Error initializing system register tables");
2542 in_hyp_mode = is_kernel_in_hyp_mode();
2544 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2545 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2546 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2547 "Only trusted guests should be used on this system.\n");
2549 err = kvm_set_ipa_limit();
2553 err = kvm_arm_init_sve();
2557 err = kvm_arm_vmid_alloc_init();
2559 kvm_err("Failed to initialize VMID allocator.\n");
2564 err = init_hyp_mode();
2569 err = kvm_init_vector_slots();
2571 kvm_err("Cannot initialise vector slots\n");
2575 err = init_subsystems();
2579 if (is_protected_kvm_enabled()) {
2580 kvm_info("Protected nVHE mode initialized successfully\n");
2581 } else if (in_hyp_mode) {
2582 kvm_info("VHE mode initialized successfully\n");
2584 kvm_info("Hyp mode initialized successfully\n");
2588 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2589 * hypervisor protection is finalized.
2591 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2595 kvm_arm_initialised = true;
2600 teardown_subsystems();
2603 teardown_hyp_mode();
2605 kvm_arm_vmid_alloc_free();
2609 static int __init early_kvm_mode_cfg(char *arg)
2614 if (strcmp(arg, "none") == 0) {
2615 kvm_mode = KVM_MODE_NONE;
2619 if (!is_hyp_mode_available()) {
2620 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2624 if (strcmp(arg, "protected") == 0) {
2625 if (!is_kernel_in_hyp_mode())
2626 kvm_mode = KVM_MODE_PROTECTED;
2628 pr_warn_once("Protected KVM not available with VHE\n");
2633 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2634 kvm_mode = KVM_MODE_DEFAULT;
2638 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2639 kvm_mode = KVM_MODE_NV;
2645 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2647 enum kvm_mode kvm_get_mode(void)
2652 module_init(kvm_arm_init);