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;
193 void kvm_arch_create_vm_debugfs(struct kvm *kvm)
195 kvm_sys_regs_create_debugfs(kvm);
199 * kvm_arch_destroy_vm - destroy the VM data structure
200 * @kvm: pointer to the KVM struct
202 void kvm_arch_destroy_vm(struct kvm *kvm)
204 bitmap_free(kvm->arch.pmu_filter);
205 free_cpumask_var(kvm->arch.supported_cpus);
207 kvm_vgic_destroy(kvm);
209 if (is_protected_kvm_enabled())
210 pkvm_destroy_hyp_vm(kvm);
212 kfree(kvm->arch.mpidr_data);
213 kfree(kvm->arch.sysreg_masks);
214 kvm_destroy_vcpus(kvm);
216 kvm_unshare_hyp(kvm, kvm + 1);
218 kvm_arm_teardown_hypercalls(kvm);
221 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
225 case KVM_CAP_IRQCHIP:
228 case KVM_CAP_IOEVENTFD:
229 case KVM_CAP_USER_MEMORY:
230 case KVM_CAP_SYNC_MMU:
231 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
232 case KVM_CAP_ONE_REG:
233 case KVM_CAP_ARM_PSCI:
234 case KVM_CAP_ARM_PSCI_0_2:
235 case KVM_CAP_READONLY_MEM:
236 case KVM_CAP_MP_STATE:
237 case KVM_CAP_IMMEDIATE_EXIT:
238 case KVM_CAP_VCPU_EVENTS:
239 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
240 case KVM_CAP_ARM_NISV_TO_USER:
241 case KVM_CAP_ARM_INJECT_EXT_DABT:
242 case KVM_CAP_SET_GUEST_DEBUG:
243 case KVM_CAP_VCPU_ATTRIBUTES:
244 case KVM_CAP_PTP_KVM:
245 case KVM_CAP_ARM_SYSTEM_SUSPEND:
246 case KVM_CAP_IRQFD_RESAMPLE:
247 case KVM_CAP_COUNTER_OFFSET:
250 case KVM_CAP_SET_GUEST_DEBUG2:
251 return KVM_GUESTDBG_VALID_MASK;
252 case KVM_CAP_ARM_SET_DEVICE_ADDR:
255 case KVM_CAP_NR_VCPUS:
257 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
258 * architectures, as it does not always bound it to
259 * KVM_CAP_MAX_VCPUS. It should not matter much because
260 * this is just an advisory value.
262 r = min_t(unsigned int, num_online_cpus(),
263 kvm_arm_default_max_vcpus());
265 case KVM_CAP_MAX_VCPUS:
266 case KVM_CAP_MAX_VCPU_ID:
270 r = kvm_arm_default_max_vcpus();
272 case KVM_CAP_MSI_DEVID:
276 r = kvm->arch.vgic.msis_require_devid;
278 case KVM_CAP_ARM_USER_IRQ:
280 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
281 * (bump this number if adding more devices)
285 case KVM_CAP_ARM_MTE:
286 r = system_supports_mte();
288 case KVM_CAP_STEAL_TIME:
289 r = kvm_arm_pvtime_supported();
291 case KVM_CAP_ARM_EL1_32BIT:
292 r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
294 case KVM_CAP_GUEST_DEBUG_HW_BPS:
297 case KVM_CAP_GUEST_DEBUG_HW_WPS:
300 case KVM_CAP_ARM_PMU_V3:
301 r = kvm_arm_support_pmu_v3();
303 case KVM_CAP_ARM_INJECT_SERROR_ESR:
304 r = cpus_have_final_cap(ARM64_HAS_RAS_EXTN);
306 case KVM_CAP_ARM_VM_IPA_SIZE:
307 r = get_kvm_ipa_limit();
309 case KVM_CAP_ARM_SVE:
310 r = system_supports_sve();
312 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
313 case KVM_CAP_ARM_PTRAUTH_GENERIC:
314 r = system_has_full_ptr_auth();
316 case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
318 r = kvm->arch.mmu.split_page_chunk_size;
320 r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
322 case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
323 r = kvm_supported_block_sizes();
325 case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
335 long kvm_arch_dev_ioctl(struct file *filp,
336 unsigned int ioctl, unsigned long arg)
341 struct kvm *kvm_arch_alloc_vm(void)
343 size_t sz = sizeof(struct kvm);
346 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
348 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
351 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
353 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
356 if (id >= kvm->max_vcpus)
362 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
366 spin_lock_init(&vcpu->arch.mp_state_lock);
368 #ifdef CONFIG_LOCKDEP
369 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
370 mutex_lock(&vcpu->mutex);
371 mutex_lock(&vcpu->kvm->arch.config_lock);
372 mutex_unlock(&vcpu->kvm->arch.config_lock);
373 mutex_unlock(&vcpu->mutex);
376 /* Force users to call KVM_ARM_VCPU_INIT */
377 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
379 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
382 * Default value for the FP state, will be overloaded at load
383 * time if we support FP (pretty likely)
385 vcpu->arch.fp_state = FP_STATE_FREE;
387 /* Set up the timer */
388 kvm_timer_vcpu_init(vcpu);
390 kvm_pmu_vcpu_init(vcpu);
392 kvm_arm_reset_debug_ptr(vcpu);
394 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
396 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
398 err = kvm_vgic_vcpu_init(vcpu);
402 return kvm_share_hyp(vcpu, vcpu + 1);
405 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
409 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
411 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
412 static_branch_dec(&userspace_irqchip_in_use);
414 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
415 kvm_timer_vcpu_terminate(vcpu);
416 kvm_pmu_vcpu_destroy(vcpu);
417 kvm_vgic_vcpu_destroy(vcpu);
418 kvm_arm_vcpu_destroy(vcpu);
421 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
426 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
431 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
433 struct kvm_s2_mmu *mmu;
436 mmu = vcpu->arch.hw_mmu;
437 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
440 * We guarantee that both TLBs and I-cache are private to each
441 * vcpu. If detecting that a vcpu from the same VM has
442 * previously run on the same physical CPU, call into the
443 * hypervisor code to nuke the relevant contexts.
