1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
8 #include <linux/cpu_pm.h>
9 #include <linux/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kvm.h>
20 #include <linux/kvm_irqfd.h>
21 #include <linux/irqbypass.h>
22 #include <linux/sched/stat.h>
23 #include <linux/psci.h>
24 #include <trace/events/kvm.h>
26 #define CREATE_TRACE_POINTS
27 #include "trace_arm.h"
29 #include <linux/uaccess.h>
30 #include <asm/ptrace.h>
32 #include <asm/tlbflush.h>
33 #include <asm/cacheflush.h>
34 #include <asm/cpufeature.h>
36 #include <asm/kvm_arm.h>
37 #include <asm/kvm_asm.h>
38 #include <asm/kvm_mmu.h>
39 #include <asm/kvm_pkvm.h>
40 #include <asm/kvm_emulate.h>
41 #include <asm/sections.h>
43 #include <kvm/arm_hypercalls.h>
44 #include <kvm/arm_pmu.h>
45 #include <kvm/arm_psci.h>
47 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
49 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
51 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
56 static bool vgic_present;
58 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
59 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
61 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
63 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
66 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
67 struct kvm_enable_cap *cap)
76 case KVM_CAP_ARM_NISV_TO_USER:
78 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
82 mutex_lock(&kvm->lock);
83 if (!system_supports_mte() || kvm->created_vcpus) {
87 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
89 mutex_unlock(&kvm->lock);
91 case KVM_CAP_ARM_SYSTEM_SUSPEND:
93 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
95 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
96 new_cap = cap->args[0];
98 mutex_lock(&kvm->slots_lock);
100 * To keep things simple, allow changing the chunk
101 * size only when no memory slots have been created.
103 if (!kvm_are_all_memslots_empty(kvm)) {
105 } else if (new_cap && !kvm_is_block_size_supported(new_cap)) {
109 kvm->arch.mmu.split_page_chunk_size = new_cap;
111 mutex_unlock(&kvm->slots_lock);
121 static int kvm_arm_default_max_vcpus(void)
123 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
127 * kvm_arch_init_vm - initializes a VM data structure
128 * @kvm: pointer to the KVM struct
130 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
134 mutex_init(&kvm->arch.config_lock);
136 #ifdef CONFIG_LOCKDEP
137 /* Clue in lockdep that the config_lock must be taken inside kvm->lock */
138 mutex_lock(&kvm->lock);
139 mutex_lock(&kvm->arch.config_lock);
140 mutex_unlock(&kvm->arch.config_lock);
141 mutex_unlock(&kvm->lock);
144 ret = kvm_share_hyp(kvm, kvm + 1);
148 ret = pkvm_init_host_vm(kvm);
150 goto err_unshare_kvm;
152 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
154 goto err_unshare_kvm;
156 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
158 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
160 goto err_free_cpumask;
162 kvm_vgic_early_init(kvm);
164 kvm_timer_init_vm(kvm);
166 /* The maximum number of VCPUs is limited by the host's GIC model */
167 kvm->max_vcpus = kvm_arm_default_max_vcpus();
169 kvm_arm_init_hypercalls(kvm);
171 bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
176 free_cpumask_var(kvm->arch.supported_cpus);
178 kvm_unshare_hyp(kvm, kvm + 1);
182 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
184 return VM_FAULT_SIGBUS;
189 * kvm_arch_destroy_vm - destroy the VM data structure
190 * @kvm: pointer to the KVM struct
192 void kvm_arch_destroy_vm(struct kvm *kvm)
194 bitmap_free(kvm->arch.pmu_filter);
195 free_cpumask_var(kvm->arch.supported_cpus);
197 kvm_vgic_destroy(kvm);
199 if (is_protected_kvm_enabled())
200 pkvm_destroy_hyp_vm(kvm);
202 kvm_destroy_vcpus(kvm);
204 kvm_unshare_hyp(kvm, kvm + 1);
206 kvm_arm_teardown_hypercalls(kvm);
209 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
213 case KVM_CAP_IRQCHIP:
216 case KVM_CAP_IOEVENTFD:
217 case KVM_CAP_DEVICE_CTRL:
218 case KVM_CAP_USER_MEMORY:
219 case KVM_CAP_SYNC_MMU:
220 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
221 case KVM_CAP_ONE_REG:
222 case KVM_CAP_ARM_PSCI:
223 case KVM_CAP_ARM_PSCI_0_2:
224 case KVM_CAP_READONLY_MEM:
225 case KVM_CAP_MP_STATE:
226 case KVM_CAP_IMMEDIATE_EXIT:
227 case KVM_CAP_VCPU_EVENTS:
228 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
229 case KVM_CAP_ARM_NISV_TO_USER:
230 case KVM_CAP_ARM_INJECT_EXT_DABT:
231 case KVM_CAP_SET_GUEST_DEBUG:
232 case KVM_CAP_VCPU_ATTRIBUTES:
233 case KVM_CAP_PTP_KVM:
234 case KVM_CAP_ARM_SYSTEM_SUSPEND:
235 case KVM_CAP_IRQFD_RESAMPLE:
236 case KVM_CAP_COUNTER_OFFSET:
239 case KVM_CAP_SET_GUEST_DEBUG2:
240 return KVM_GUESTDBG_VALID_MASK;
241 case KVM_CAP_ARM_SET_DEVICE_ADDR:
244 case KVM_CAP_NR_VCPUS:
246 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
247 * architectures, as it does not always bound it to
248 * KVM_CAP_MAX_VCPUS. It should not matter much because
249 * this is just an advisory value.