445 * We might get preempted before the vCPU actually runs, but
446 * over-invalidation doesn't affect correctness.
448 if (*last_ran != vcpu->vcpu_idx) {
449 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
450 *last_ran = vcpu->vcpu_idx;
456 kvm_timer_vcpu_load(vcpu);
458 kvm_vcpu_load_vhe(vcpu);
459 kvm_arch_vcpu_load_fp(vcpu);
460 kvm_vcpu_pmu_restore_guest(vcpu);
461 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
462 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
464 if (single_task_running())
465 vcpu_clear_wfx_traps(vcpu);
467 vcpu_set_wfx_traps(vcpu);
469 if (vcpu_has_ptrauth(vcpu))
470 vcpu_ptrauth_disable(vcpu);
471 kvm_arch_vcpu_load_debug_state_flags(vcpu);
473 if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
474 vcpu_set_on_unsupported_cpu(vcpu);
477 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
479 kvm_arch_vcpu_put_debug_state_flags(vcpu);
480 kvm_arch_vcpu_put_fp(vcpu);
482 kvm_vcpu_put_vhe(vcpu);
483 kvm_timer_vcpu_put(vcpu);
485 kvm_vcpu_pmu_restore_host(vcpu);
486 kvm_arm_vmid_clear_active();
488 vcpu_clear_on_unsupported_cpu(vcpu);
492 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
494 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
495 kvm_make_request(KVM_REQ_SLEEP, vcpu);
499 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
501 spin_lock(&vcpu->arch.mp_state_lock);
502 __kvm_arm_vcpu_power_off(vcpu);
503 spin_unlock(&vcpu->arch.mp_state_lock);
506 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
508 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
511 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
513 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
514 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
518 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
520 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
523 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
524 struct kvm_mp_state *mp_state)
526 *mp_state = READ_ONCE(vcpu->arch.mp_state);
531 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
532 struct kvm_mp_state *mp_state)
536 spin_lock(&vcpu->arch.mp_state_lock);
538 switch (mp_state->mp_state) {
539 case KVM_MP_STATE_RUNNABLE:
540 WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
542 case KVM_MP_STATE_STOPPED:
543 __kvm_arm_vcpu_power_off(vcpu);
545 case KVM_MP_STATE_SUSPENDED:
546 kvm_arm_vcpu_suspend(vcpu);
552 spin_unlock(&vcpu->arch.mp_state_lock);
558 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
559 * @v: The VCPU pointer
561 * If the guest CPU is not waiting for interrupts or an interrupt line is
562 * asserted, the CPU is by definition runnable.
564 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
566 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
567 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
568 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
571 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
573 return vcpu_mode_priv(vcpu);
576 #ifdef CONFIG_GUEST_PERF_EVENTS
577 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
579 return *vcpu_pc(vcpu);
583 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
585 return vcpu_get_flag(vcpu, VCPU_INITIALIZED);
588 static void kvm_init_mpidr_data(struct kvm *kvm)
590 struct kvm_mpidr_data *data = NULL;
591 unsigned long c, mask, nr_entries;
592 u64 aff_set = 0, aff_clr = ~0UL;
593 struct kvm_vcpu *vcpu;
595 mutex_lock(&kvm->arch.config_lock);
597 if (kvm->arch.mpidr_data || atomic_read(&kvm->online_vcpus) == 1)
600 kvm_for_each_vcpu(c, vcpu, kvm) {
601 u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
607 * A significant bit can be either 0 or 1, and will only appear in
608 * aff_set. Use aff_clr to weed out the useless stuff.
610 mask = aff_set ^ aff_clr;
611 nr_entries = BIT_ULL(hweight_long(mask));
614 * Don't let userspace fool us. If we need more than a single page
615 * to describe the compressed MPIDR array, just fall back to the
616 * iterative method. Single vcpu VMs do not need this either.
618 if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
619 data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
625 data->mpidr_mask = mask;
627 kvm_for_each_vcpu(c, vcpu, kvm) {
628 u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
629 u16 index = kvm_mpidr_index(data, aff);
631 data->cmpidr_to_idx[index] = c;
634 kvm->arch.mpidr_data = data;
636 mutex_unlock(&kvm->arch.config_lock);
640 * Handle both the initialisation that is being done when the vcpu is
641 * run for the first time, as well as the updates that must be
642 * performed each time we get a new thread dealing with this vcpu.
644 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
646 struct kvm *kvm = vcpu->kvm;
649 if (!kvm_vcpu_initialized(vcpu))
652 if (!kvm_arm_vcpu_is_finalized(vcpu))
655 ret = kvm_arch_vcpu_run_map_fp(vcpu);
659 if (likely(vcpu_has_run_once(vcpu)))
662 kvm_init_mpidr_data(kvm);
664 kvm_arm_vcpu_init_debug(vcpu);
666 if (likely(irqchip_in_kernel(kvm))) {
668 * Map the VGIC hardware resources before running a vcpu the
669 * first time on this VM.
671 ret = kvm_vgic_map_resources(kvm);
676 if (vcpu_has_nv(vcpu)) {
677 ret = kvm_init_nv_sysregs(vcpu->kvm);
683 * This needs to happen after NV has imposed its own restrictions on
686 kvm_init_sysreg(vcpu);
688 ret = kvm_timer_enable(vcpu);
692 ret = kvm_arm_pmu_v3_enable(vcpu);
696 if (is_protected_kvm_enabled()) {
697 ret = pkvm_create_hyp_vm(kvm);
702 if (!irqchip_in_kernel(kvm)) {
704 * Tell the rest of the code that there are userspace irqchip
707 static_branch_inc(&userspace_irqchip_in_use);
711 * Initialize traps for protected VMs.
712 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
713 * the code is in place for first run initialization at EL2.
715 if (kvm_vm_is_protected(kvm))
716 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
718 mutex_lock(&kvm->arch.config_lock);
719 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
720 mutex_unlock(&kvm->arch.config_lock);
725 bool kvm_arch_intc_initialized(struct kvm *kvm)
727 return vgic_initialized(kvm);
730 void kvm_arm_halt_guest(struct kvm *kvm)
733 struct kvm_vcpu *vcpu;
735 kvm_for_each_vcpu(i, vcpu, kvm)
736 vcpu->arch.pause = true;
737 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
740 void kvm_arm_resume_guest(struct kvm *kvm)
743 struct kvm_vcpu *vcpu;
745 kvm_for_each_vcpu(i, vcpu, kvm) {
746 vcpu->arch.pause = false;
747 __kvm_vcpu_wake_up(vcpu);
751 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
753 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
755 rcuwait_wait_event(wait,
756 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
759 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
760 /* Awaken to handle a signal, request we sleep again later. */
761 kvm_make_request(KVM_REQ_SLEEP, vcpu);
765 * Make sure we will observe a potential reset request if we've
766 * observed a change to the power state. Pairs with the smp_wmb() in
767 * kvm_psci_vcpu_on().
773 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
774 * @vcpu: The VCPU pointer
776 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
777 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
778 * on when a wake event arrives, e.g. there may already be a pending wake event.
780 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
783 * Sync back the state of the GIC CPU interface so that we have
784 * the latest PMR and group enables. This ensures that
785 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
786 * we have pending interrupts, e.g. when determining if the
789 * For the same reason, we want to tell GICv4 that we need
790 * doorbells to be signalled, should an interrupt become pending.
793 kvm_vgic_vmcr_sync(vcpu);
794 vcpu_set_flag(vcpu, IN_WFI);
799 vcpu_clear_flag(vcpu, IN_WFIT);
802 vcpu_clear_flag(vcpu, IN_WFI);
807 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
809 if (!kvm_arm_vcpu_suspended(vcpu))
815 * The suspend state is sticky; we do not leave it until userspace
816 * explicitly marks the vCPU as runnable. Request that we suspend again
819 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
822 * Check to make sure the vCPU is actually runnable. If so, exit to
823 * userspace informing it of the wakeup condition.
825 if (kvm_arch_vcpu_runnable(vcpu)) {
826 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
827 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
828 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
833 * Otherwise, we were unblocked to process a different event, such as a
834 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
841 * check_vcpu_requests - check and handle pending vCPU requests
842 * @vcpu: the VCPU pointer
844 * Return: 1 if we should enter the guest
845 * 0 if we should exit to userspace
846 * < 0 if we should exit to userspace, where the return value indicates
849 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
851 if (kvm_request_pending(vcpu)) {
852 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
853 kvm_vcpu_sleep(vcpu);
855 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
856 kvm_reset_vcpu(vcpu);
859 * Clear IRQ_PENDING requests that were made to guarantee
860 * that a VCPU sees new virtual interrupts.
862 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
864 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
865 kvm_update_stolen_time(vcpu);
867 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
868 /* The distributor enable bits were changed */
875 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
876 kvm_vcpu_reload_pmu(vcpu);
878 if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
879 kvm_vcpu_pmu_restore_guest(vcpu);
881 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
882 return kvm_vcpu_suspend(vcpu);
884 if (kvm_dirty_ring_check_request(vcpu))
891 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
893 if (likely(!vcpu_mode_is_32bit(vcpu)))
896 if (vcpu_has_nv(vcpu))
899 return !kvm_supports_32bit_el0();
903 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
904 * @vcpu: The VCPU pointer
905 * @ret: Pointer to write optional return code
907 * Returns: true if the VCPU needs to return to a preemptible + interruptible
908 * and skip guest entry.