251 r = min_t(unsigned int, num_online_cpus(),
252 kvm_arm_default_max_vcpus());
254 case KVM_CAP_MAX_VCPUS:
255 case KVM_CAP_MAX_VCPU_ID:
259 r = kvm_arm_default_max_vcpus();
261 case KVM_CAP_MSI_DEVID:
265 r = kvm->arch.vgic.msis_require_devid;
267 case KVM_CAP_ARM_USER_IRQ:
269 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
270 * (bump this number if adding more devices)
274 case KVM_CAP_ARM_MTE:
275 r = system_supports_mte();
277 case KVM_CAP_STEAL_TIME:
278 r = kvm_arm_pvtime_supported();
280 case KVM_CAP_ARM_EL1_32BIT:
281 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
283 case KVM_CAP_GUEST_DEBUG_HW_BPS:
286 case KVM_CAP_GUEST_DEBUG_HW_WPS:
289 case KVM_CAP_ARM_PMU_V3:
290 r = kvm_arm_support_pmu_v3();
292 case KVM_CAP_ARM_INJECT_SERROR_ESR:
293 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
295 case KVM_CAP_ARM_VM_IPA_SIZE:
296 r = get_kvm_ipa_limit();
298 case KVM_CAP_ARM_SVE:
299 r = system_supports_sve();
301 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
302 case KVM_CAP_ARM_PTRAUTH_GENERIC:
303 r = system_has_full_ptr_auth();
305 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
307 r = kvm->arch.mmu.split_page_chunk_size;
309 r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
311 case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
312 r = kvm_supported_block_sizes();
321 long kvm_arch_dev_ioctl(struct file *filp,
322 unsigned int ioctl, unsigned long arg)
327 struct kvm *kvm_arch_alloc_vm(void)
329 size_t sz = sizeof(struct kvm);
332 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
334 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
337 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
339 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
342 if (id >= kvm->max_vcpus)
348 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
352 spin_lock_init(&vcpu->arch.mp_state_lock);
354 #ifdef CONFIG_LOCKDEP
355 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
356 mutex_lock(&vcpu->mutex);
357 mutex_lock(&vcpu->kvm->arch.config_lock);
358 mutex_unlock(&vcpu->kvm->arch.config_lock);
359 mutex_unlock(&vcpu->mutex);
362 /* Force users to call KVM_ARM_VCPU_INIT */
363 vcpu->arch.target = -1;
364 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
366 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
369 * Default value for the FP state, will be overloaded at load
370 * time if we support FP (pretty likely)
372 vcpu->arch.fp_state = FP_STATE_FREE;
374 /* Set up the timer */
375 kvm_timer_vcpu_init(vcpu);
377 kvm_pmu_vcpu_init(vcpu);
379 kvm_arm_reset_debug_ptr(vcpu);
381 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
383 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
385 err = kvm_vgic_vcpu_init(vcpu);
389 return kvm_share_hyp(vcpu, vcpu + 1);
392 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
396 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
398 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
399 static_branch_dec(&userspace_irqchip_in_use);
401 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
402 kvm_timer_vcpu_terminate(vcpu);
403 kvm_pmu_vcpu_destroy(vcpu);
405 kvm_arm_vcpu_destroy(vcpu);
408 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
413 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
418 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
420 struct kvm_s2_mmu *mmu;
423 mmu = vcpu->arch.hw_mmu;
424 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
427 * We guarantee that both TLBs and I-cache are private to each
428 * vcpu. If detecting that a vcpu from the same VM has
429 * previously run on the same physical CPU, call into the
430 * hypervisor code to nuke the relevant contexts.
432 * We might get preempted before the vCPU actually runs, but
433 * over-invalidation doesn't affect correctness.
435 if (*last_ran != vcpu->vcpu_id) {
436 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
437 *last_ran = vcpu->vcpu_id;
443 kvm_timer_vcpu_load(vcpu);
445 kvm_vcpu_load_sysregs_vhe(vcpu);
446 kvm_arch_vcpu_load_fp(vcpu);
447 kvm_vcpu_pmu_restore_guest(vcpu);
448 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
449 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
451 if (single_task_running())
452 vcpu_clear_wfx_traps(vcpu);
454 vcpu_set_wfx_traps(vcpu);
456 if (vcpu_has_ptrauth(vcpu))
457 vcpu_ptrauth_disable(vcpu);
458 kvm_arch_vcpu_load_debug_state_flags(vcpu);
460 if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
461 vcpu_set_on_unsupported_cpu(vcpu);
464 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
466 kvm_arch_vcpu_put_debug_state_flags(vcpu);
467 kvm_arch_vcpu_put_fp(vcpu);
469 kvm_vcpu_put_sysregs_vhe(vcpu);
470 kvm_timer_vcpu_put(vcpu);
472 kvm_vcpu_pmu_restore_host(vcpu);
473 kvm_arm_vmid_clear_active();
475 vcpu_clear_on_unsupported_cpu(vcpu);
479 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
481 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
482 kvm_make_request(KVM_REQ_SLEEP, vcpu);
486 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
488 spin_lock(&vcpu->arch.mp_state_lock);
489 __kvm_arm_vcpu_power_off(vcpu);
490 spin_unlock(&vcpu->arch.mp_state_lock);
493 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
495 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
498 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
500 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
501 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
505 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
507 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
510 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
511 struct kvm_mp_state *mp_state)
513 *mp_state = READ_ONCE(vcpu->arch.mp_state);
518 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
519 struct kvm_mp_state *mp_state)
523 spin_lock(&vcpu->arch.mp_state_lock);
525 switch (mp_state->mp_state) {
526 case KVM_MP_STATE_RUNNABLE:
527 WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
529 case KVM_MP_STATE_STOPPED:
530 __kvm_arm_vcpu_power_off(vcpu);
532 case KVM_MP_STATE_SUSPENDED:
533 kvm_arm_vcpu_suspend(vcpu);
539 spin_unlock(&vcpu->arch.mp_state_lock);
545 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
546 * @v: The VCPU pointer
548 * If the guest CPU is not waiting for interrupts or an interrupt line is
549 * asserted, the CPU is by definition runnable.
551 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
553 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
554 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
555 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
558 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
560 return vcpu_mode_priv(vcpu);
563 #ifdef CONFIG_GUEST_PERF_EVENTS
564 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
566 return *vcpu_pc(vcpu);
570 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
572 return vcpu->arch.target >= 0;
576 * Handle both the initialisation that is being done when the vcpu is
577 * run for the first time, as well as the updates that must be
578 * performed each time we get a new thread dealing with this vcpu.