910 * This function disambiguates between two different types of exits: exits to a
911 * preemptible + interruptible kernel context and exits to userspace. For an
912 * exit to userspace, this function will write the return code to ret and return
913 * true. For an exit to preemptible + interruptible kernel context (i.e. check
914 * for pending work and re-enter), return true without writing to ret.
916 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
918 struct kvm_run *run = vcpu->run;
921 * If we're using a userspace irqchip, then check if we need
922 * to tell a userspace irqchip about timer or PMU level
923 * changes and if so, exit to userspace (the actual level
924 * state gets updated in kvm_timer_update_run and
925 * kvm_pmu_update_run below).
927 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
928 if (kvm_timer_should_notify_user(vcpu) ||
929 kvm_pmu_should_notify_user(vcpu)) {
931 run->exit_reason = KVM_EXIT_INTR;
936 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
937 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
938 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
939 run->fail_entry.cpu = smp_processor_id();
944 return kvm_request_pending(vcpu) ||
945 xfer_to_guest_mode_work_pending();
949 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
950 * the vCPU is running.
952 * This must be noinstr as instrumentation may make use of RCU, and this is not
953 * safe during the EQS.
955 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
959 guest_state_enter_irqoff();
960 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
961 guest_state_exit_irqoff();
967 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
968 * @vcpu: The VCPU pointer
970 * This function is called through the VCPU_RUN ioctl called from user space. It
971 * will execute VM code in a loop until the time slice for the process is used
972 * or some emulation is needed from user space in which case the function will
973 * return with return value 0 and with the kvm_run structure filled in with the
974 * required data for the requested emulation.
976 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
978 struct kvm_run *run = vcpu->run;
981 if (run->exit_reason == KVM_EXIT_MMIO) {
982 ret = kvm_handle_mmio_return(vcpu);
989 if (run->immediate_exit) {
994 kvm_sigset_activate(vcpu);
997 run->exit_reason = KVM_EXIT_UNKNOWN;
1001 * Check conditions before entering the guest
1003 ret = xfer_to_guest_mode_handle_work(vcpu);
1008 ret = check_vcpu_requests(vcpu);
1011 * Preparing the interrupts to be injected also
1012 * involves poking the GIC, which must be done in a
1013 * non-preemptible context.
1018 * The VMID allocator only tracks active VMIDs per
1019 * physical CPU, and therefore the VMID allocated may not be
1020 * preserved on VMID roll-over if the task was preempted,
1021 * making a thread's VMID inactive. So we need to call
1022 * kvm_arm_vmid_update() in non-premptible context.
1024 if (kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid) &&
1026 __load_stage2(vcpu->arch.hw_mmu,
1027 vcpu->arch.hw_mmu->arch);
1029 kvm_pmu_flush_hwstate(vcpu);
1031 local_irq_disable();
1033 kvm_vgic_flush_hwstate(vcpu);
1035 kvm_pmu_update_vcpu_events(vcpu);
1038 * Ensure we set mode to IN_GUEST_MODE after we disable
1039 * interrupts and before the final VCPU requests check.
1040 * See the comment in kvm_vcpu_exiting_guest_mode() and
1041 * Documentation/virt/kvm/vcpu-requests.rst
1043 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1045 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1046 vcpu->mode = OUTSIDE_GUEST_MODE;
1047 isb(); /* Ensure work in x_flush_hwstate is committed */
1048 kvm_pmu_sync_hwstate(vcpu);
1049 if (static_branch_unlikely(&userspace_irqchip_in_use))
1050 kvm_timer_sync_user(vcpu);
1051 kvm_vgic_sync_hwstate(vcpu);
1057 kvm_arm_setup_debug(vcpu);
1058 kvm_arch_vcpu_ctxflush_fp(vcpu);
1060 /**************************************************************
1063 trace_kvm_entry(*vcpu_pc(vcpu));
1064 guest_timing_enter_irqoff();
1066 ret = kvm_arm_vcpu_enter_exit(vcpu);
1068 vcpu->mode = OUTSIDE_GUEST_MODE;
1072 *************************************************************/
1074 kvm_arm_clear_debug(vcpu);
1077 * We must sync the PMU state before the vgic state so
1078 * that the vgic can properly sample the updated state of the
1081 kvm_pmu_sync_hwstate(vcpu);
1084 * Sync the vgic state before syncing the timer state because
1085 * the timer code needs to know if the virtual timer
1086 * interrupts are active.
1088 kvm_vgic_sync_hwstate(vcpu);
1091 * Sync the timer hardware state before enabling interrupts as
1092 * we don't want vtimer interrupts to race with syncing the
1093 * timer virtual interrupt state.
1095 if (static_branch_unlikely(&userspace_irqchip_in_use))
1096 kvm_timer_sync_user(vcpu);
1098 kvm_arch_vcpu_ctxsync_fp(vcpu);
1101 * We must ensure that any pending interrupts are taken before
1102 * we exit guest timing so that timer ticks are accounted as
1103 * guest time. Transiently unmask interrupts so that any
1104 * pending interrupts are taken.
1106 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1107 * context synchronization event) is necessary to ensure that
1108 * pending interrupts are taken.
1110 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1113 local_irq_disable();
1116 guest_timing_exit_irqoff();
1120 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1122 /* Exit types that need handling before we can be preempted */
1123 handle_exit_early(vcpu, ret);
1128 * The ARMv8 architecture doesn't give the hypervisor
1129 * a mechanism to prevent a guest from dropping to AArch32 EL0
1130 * if implemented by the CPU. If we spot the guest in such
1131 * state and that we decided it wasn't supposed to do so (like
1132 * with the asymmetric AArch32 case), return to userspace with
1135 if (vcpu_mode_is_bad_32bit(vcpu)) {
1137 * As we have caught the guest red-handed, decide that
1138 * it isn't fit for purpose anymore by making the vcpu
1139 * invalid. The VMM can try and fix it by issuing a
1140 * KVM_ARM_VCPU_INIT if it really wants to.
1142 vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1143 ret = ARM_EXCEPTION_IL;
1146 ret = handle_exit(vcpu, ret);
1149 /* Tell userspace about in-kernel device output levels */
1150 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1151 kvm_timer_update_run(vcpu);
1152 kvm_pmu_update_run(vcpu);
1155 kvm_sigset_deactivate(vcpu);
1159 * In the unlikely event that we are returning to userspace
1160 * with pending exceptions or PC adjustment, commit these
1161 * adjustments in order to give userspace a consistent view of
1162 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1163 * being preempt-safe on VHE.
1165 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1166 vcpu_get_flag(vcpu, INCREMENT_PC)))
1167 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1173 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1179 if (number == KVM_ARM_IRQ_CPU_IRQ)
1180 bit_index = __ffs(HCR_VI);
1181 else /* KVM_ARM_IRQ_CPU_FIQ */
1182 bit_index = __ffs(HCR_VF);
1184 hcr = vcpu_hcr(vcpu);
1186 set = test_and_set_bit(bit_index, hcr);
1188 set = test_and_clear_bit(bit_index, hcr);
1191 * If we didn't change anything, no need to wake up or kick other CPUs
1197 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1198 * trigger a world-switch round on the running physical CPU to set the
1199 * virtual IRQ/FIQ fields in the HCR appropriately.
1201 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1202 kvm_vcpu_kick(vcpu);
1207 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1210 u32 irq = irq_level->irq;
1211 unsigned int irq_type, vcpu_id, irq_num;
1212 struct kvm_vcpu *vcpu = NULL;
1213 bool level = irq_level->level;
1215 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1216 vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1217 vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1218 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1220 trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1223 case KVM_ARM_IRQ_TYPE_CPU:
1224 if (irqchip_in_kernel(kvm))
1227 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1231 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1234 return vcpu_interrupt_line(vcpu, irq_num, level);
1235 case KVM_ARM_IRQ_TYPE_PPI:
1236 if (!irqchip_in_kernel(kvm))
1239 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1243 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1246 return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1247 case KVM_ARM_IRQ_TYPE_SPI:
1248 if (!irqchip_in_kernel(kvm))
1251 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1254 return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1260 static unsigned long system_supported_vcpu_features(void)
1262 unsigned long features = KVM_VCPU_VALID_FEATURES;
1264 if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1265 clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1267 if (!kvm_arm_support_pmu_v3())
1268 clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1270 if (!system_supports_sve())
1271 clear_bit(KVM_ARM_VCPU_SVE, &features);
1273 if (!system_has_full_ptr_auth()) {
1274 clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1275 clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1278 if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1279 clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1284 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1285 const struct kvm_vcpu_init *init)
1287 unsigned long features = init->features[0];
1290 if (features & ~KVM_VCPU_VALID_FEATURES)
1293 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1294 if (init->features[i])
1298 if (features & ~system_supported_vcpu_features())
1302 * For now make sure that both address/generic pointer authentication
1303 * features are requested by the userspace together.
1305 if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1306 test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1309 /* Disallow NV+SVE for the time being */
1310 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features) &&
1311 test_bit(KVM_ARM_VCPU_SVE, &features))
1314 if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1317 /* MTE is incompatible with AArch32 */
1318 if (kvm_has_mte(vcpu->kvm))
1321 /* NV is incompatible with AArch32 */
1322 if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1328 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1329 const struct kvm_vcpu_init *init)
1331 unsigned long features = init->features[0];
1333 return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1334 KVM_VCPU_MAX_FEATURES);
1337 static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1339 struct kvm *kvm = vcpu->kvm;
1343 * When the vCPU has a PMU, but no PMU is set for the guest
1344 * yet, set the default one.
1346 if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1347 ret = kvm_arm_set_default_pmu(kvm);
1352 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1353 const struct kvm_vcpu_init *init)
1355 unsigned long features = init->features[0];
1356 struct kvm *kvm = vcpu->kvm;
1359 mutex_lock(&kvm->arch.config_lock);
1361 if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1362 kvm_vcpu_init_changed(vcpu, init))
1365 bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1367 ret = kvm_setup_vcpu(vcpu);
1371 /* Now we know what it is, we can reset it. */
1372 kvm_reset_vcpu(vcpu);
1374 set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1375 vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1378 mutex_unlock(&kvm->arch.config_lock);
1382 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1383 const struct kvm_vcpu_init *init)
1387 if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1388 init->target != kvm_target_cpu())
1391 ret = kvm_vcpu_init_check_features(vcpu, init);
1395 if (!kvm_vcpu_initialized(vcpu))
1396 return __kvm_vcpu_set_target(vcpu, init);
1398 if (kvm_vcpu_init_changed(vcpu, init))
1401 kvm_reset_vcpu(vcpu);
1405 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1406 struct kvm_vcpu_init *init)
1408 bool power_off = false;
1412 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1413 * reflecting it in the finalized feature set, thus limiting its scope
1414 * to a single KVM_ARM_VCPU_INIT call.