580 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
582 struct kvm *kvm = vcpu->kvm;
585 if (!kvm_vcpu_initialized(vcpu))
588 if (!kvm_arm_vcpu_is_finalized(vcpu))
591 ret = kvm_arch_vcpu_run_map_fp(vcpu);
595 if (likely(vcpu_has_run_once(vcpu)))
598 kvm_arm_vcpu_init_debug(vcpu);
600 if (likely(irqchip_in_kernel(kvm))) {
602 * Map the VGIC hardware resources before running a vcpu the
603 * first time on this VM.
605 ret = kvm_vgic_map_resources(kvm);
610 ret = kvm_timer_enable(vcpu);
614 ret = kvm_arm_pmu_v3_enable(vcpu);
618 if (is_protected_kvm_enabled()) {
619 ret = pkvm_create_hyp_vm(kvm);
624 if (!irqchip_in_kernel(kvm)) {
626 * Tell the rest of the code that there are userspace irqchip
629 static_branch_inc(&userspace_irqchip_in_use);
633 * Initialize traps for protected VMs.
634 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
635 * the code is in place for first run initialization at EL2.
637 if (kvm_vm_is_protected(kvm))
638 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
640 mutex_lock(&kvm->arch.config_lock);
641 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
642 mutex_unlock(&kvm->arch.config_lock);
647 bool kvm_arch_intc_initialized(struct kvm *kvm)
649 return vgic_initialized(kvm);
652 void kvm_arm_halt_guest(struct kvm *kvm)
655 struct kvm_vcpu *vcpu;
657 kvm_for_each_vcpu(i, vcpu, kvm)
658 vcpu->arch.pause = true;
659 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
662 void kvm_arm_resume_guest(struct kvm *kvm)
665 struct kvm_vcpu *vcpu;
667 kvm_for_each_vcpu(i, vcpu, kvm) {
668 vcpu->arch.pause = false;
669 __kvm_vcpu_wake_up(vcpu);
673 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
675 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
677 rcuwait_wait_event(wait,
678 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
681 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
682 /* Awaken to handle a signal, request we sleep again later. */
683 kvm_make_request(KVM_REQ_SLEEP, vcpu);
687 * Make sure we will observe a potential reset request if we've
688 * observed a change to the power state. Pairs with the smp_wmb() in
689 * kvm_psci_vcpu_on().
695 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
696 * @vcpu: The VCPU pointer
698 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
699 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
700 * on when a wake event arrives, e.g. there may already be a pending wake event.
702 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
705 * Sync back the state of the GIC CPU interface so that we have
706 * the latest PMR and group enables. This ensures that
707 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
708 * we have pending interrupts, e.g. when determining if the
711 * For the same reason, we want to tell GICv4 that we need
712 * doorbells to be signalled, should an interrupt become pending.
715 kvm_vgic_vmcr_sync(vcpu);
716 vgic_v4_put(vcpu, true);
720 vcpu_clear_flag(vcpu, IN_WFIT);
727 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
729 if (!kvm_arm_vcpu_suspended(vcpu))
735 * The suspend state is sticky; we do not leave it until userspace
736 * explicitly marks the vCPU as runnable. Request that we suspend again
739 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
742 * Check to make sure the vCPU is actually runnable. If so, exit to
743 * userspace informing it of the wakeup condition.
745 if (kvm_arch_vcpu_runnable(vcpu)) {
746 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
747 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
748 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
753 * Otherwise, we were unblocked to process a different event, such as a
754 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
761 * check_vcpu_requests - check and handle pending vCPU requests
762 * @vcpu: the VCPU pointer
764 * Return: 1 if we should enter the guest
765 * 0 if we should exit to userspace
766 * < 0 if we should exit to userspace, where the return value indicates
769 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
771 if (kvm_request_pending(vcpu)) {
772 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
773 kvm_vcpu_sleep(vcpu);
775 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
776 kvm_reset_vcpu(vcpu);
779 * Clear IRQ_PENDING requests that were made to guarantee
780 * that a VCPU sees new virtual interrupts.
782 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
784 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
785 kvm_update_stolen_time(vcpu);
787 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
788 /* The distributor enable bits were changed */
790 vgic_v4_put(vcpu, false);
795 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
796 kvm_pmu_handle_pmcr(vcpu,
797 __vcpu_sys_reg(vcpu, PMCR_EL0));
799 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
800 return kvm_vcpu_suspend(vcpu);
802 if (kvm_dirty_ring_check_request(vcpu))
809 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
811 if (likely(!vcpu_mode_is_32bit(vcpu)))
814 return !kvm_supports_32bit_el0();
818 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
819 * @vcpu: The VCPU pointer
820 * @ret: Pointer to write optional return code
822 * Returns: true if the VCPU needs to return to a preemptible + interruptible
823 * and skip guest entry.
825 * This function disambiguates between two different types of exits: exits to a
826 * preemptible + interruptible kernel context and exits to userspace. For an
827 * exit to userspace, this function will write the return code to ret and return
828 * true. For an exit to preemptible + interruptible kernel context (i.e. check
829 * for pending work and re-enter), return true without writing to ret.
831 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
833 struct kvm_run *run = vcpu->run;
836 * If we're using a userspace irqchip, then check if we need
837 * to tell a userspace irqchip about timer or PMU level
838 * changes and if so, exit to userspace (the actual level
839 * state gets updated in kvm_timer_update_run and
840 * kvm_pmu_update_run below).
842 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
843 if (kvm_timer_should_notify_user(vcpu) ||
844 kvm_pmu_should_notify_user(vcpu)) {
846 run->exit_reason = KVM_EXIT_INTR;
851 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
852 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
853 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
854 run->fail_entry.cpu = smp_processor_id();
859 return kvm_request_pending(vcpu) ||
860 xfer_to_guest_mode_work_pending();
864 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
865 * the vCPU is running.
867 * This must be noinstr as instrumentation may make use of RCU, and this is not
868 * safe during the EQS.
870 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
874 guest_state_enter_irqoff();
875 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
876 guest_state_exit_irqoff();
882 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
883 * @vcpu: The VCPU pointer
885 * This function is called through the VCPU_RUN ioctl called from user space. It
886 * will execute VM code in a loop until the time slice for the process is used
887 * or some emulation is needed from user space in which case the function will
888 * return with return value 0 and with the kvm_run structure filled in with the
889 * required data for the requested emulation.