1416 if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1417 init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1421 ret = kvm_vcpu_set_target(vcpu, init);
1426 * Ensure a rebooted VM will fault in RAM pages and detect if the
1427 * guest MMU is turned off and flush the caches as needed.
1429 * S2FWB enforces all memory accesses to RAM being cacheable,
1430 * ensuring that the data side is always coherent. We still
1431 * need to invalidate the I-cache though, as FWB does *not*
1432 * imply CTR_EL0.DIC.
1434 if (vcpu_has_run_once(vcpu)) {
1435 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1436 stage2_unmap_vm(vcpu->kvm);
1438 icache_inval_all_pou();
1441 vcpu_reset_hcr(vcpu);
1442 vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1445 * Handle the "start in power-off" case.
1447 spin_lock(&vcpu->arch.mp_state_lock);
1450 __kvm_arm_vcpu_power_off(vcpu);
1452 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1454 spin_unlock(&vcpu->arch.mp_state_lock);
1459 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1460 struct kvm_device_attr *attr)
1464 switch (attr->group) {
1466 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1473 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1474 struct kvm_device_attr *attr)
1478 switch (attr->group) {
1480 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1487 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1488 struct kvm_device_attr *attr)
1492 switch (attr->group) {
1494 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1501 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1502 struct kvm_vcpu_events *events)
1504 memset(events, 0, sizeof(*events));
1506 return __kvm_arm_vcpu_get_events(vcpu, events);
1509 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1510 struct kvm_vcpu_events *events)
1514 /* check whether the reserved field is zero */
1515 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1516 if (events->reserved[i])
1519 /* check whether the pad field is zero */
1520 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1521 if (events->exception.pad[i])
1524 return __kvm_arm_vcpu_set_events(vcpu, events);
1527 long kvm_arch_vcpu_ioctl(struct file *filp,
1528 unsigned int ioctl, unsigned long arg)
1530 struct kvm_vcpu *vcpu = filp->private_data;
1531 void __user *argp = (void __user *)arg;
1532 struct kvm_device_attr attr;
1536 case KVM_ARM_VCPU_INIT: {
1537 struct kvm_vcpu_init init;
1540 if (copy_from_user(&init, argp, sizeof(init)))
1543 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1546 case KVM_SET_ONE_REG:
1547 case KVM_GET_ONE_REG: {
1548 struct kvm_one_reg reg;
1551 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1555 if (copy_from_user(®, argp, sizeof(reg)))
1559 * We could owe a reset due to PSCI. Handle the pending reset
1560 * here to ensure userspace register accesses are ordered after
1563 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1564 kvm_reset_vcpu(vcpu);
1566 if (ioctl == KVM_SET_ONE_REG)
1567 r = kvm_arm_set_reg(vcpu, ®);
1569 r = kvm_arm_get_reg(vcpu, ®);
1572 case KVM_GET_REG_LIST: {
1573 struct kvm_reg_list __user *user_list = argp;
1574 struct kvm_reg_list reg_list;
1578 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1582 if (!kvm_arm_vcpu_is_finalized(vcpu))
1586 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1589 reg_list.n = kvm_arm_num_regs(vcpu);
1590 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1595 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1598 case KVM_SET_DEVICE_ATTR: {
1600 if (copy_from_user(&attr, argp, sizeof(attr)))
1602 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1605 case KVM_GET_DEVICE_ATTR: {
1607 if (copy_from_user(&attr, argp, sizeof(attr)))
1609 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1612 case KVM_HAS_DEVICE_ATTR: {
1614 if (copy_from_user(&attr, argp, sizeof(attr)))
1616 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1619 case KVM_GET_VCPU_EVENTS: {
1620 struct kvm_vcpu_events events;
1622 if (kvm_arm_vcpu_get_events(vcpu, &events))
1625 if (copy_to_user(argp, &events, sizeof(events)))
1630 case KVM_SET_VCPU_EVENTS: {
1631 struct kvm_vcpu_events events;
1633 if (copy_from_user(&events, argp, sizeof(events)))
1636 return kvm_arm_vcpu_set_events(vcpu, &events);
1638 case KVM_ARM_VCPU_FINALIZE: {
1641 if (!kvm_vcpu_initialized(vcpu))
1644 if (get_user(what, (const int __user *)argp))
1647 return kvm_arm_vcpu_finalize(vcpu, what);
1656 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1661 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1662 struct kvm_arm_device_addr *dev_addr)
1664 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1665 case KVM_ARM_DEVICE_VGIC_V2:
1668 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1674 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1676 switch (attr->group) {
1677 case KVM_ARM_VM_SMCCC_CTRL:
1678 return kvm_vm_smccc_has_attr(kvm, attr);
1684 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1686 switch (attr->group) {
1687 case KVM_ARM_VM_SMCCC_CTRL:
1688 return kvm_vm_smccc_set_attr(kvm, attr);
1694 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1696 struct kvm *kvm = filp->private_data;
1697 void __user *argp = (void __user *)arg;
1698 struct kvm_device_attr attr;
1701 case KVM_CREATE_IRQCHIP: {
1705 mutex_lock(&kvm->lock);
1706 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1707 mutex_unlock(&kvm->lock);
1710 case KVM_ARM_SET_DEVICE_ADDR: {
1711 struct kvm_arm_device_addr dev_addr;
1713 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1715 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1717 case KVM_ARM_PREFERRED_TARGET: {
1718 struct kvm_vcpu_init init = {
1719 .target = KVM_ARM_TARGET_GENERIC_V8,
1722 if (copy_to_user(argp, &init, sizeof(init)))
1727 case KVM_ARM_MTE_COPY_TAGS: {
1728 struct kvm_arm_copy_mte_tags copy_tags;
1730 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1732 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1734 case KVM_ARM_SET_COUNTER_OFFSET: {
1735 struct kvm_arm_counter_offset offset;
1737 if (copy_from_user(&offset, argp, sizeof(offset)))
1739 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1741 case KVM_HAS_DEVICE_ATTR: {
1742 if (copy_from_user(&attr, argp, sizeof(attr)))
1745 return kvm_vm_has_attr(kvm, &attr);
1747 case KVM_SET_DEVICE_ATTR: {
1748 if (copy_from_user(&attr, argp, sizeof(attr)))
1751 return kvm_vm_set_attr(kvm, &attr);
1753 case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1754 struct reg_mask_range range;
1756 if (copy_from_user(&range, argp, sizeof(range)))
1758 return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1765 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1766 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1768 struct kvm_vcpu *tmp_vcpu;
1770 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1771 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1772 mutex_unlock(&tmp_vcpu->mutex);
1776 void unlock_all_vcpus(struct kvm *kvm)
1778 lockdep_assert_held(&kvm->lock);
1780 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1783 /* Returns true if all vcpus were locked, false otherwise */
1784 bool lock_all_vcpus(struct kvm *kvm)
1786 struct kvm_vcpu *tmp_vcpu;
1789 lockdep_assert_held(&kvm->lock);
1792 * Any time a vcpu is in an ioctl (including running), the
1793 * core KVM code tries to grab the vcpu->mutex.
1795 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1796 * other VCPUs can fiddle with the state while we access it.
1798 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1799 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1800 unlock_vcpus(kvm, c - 1);
1808 static unsigned long nvhe_percpu_size(void)
1810 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1811 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1814 static unsigned long nvhe_percpu_order(void)
1816 unsigned long size = nvhe_percpu_size();
1818 return size ? get_order(size) : 0;
1821 /* A lookup table holding the hypervisor VA for each vector slot */
1822 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1824 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1826 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1829 static int kvm_init_vector_slots(void)
1834 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1835 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1837 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1838 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1840 if (kvm_system_needs_idmapped_vectors() &&
1841 !is_protected_kvm_enabled()) {
1842 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1843 __BP_HARDEN_HYP_VECS_SZ, &base);
1848 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1849 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1853 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1855 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1856 u64 mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1860 * Calculate the raw per-cpu offset without a translation from the
1861 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1862 * so that we can use adr_l to access per-cpu variables in EL2.
1863 * Also drop the KASAN tag which gets in the way...
1865 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1866 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1868 params->mair_el2 = read_sysreg(mair_el1);
1870 tcr = read_sysreg(tcr_el1);
1871 if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
1872 tcr |= TCR_EPD1_MASK;
1874 tcr &= TCR_EL2_MASK;
1875 tcr |= TCR_EL2_RES1;
1877 tcr &= ~TCR_T0SZ_MASK;
1878 tcr |= TCR_T0SZ(hyp_va_bits);
1879 tcr &= ~TCR_EL2_PS_MASK;
1880 tcr |= FIELD_PREP(TCR_EL2_PS_MASK, kvm_get_parange(mmfr0));
1881 if (kvm_lpa2_is_enabled())
1883 params->tcr_el2 = tcr;
1885 params->pgd_pa = kvm_mmu_get_httbr();
1886 if (is_protected_kvm_enabled())
1887 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1889 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1890 if (cpus_have_final_cap(ARM64_KVM_HVHE))
1891 params->hcr_el2 |= HCR_E2H;
1892 params->vttbr = params->vtcr = 0;
1895 * Flush the init params from the data cache because the struct will
1896 * be read while the MMU is off.
1898 kvm_flush_dcache_to_poc(params, sizeof(*params));
1901 static void hyp_install_host_vector(void)
1903 struct kvm_nvhe_init_params *params;
1904 struct arm_smccc_res res;
1906 /* Switch from the HYP stub to our own HYP init vector */
1907 __hyp_set_vectors(kvm_get_idmap_vector());
1910 * Call initialization code, and switch to the full blown HYP code.
1911 * If the cpucaps haven't been finalized yet, something has gone very
1912 * wrong, and hyp will crash and burn when it uses any
1913 * cpus_have_*_cap() wrapper.