891 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
893 struct kvm_run *run = vcpu->run;
896 if (run->exit_reason == KVM_EXIT_MMIO) {
897 ret = kvm_handle_mmio_return(vcpu);
904 if (run->immediate_exit) {
909 kvm_sigset_activate(vcpu);
912 run->exit_reason = KVM_EXIT_UNKNOWN;
916 * Check conditions before entering the guest
918 ret = xfer_to_guest_mode_handle_work(vcpu);
923 ret = check_vcpu_requests(vcpu);
926 * Preparing the interrupts to be injected also
927 * involves poking the GIC, which must be done in a
928 * non-preemptible context.
933 * The VMID allocator only tracks active VMIDs per
934 * physical CPU, and therefore the VMID allocated may not be
935 * preserved on VMID roll-over if the task was preempted,
936 * making a thread's VMID inactive. So we need to call
937 * kvm_arm_vmid_update() in non-premptible context.
939 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
941 kvm_pmu_flush_hwstate(vcpu);
945 kvm_vgic_flush_hwstate(vcpu);
947 kvm_pmu_update_vcpu_events(vcpu);
950 * Ensure we set mode to IN_GUEST_MODE after we disable
951 * interrupts and before the final VCPU requests check.
952 * See the comment in kvm_vcpu_exiting_guest_mode() and
953 * Documentation/virt/kvm/vcpu-requests.rst
955 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
957 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
958 vcpu->mode = OUTSIDE_GUEST_MODE;
959 isb(); /* Ensure work in x_flush_hwstate is committed */
960 kvm_pmu_sync_hwstate(vcpu);
961 if (static_branch_unlikely(&userspace_irqchip_in_use))
962 kvm_timer_sync_user(vcpu);
963 kvm_vgic_sync_hwstate(vcpu);
969 kvm_arm_setup_debug(vcpu);
970 kvm_arch_vcpu_ctxflush_fp(vcpu);
972 /**************************************************************
975 trace_kvm_entry(*vcpu_pc(vcpu));
976 guest_timing_enter_irqoff();
978 ret = kvm_arm_vcpu_enter_exit(vcpu);
980 vcpu->mode = OUTSIDE_GUEST_MODE;
984 *************************************************************/
986 kvm_arm_clear_debug(vcpu);
989 * We must sync the PMU state before the vgic state so
990 * that the vgic can properly sample the updated state of the
993 kvm_pmu_sync_hwstate(vcpu);
996 * Sync the vgic state before syncing the timer state because
997 * the timer code needs to know if the virtual timer
998 * interrupts are active.
1000 kvm_vgic_sync_hwstate(vcpu);
1003 * Sync the timer hardware state before enabling interrupts as
1004 * we don't want vtimer interrupts to race with syncing the
1005 * timer virtual interrupt state.
1007 if (static_branch_unlikely(&userspace_irqchip_in_use))
1008 kvm_timer_sync_user(vcpu);
1010 kvm_arch_vcpu_ctxsync_fp(vcpu);
1013 * We must ensure that any pending interrupts are taken before
1014 * we exit guest timing so that timer ticks are accounted as
1015 * guest time. Transiently unmask interrupts so that any
1016 * pending interrupts are taken.
1018 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1019 * context synchronization event) is necessary to ensure that
1020 * pending interrupts are taken.
1022 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1025 local_irq_disable();
1028 guest_timing_exit_irqoff();
1032 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1034 /* Exit types that need handling before we can be preempted */
1035 handle_exit_early(vcpu, ret);
1040 * The ARMv8 architecture doesn't give the hypervisor
1041 * a mechanism to prevent a guest from dropping to AArch32 EL0
1042 * if implemented by the CPU. If we spot the guest in such
1043 * state and that we decided it wasn't supposed to do so (like
1044 * with the asymmetric AArch32 case), return to userspace with
1047 if (vcpu_mode_is_bad_32bit(vcpu)) {
1049 * As we have caught the guest red-handed, decide that
1050 * it isn't fit for purpose anymore by making the vcpu
1051 * invalid. The VMM can try and fix it by issuing a
1052 * KVM_ARM_VCPU_INIT if it really wants to.
1054 vcpu->arch.target = -1;
1055 ret = ARM_EXCEPTION_IL;
1058 ret = handle_exit(vcpu, ret);
1061 /* Tell userspace about in-kernel device output levels */
1062 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1063 kvm_timer_update_run(vcpu);
1064 kvm_pmu_update_run(vcpu);
1067 kvm_sigset_deactivate(vcpu);
1071 * In the unlikely event that we are returning to userspace
1072 * with pending exceptions or PC adjustment, commit these
1073 * adjustments in order to give userspace a consistent view of
1074 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1075 * being preempt-safe on VHE.
1077 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1078 vcpu_get_flag(vcpu, INCREMENT_PC)))
1079 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1085 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1091 if (number == KVM_ARM_IRQ_CPU_IRQ)
1092 bit_index = __ffs(HCR_VI);
1093 else /* KVM_ARM_IRQ_CPU_FIQ */
1094 bit_index = __ffs(HCR_VF);
1096 hcr = vcpu_hcr(vcpu);
1098 set = test_and_set_bit(bit_index, hcr);
1100 set = test_and_clear_bit(bit_index, hcr);
1103 * If we didn't change anything, no need to wake up or kick other CPUs
1109 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1110 * trigger a world-switch round on the running physical CPU to set the
1111 * virtual IRQ/FIQ fields in the HCR appropriately.
1113 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1114 kvm_vcpu_kick(vcpu);
1119 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1122 u32 irq = irq_level->irq;
1123 unsigned int irq_type, vcpu_idx, irq_num;
1124 int nrcpus = atomic_read(&kvm->online_vcpus);
1125 struct kvm_vcpu *vcpu = NULL;
1126 bool level = irq_level->level;
1128 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1129 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1130 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1131 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1133 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1136 case KVM_ARM_IRQ_TYPE_CPU:
1137 if (irqchip_in_kernel(kvm))
1140 if (vcpu_idx >= nrcpus)
1143 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1147 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1150 return vcpu_interrupt_line(vcpu, irq_num, level);
1151 case KVM_ARM_IRQ_TYPE_PPI:
1152 if (!irqchip_in_kernel(kvm))
1155 if (vcpu_idx >= nrcpus)
1158 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1162 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1165 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1166 case KVM_ARM_IRQ_TYPE_SPI:
1167 if (!irqchip_in_kernel(kvm))
1170 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1173 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1179 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1180 const struct kvm_vcpu_init *init)
1182 unsigned long features = init->features[0];
1185 if (features & ~KVM_VCPU_VALID_FEATURES)
1188 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1189 if (init->features[i])
1193 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1196 if (!cpus_have_const_cap(ARM64_HAS_32BIT_EL1))
1199 /* MTE is incompatible with AArch32 */
1200 if (kvm_has_mte(vcpu->kvm))
1203 /* NV is incompatible with AArch32 */
1204 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1210 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1211 const struct kvm_vcpu_init *init)
1213 unsigned long features = init->features[0];
1215 return !bitmap_equal(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES) ||
1216 vcpu->arch.target != init->target;
1219 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1220 const struct kvm_vcpu_init *init)
1222 unsigned long features = init->features[0];
1223 struct kvm *kvm = vcpu->kvm;
1226 mutex_lock(&kvm->arch.config_lock);
1228 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1229 !bitmap_equal(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES))
1232 vcpu->arch.target = init->target;
1233 bitmap_copy(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES);
1235 /* Now we know what it is, we can reset it. */
1236 ret = kvm_reset_vcpu(vcpu);
1238 vcpu->arch.target = -1;
1239 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1243 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1244 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1247 mutex_unlock(&kvm->arch.config_lock);
1251 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1252 const struct kvm_vcpu_init *init)
1256 if (init->target != kvm_target_cpu())
1259 ret = kvm_vcpu_init_check_features(vcpu, init);
1263 if (vcpu->arch.target == -1)
1264 return __kvm_vcpu_set_target(vcpu, init);
1266 if (kvm_vcpu_init_changed(vcpu, init))
1269 return kvm_reset_vcpu(vcpu);
1272 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1273 struct kvm_vcpu_init *init)
1275 bool power_off = false;
1279 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1280 * reflecting it in the finalized feature set, thus limiting its scope
1281 * to a single KVM_ARM_VCPU_INIT call.
1283 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1284 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1288 ret = kvm_vcpu_set_target(vcpu, init);
1293 * Ensure a rebooted VM will fault in RAM pages and detect if the
1294 * guest MMU is turned off and flush the caches as needed.
1296 * S2FWB enforces all memory accesses to RAM being cacheable,
1297 * ensuring that the data side is always coherent. We still
1298 * need to invalidate the I-cache though, as FWB does *not*
1299 * imply CTR_EL0.DIC.
1301 if (vcpu_has_run_once(vcpu)) {
1302 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1303 stage2_unmap_vm(vcpu->kvm);
1305 icache_inval_all_pou();
1308 vcpu_reset_hcr(vcpu);
1309 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1312 * Handle the "start in power-off" case.
1314 spin_lock(&vcpu->arch.mp_state_lock);
1317 __kvm_arm_vcpu_power_off(vcpu);
1319 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1321 spin_unlock(&vcpu->arch.mp_state_lock);
1326 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1327 struct kvm_device_attr *attr)
1331 switch (attr->group) {
1333 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1340 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1341 struct kvm_device_attr *attr)
1345 switch (attr->group) {
1347 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1354 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1355 struct kvm_device_attr *attr)
1359 switch (attr->group) {
1361 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1368 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1369 struct kvm_vcpu_events *events)
1371 memset(events, 0, sizeof(*events));
1373 return __kvm_arm_vcpu_get_events(vcpu, events);
1376 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1377 struct kvm_vcpu_events *events)
1381 /* check whether the reserved field is zero */
1382 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1383 if (events->reserved[i])
1386 /* check whether the pad field is zero */
1387 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1388 if (events->exception.pad[i])
1391 return __kvm_arm_vcpu_set_events(vcpu, events);
1394 long kvm_arch_vcpu_ioctl(struct file *filp,
1395 unsigned int ioctl, unsigned long arg)
1397 struct kvm_vcpu *vcpu = filp->private_data;
1398 void __user *argp = (void __user *)arg;
1399 struct kvm_device_attr attr;
1403 case KVM_ARM_VCPU_INIT: {
1404 struct kvm_vcpu_init init;
1407 if (copy_from_user(&init, argp, sizeof(init)))
1410 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1413 case KVM_SET_ONE_REG:
1414 case KVM_GET_ONE_REG: {
1415 struct kvm_one_reg reg;
1418 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1422 if (copy_from_user(®, argp, sizeof(reg)))
1426 * We could owe a reset due to PSCI. Handle the pending reset
1427 * here to ensure userspace register accesses are ordered after
1430 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1431 kvm_reset_vcpu(vcpu);
1433 if (ioctl == KVM_SET_ONE_REG)
1434 r = kvm_arm_set_reg(vcpu, ®);
1436 r = kvm_arm_get_reg(vcpu, ®);
1439 case KVM_GET_REG_LIST: {
1440 struct kvm_reg_list __user *user_list = argp;
1441 struct kvm_reg_list reg_list;
1445 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1449 if (!kvm_arm_vcpu_is_finalized(vcpu))
1453 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1456 reg_list.n = kvm_arm_num_regs(vcpu);
1457 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1462 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1465 case KVM_SET_DEVICE_ATTR: {
1467 if (copy_from_user(&attr, argp, sizeof(attr)))
1469 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1472 case KVM_GET_DEVICE_ATTR: {
1474 if (copy_from_user(&attr, argp, sizeof(attr)))
1476 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1479 case KVM_HAS_DEVICE_ATTR: {
1481 if (copy_from_user(&attr, argp, sizeof(attr)))
1483 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1486 case KVM_GET_VCPU_EVENTS: {
1487 struct kvm_vcpu_events events;
1489 if (kvm_arm_vcpu_get_events(vcpu, &events))
1492 if (copy_to_user(argp, &events, sizeof(events)))
1497 case KVM_SET_VCPU_EVENTS: {
1498 struct kvm_vcpu_events events;
1500 if (copy_from_user(&events, argp, sizeof(events)))
1503 return kvm_arm_vcpu_set_events(vcpu, &events);
1505 case KVM_ARM_VCPU_FINALIZE: {
1508 if (!kvm_vcpu_initialized(vcpu))
1511 if (get_user(what, (const int __user *)argp))
1514 return kvm_arm_vcpu_finalize(vcpu, what);
1523 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1528 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1529 const struct kvm_memory_slot *memslot)
1531 kvm_flush_remote_tlbs(kvm);
1534 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1535 struct kvm_arm_device_addr *dev_addr)
1537 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1538 case KVM_ARM_DEVICE_VGIC_V2:
1541 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1547 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1549 switch (attr->group) {
1550 case KVM_ARM_VM_SMCCC_CTRL:
1551 return kvm_vm_smccc_has_attr(kvm, attr);
1557 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1559 switch (attr->group) {
1560 case KVM_ARM_VM_SMCCC_CTRL:
1561 return kvm_vm_smccc_set_attr(kvm, attr);
1567 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1569 struct kvm *kvm = filp->private_data;
1570 void __user *argp = (void __user *)arg;
1571 struct kvm_device_attr attr;
1574 case KVM_CREATE_IRQCHIP: {
1578 mutex_lock(&kvm->lock);
1579 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1580 mutex_unlock(&kvm->lock);
1583 case KVM_ARM_SET_DEVICE_ADDR: {
1584 struct kvm_arm_device_addr dev_addr;
1586 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1588 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1590 case KVM_ARM_PREFERRED_TARGET: {
1591 struct kvm_vcpu_init init;
1593 kvm_vcpu_preferred_target(&init);
1595 if (copy_to_user(argp, &init, sizeof(init)))
1600 case KVM_ARM_MTE_COPY_TAGS: {
1601 struct kvm_arm_copy_mte_tags copy_tags;
1603 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1605 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1607 case KVM_ARM_SET_COUNTER_OFFSET: {
1608 struct kvm_arm_counter_offset offset;
1610 if (copy_from_user(&offset, argp, sizeof(offset)))
1612 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1614 case KVM_HAS_DEVICE_ATTR: {
1615 if (copy_from_user(&attr, argp, sizeof(attr)))
1618 return kvm_vm_has_attr(kvm, &attr);
1620 case KVM_SET_DEVICE_ATTR: {
1621 if (copy_from_user(&attr, argp, sizeof(attr)))
1624 return kvm_vm_set_attr(kvm, &attr);
1631 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1632 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1634 struct kvm_vcpu *tmp_vcpu;
1636 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1637 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1638 mutex_unlock(&tmp_vcpu->mutex);
1642 void unlock_all_vcpus(struct kvm *kvm)
1644 lockdep_assert_held(&kvm->lock);
1646 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1649 /* Returns true if all vcpus were locked, false otherwise */
1650 bool lock_all_vcpus(struct kvm *kvm)
1652 struct kvm_vcpu *tmp_vcpu;
1655 lockdep_assert_held(&kvm->lock);
1658 * Any time a vcpu is in an ioctl (including running), the
1659 * core KVM code tries to grab the vcpu->mutex.
1661 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1662 * other VCPUs can fiddle with the state while we access it.
1664 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1665 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1666 unlock_vcpus(kvm, c - 1);
1674 static unsigned long nvhe_percpu_size(void)
1676 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1677 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1680 static unsigned long nvhe_percpu_order(void)
1682 unsigned long size = nvhe_percpu_size();
1684 return size ? get_order(size) : 0;
1687 /* A lookup table holding the hypervisor VA for each vector slot */
1688 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1690 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1692 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1695 static int kvm_init_vector_slots(void)
1700 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1701 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1703 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1704 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1706 if (kvm_system_needs_idmapped_vectors() &&
1707 !is_protected_kvm_enabled()) {
1708 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1709 __BP_HARDEN_HYP_VECS_SZ, &base);
1714 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1715 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1719 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1721 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1725 * Calculate the raw per-cpu offset without a translation from the
1726 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1727 * so that we can use adr_l to access per-cpu variables in EL2.
1728 * Also drop the KASAN tag which gets in the way...
1730 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1731 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1733 params->mair_el2 = read_sysreg(mair_el1);
1735 tcr = read_sysreg(tcr_el1);
1736 if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1737 tcr |= TCR_EPD1_MASK;
1739 tcr &= TCR_EL2_MASK;
1740 tcr |= TCR_EL2_RES1;
1742 tcr &= ~TCR_T0SZ_MASK;
1743 tcr |= TCR_T0SZ(hyp_va_bits);
1744 params->tcr_el2 = tcr;
1746 params->pgd_pa = kvm_mmu_get_httbr();
1747 if (is_protected_kvm_enabled())
1748 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1750 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1751 if (cpus_have_final_cap(ARM64_KVM_HVHE))
1752 params->hcr_el2 |= HCR_E2H;
1753 params->vttbr = params->vtcr = 0;
1756 * Flush the init params from the data cache because the struct will
1757 * be read while the MMU is off.
1759 kvm_flush_dcache_to_poc(params, sizeof(*params));
1762 static void hyp_install_host_vector(void)
1764 struct kvm_nvhe_init_params *params;
1765 struct arm_smccc_res res;
1767 /* Switch from the HYP stub to our own HYP init vector */
1768 __hyp_set_vectors(kvm_get_idmap_vector());
1771 * Call initialization code, and switch to the full blown HYP code.
1772 * If the cpucaps haven't been finalized yet, something has gone very
1773 * wrong, and hyp will crash and burn when it uses any
1774 * cpus_have_const_cap() wrapper.
1776 BUG_ON(!system_capabilities_finalized());
1777 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1778 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1779 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1782 static void cpu_init_hyp_mode(void)
1784 hyp_install_host_vector();
1787 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1790 if (this_cpu_has_cap(ARM64_SSBS) &&
1791 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1792 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1796 static void cpu_hyp_reset(void)
1798 if (!is_kernel_in_hyp_mode())
1799 __hyp_reset_vectors();
1803 * EL2 vectors can be mapped and rerouted in a number of ways,
1804 * depending on the kernel configuration and CPU present:
1806 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1807 * placed in one of the vector slots, which is executed before jumping
1808 * to the real vectors.
1810 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1811 * containing the hardening sequence is mapped next to the idmap page,
1812 * and executed before jumping to the real vectors.
1814 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1815 * empty slot is selected, mapped next to the idmap page, and
1816 * executed before jumping to the real vectors.
1818 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1819 * VHE, as we don't have hypervisor-specific mappings. If the system
1820 * is VHE and yet selects this capability, it will be ignored.
1822 static void cpu_set_hyp_vector(void)
1824 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1825 void *vector = hyp_spectre_vector_selector[data->slot];
1827 if (!is_protected_kvm_enabled())
1828 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1830 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1833 static void cpu_hyp_init_context(void)
1835 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1837 if (!is_kernel_in_hyp_mode())
1838 cpu_init_hyp_mode();
1841 static void cpu_hyp_init_features(void)
1843 cpu_set_hyp_vector();
1844 kvm_arm_init_debug();
1846 if (is_kernel_in_hyp_mode())
1847 kvm_timer_init_vhe();
1850 kvm_vgic_init_cpu_hardware();
1853 static void cpu_hyp_reinit(void)
1856 cpu_hyp_init_context();
1857 cpu_hyp_init_features();
1860 static void _kvm_arch_hardware_enable(void *discard)
1862 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1864 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1868 int kvm_arch_hardware_enable(void)
1870 int was_enabled = __this_cpu_read(kvm_arm_hardware_enabled);
1872 _kvm_arch_hardware_enable(NULL);
1882 static void _kvm_arch_hardware_disable(void *discard)
1884 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1886 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1890 void kvm_arch_hardware_disable(void)
1892 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1893 kvm_timer_cpu_down();
1894 kvm_vgic_cpu_down();
1897 if (!is_protected_kvm_enabled())
1898 _kvm_arch_hardware_disable(NULL);
1901 #ifdef CONFIG_CPU_PM
1902 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1907 * kvm_arm_hardware_enabled is left with its old value over
1908 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1913 if (__this_cpu_read(kvm_arm_hardware_enabled))
1915 * don't update kvm_arm_hardware_enabled here
1916 * so that the hardware will be re-enabled
1917 * when we resume. See below.
1922 case CPU_PM_ENTER_FAILED:
1924 if (__this_cpu_read(kvm_arm_hardware_enabled))
1925 /* The hardware was enabled before suspend. */
1935 static struct notifier_block hyp_init_cpu_pm_nb = {
1936 .notifier_call = hyp_init_cpu_pm_notifier,
1939 static void __init hyp_cpu_pm_init(void)
1941 if (!is_protected_kvm_enabled())
1942 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1944 static void __init hyp_cpu_pm_exit(void)
1946 if (!is_protected_kvm_enabled())
1947 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1950 static inline void __init hyp_cpu_pm_init(void)
1953 static inline void __init hyp_cpu_pm_exit(void)
1958 static void __init init_cpu_logical_map(void)
1963 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1964 * Only copy the set of online CPUs whose features have been checked
1965 * against the finalized system capabilities. The hypervisor will not
1966 * allow any other CPUs from the `possible` set to boot.
1968 for_each_online_cpu(cpu)
1969 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1972 #define init_psci_0_1_impl_state(config, what) \
1973 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1975 static bool __init init_psci_relay(void)
1978 * If PSCI has not been initialized, protected KVM cannot install
1979 * itself on newly booted CPUs.
1981 if (!psci_ops.get_version) {
1982 kvm_err("Cannot initialize protected mode without PSCI\n");
1986 kvm_host_psci_config.version = psci_ops.get_version();
1987 kvm_host_psci_config.smccc_version = arm_smccc_get_version();
1989 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1990 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1991 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1992 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1993 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1994 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1999 static int __init init_subsystems(void)
2004 * Enable hardware so that subsystem initialisation can access EL2.
2006 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
2009 * Register CPU lower-power notifier
2014 * Init HYP view of VGIC
2016 err = kvm_vgic_hyp_init();
2019 vgic_present = true;
2023 vgic_present = false;
2031 * Init HYP architected timer support
2033 err = kvm_timer_hyp_init(vgic_present);
2037 kvm_register_perf_callbacks(NULL);
2043 if (err || !is_protected_kvm_enabled())
2044 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
2049 static void __init teardown_subsystems(void)
2051 kvm_unregister_perf_callbacks();
2055 static void __init teardown_hyp_mode(void)
2060 for_each_possible_cpu(cpu) {
2061 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2062 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2066 static int __init do_pkvm_init(u32 hyp_va_bits)
2068 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2072 cpu_hyp_init_context();
2073 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2074 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2076 cpu_hyp_init_features();
2079 * The stub hypercalls are now disabled, so set our local flag to
2080 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2082 __this_cpu_write(kvm_arm_hardware_enabled, 1);
2088 static u64 get_hyp_id_aa64pfr0_el1(void)
2091 * Track whether the system isn't affected by spectre/meltdown in the
2092 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2093 * Although this is per-CPU, we make it global for simplicity, e.g., not
2094 * to have to worry about vcpu migration.
2096 * Unlike for non-protected VMs, userspace cannot override this for
2099 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2101 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2102 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2104 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2105 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2106 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2107 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2112 static void kvm_hyp_init_symbols(void)
2114 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2115 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2116 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2117 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2118 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2119 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2120 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2121 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2122 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2123 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2124 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2127 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2129 void *addr = phys_to_virt(hyp_mem_base);
2132 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2136 ret = do_pkvm_init(hyp_va_bits);
2145 static void pkvm_hyp_init_ptrauth(void)
2147 struct kvm_cpu_context *hyp_ctxt;
2150 for_each_possible_cpu(cpu) {
2151 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2152 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2153 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2154 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2155 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2156 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2157 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2158 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2159 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2160 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2161 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2165 /* Inits Hyp-mode on all online CPUs */
2166 static int __init init_hyp_mode(void)
2173 * The protected Hyp-mode cannot be initialized if the memory pool
2174 * allocation has failed.
2176 if (is_protected_kvm_enabled() && !hyp_mem_base)
2180 * Allocate Hyp PGD and setup Hyp identity mapping
2182 err = kvm_mmu_init(&hyp_va_bits);
2187 * Allocate stack pages for Hypervisor-mode
2189 for_each_possible_cpu(cpu) {
2190 unsigned long stack_page;
2192 stack_page = __get_free_page(GFP_KERNEL);
2198 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2202 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2204 for_each_possible_cpu(cpu) {
2208 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2214 page_addr = page_address(page);
2215 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2216 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2220 * Map the Hyp-code called directly from the host
2222 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2223 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2225 kvm_err("Cannot map world-switch code\n");
2229 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2230 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2232 kvm_err("Cannot map .hyp.rodata section\n");
2236 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2237 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2239 kvm_err("Cannot map rodata section\n");
2244 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2245 * section thanks to an assertion in the linker script. Map it RW and
2246 * the rest of .bss RO.
2248 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2249 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2251 kvm_err("Cannot map hyp bss section: %d\n", err);
2255 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2256 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2258 kvm_err("Cannot map bss section\n");
2263 * Map the Hyp stack pages
2265 for_each_possible_cpu(cpu) {
2266 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2267 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2268 unsigned long hyp_addr;
2271 * Allocate a contiguous HYP private VA range for the stack
2272 * and guard page. The allocation is also aligned based on
2273 * the order of its size.
2275 err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr);
2277 kvm_err("Cannot allocate hyp stack guard page\n");
2282 * Since the stack grows downwards, map the stack to the page
2283 * at the higher address and leave the lower guard page
2286 * Any valid stack address now has the PAGE_SHIFT bit as 1
2287 * and addresses corresponding to the guard page have the
2288 * PAGE_SHIFT bit as 0 - this is used for overflow detection.
2290 err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE,
2291 __pa(stack_page), PAGE_HYP);
2293 kvm_err("Cannot map hyp stack\n");
2298 * Save the stack PA in nvhe_init_params. This will be needed
2299 * to recreate the stack mapping in protected nVHE mode.
2300 * __hyp_pa() won't do the right thing there, since the stack
2301 * has been mapped in the flexible private VA space.
2303 params->stack_pa = __pa(stack_page);
2305 params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE);
2308 for_each_possible_cpu(cpu) {
2309 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2310 char *percpu_end = percpu_begin + nvhe_percpu_size();
2312 /* Map Hyp percpu pages */
2313 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2315 kvm_err("Cannot map hyp percpu region\n");
2319 /* Prepare the CPU initialization parameters */
2320 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2323 kvm_hyp_init_symbols();
2325 if (is_protected_kvm_enabled()) {
2326 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2327 cpus_have_const_cap(ARM64_HAS_ADDRESS_AUTH))
2328 pkvm_hyp_init_ptrauth();
2330 init_cpu_logical_map();
2332 if (!init_psci_relay()) {
2337 err = kvm_hyp_init_protection(hyp_va_bits);
2339 kvm_err("Failed to init hyp memory protection\n");
2347 teardown_hyp_mode();
2348 kvm_err("error initializing Hyp mode: %d\n", err);
2352 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2354 struct kvm_vcpu *vcpu;
2357 mpidr &= MPIDR_HWID_BITMASK;
2358 kvm_for_each_vcpu(i, vcpu, kvm) {
2359 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2365 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2367 return irqchip_in_kernel(kvm);
2370 bool kvm_arch_has_irq_bypass(void)
2375 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2376 struct irq_bypass_producer *prod)
2378 struct kvm_kernel_irqfd *irqfd =
2379 container_of(cons, struct kvm_kernel_irqfd, consumer);
2381 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2384 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2385 struct irq_bypass_producer *prod)
2387 struct kvm_kernel_irqfd *irqfd =
2388 container_of(cons, struct kvm_kernel_irqfd, consumer);
2390 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2394 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2396 struct kvm_kernel_irqfd *irqfd =
2397 container_of(cons, struct kvm_kernel_irqfd, consumer);
2399 kvm_arm_halt_guest(irqfd->kvm);
2402 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2404 struct kvm_kernel_irqfd *irqfd =
2405 container_of(cons, struct kvm_kernel_irqfd, consumer);
2407 kvm_arm_resume_guest(irqfd->kvm);
2410 /* Initialize Hyp-mode and memory mappings on all CPUs */
2411 static __init int kvm_arm_init(void)
2416 if (!is_hyp_mode_available()) {
2417 kvm_info("HYP mode not available\n");
2421 if (kvm_get_mode() == KVM_MODE_NONE) {
2422 kvm_info("KVM disabled from command line\n");
2426 err = kvm_sys_reg_table_init();
2428 kvm_info("Error initializing system register tables");
2432 in_hyp_mode = is_kernel_in_hyp_mode();
2434 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2435 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2436 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2437 "Only trusted guests should be used on this system.\n");
2439 err = kvm_set_ipa_limit();
2443 err = kvm_arm_init_sve();
2447 err = kvm_arm_vmid_alloc_init();
2449 kvm_err("Failed to initialize VMID allocator.\n");
2454 err = init_hyp_mode();
2459 err = kvm_init_vector_slots();
2461 kvm_err("Cannot initialise vector slots\n");
2465 err = init_subsystems();
2469 if (is_protected_kvm_enabled()) {
2470 kvm_info("Protected nVHE mode initialized successfully\n");
2471 } else if (in_hyp_mode) {
2472 kvm_info("VHE mode initialized successfully\n");
2474 kvm_info("Hyp mode initialized successfully\n");
2478 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2479 * hypervisor protection is finalized.
2481 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2488 teardown_subsystems();
2491 teardown_hyp_mode();
2493 kvm_arm_vmid_alloc_free();
2497 static int __init early_kvm_mode_cfg(char *arg)
2502 if (strcmp(arg, "none") == 0) {
2503 kvm_mode = KVM_MODE_NONE;
2507 if (!is_hyp_mode_available()) {
2508 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2512 if (strcmp(arg, "protected") == 0) {
2513 if (!is_kernel_in_hyp_mode())
2514 kvm_mode = KVM_MODE_PROTECTED;
2516 pr_warn_once("Protected KVM not available with VHE\n");
2521 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2522 kvm_mode = KVM_MODE_DEFAULT;
2526 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2527 kvm_mode = KVM_MODE_NV;
2533 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2535 enum kvm_mode kvm_get_mode(void)
2540 module_init(kvm_arm_init);