1915 BUG_ON(!system_capabilities_finalized());
1916 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1917 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1918 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1921 static void cpu_init_hyp_mode(void)
1923 hyp_install_host_vector();
1926 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1929 if (this_cpu_has_cap(ARM64_SSBS) &&
1930 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1931 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1935 static void cpu_hyp_reset(void)
1937 if (!is_kernel_in_hyp_mode())
1938 __hyp_reset_vectors();
1942 * EL2 vectors can be mapped and rerouted in a number of ways,
1943 * depending on the kernel configuration and CPU present:
1945 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1946 * placed in one of the vector slots, which is executed before jumping
1947 * to the real vectors.
1949 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1950 * containing the hardening sequence is mapped next to the idmap page,
1951 * and executed before jumping to the real vectors.
1953 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1954 * empty slot is selected, mapped next to the idmap page, and
1955 * executed before jumping to the real vectors.
1957 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1958 * VHE, as we don't have hypervisor-specific mappings. If the system
1959 * is VHE and yet selects this capability, it will be ignored.
1961 static void cpu_set_hyp_vector(void)
1963 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1964 void *vector = hyp_spectre_vector_selector[data->slot];
1966 if (!is_protected_kvm_enabled())
1967 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1969 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1972 static void cpu_hyp_init_context(void)
1974 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1976 if (!is_kernel_in_hyp_mode())
1977 cpu_init_hyp_mode();
1980 static void cpu_hyp_init_features(void)
1982 cpu_set_hyp_vector();
1983 kvm_arm_init_debug();
1985 if (is_kernel_in_hyp_mode())
1986 kvm_timer_init_vhe();
1989 kvm_vgic_init_cpu_hardware();
1992 static void cpu_hyp_reinit(void)
1995 cpu_hyp_init_context();
1996 cpu_hyp_init_features();
1999 static void cpu_hyp_init(void *discard)
2001 if (!__this_cpu_read(kvm_hyp_initialized)) {
2003 __this_cpu_write(kvm_hyp_initialized, 1);
2007 static void cpu_hyp_uninit(void *discard)
2009 if (__this_cpu_read(kvm_hyp_initialized)) {
2011 __this_cpu_write(kvm_hyp_initialized, 0);
2015 int kvm_arch_hardware_enable(void)
2018 * Most calls to this function are made with migration
2019 * disabled, but not with preemption disabled. The former is
2020 * enough to ensure correctness, but most of the helpers
2021 * expect the later and will throw a tantrum otherwise.
2035 void kvm_arch_hardware_disable(void)
2037 kvm_timer_cpu_down();
2038 kvm_vgic_cpu_down();
2040 if (!is_protected_kvm_enabled())
2041 cpu_hyp_uninit(NULL);
2044 #ifdef CONFIG_CPU_PM
2045 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2050 * kvm_hyp_initialized is left with its old value over
2051 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2056 if (__this_cpu_read(kvm_hyp_initialized))
2058 * don't update kvm_hyp_initialized here
2059 * so that the hyp will be re-enabled
2060 * when we resume. See below.
2065 case CPU_PM_ENTER_FAILED:
2067 if (__this_cpu_read(kvm_hyp_initialized))
2068 /* The hyp was enabled before suspend. */
2078 static struct notifier_block hyp_init_cpu_pm_nb = {
2079 .notifier_call = hyp_init_cpu_pm_notifier,
2082 static void __init hyp_cpu_pm_init(void)
2084 if (!is_protected_kvm_enabled())
2085 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2087 static void __init hyp_cpu_pm_exit(void)
2089 if (!is_protected_kvm_enabled())
2090 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2093 static inline void __init hyp_cpu_pm_init(void)
2096 static inline void __init hyp_cpu_pm_exit(void)
2101 static void __init init_cpu_logical_map(void)
2106 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2107 * Only copy the set of online CPUs whose features have been checked
2108 * against the finalized system capabilities. The hypervisor will not
2109 * allow any other CPUs from the `possible` set to boot.
2111 for_each_online_cpu(cpu)
2112 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2115 #define init_psci_0_1_impl_state(config, what) \
2116 config.psci_0_1_ ## what ## _implemented = psci_ops.what
2118 static bool __init init_psci_relay(void)
2121 * If PSCI has not been initialized, protected KVM cannot install
2122 * itself on newly booted CPUs.
2124 if (!psci_ops.get_version) {
2125 kvm_err("Cannot initialize protected mode without PSCI\n");
2129 kvm_host_psci_config.version = psci_ops.get_version();
2130 kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2132 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2133 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2134 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2135 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2136 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2137 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2142 static int __init init_subsystems(void)
2147 * Enable hardware so that subsystem initialisation can access EL2.
2149 on_each_cpu(cpu_hyp_init, NULL, 1);
2152 * Register CPU lower-power notifier
2157 * Init HYP view of VGIC
2159 err = kvm_vgic_hyp_init();
2162 vgic_present = true;
2166 vgic_present = false;
2174 * Init HYP architected timer support
2176 err = kvm_timer_hyp_init(vgic_present);
2180 kvm_register_perf_callbacks(NULL);
2186 if (err || !is_protected_kvm_enabled())
2187 on_each_cpu(cpu_hyp_uninit, NULL, 1);
2192 static void __init teardown_subsystems(void)
2194 kvm_unregister_perf_callbacks();
2198 static void __init teardown_hyp_mode(void)
2203 for_each_possible_cpu(cpu) {
2204 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2205 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2209 static int __init do_pkvm_init(u32 hyp_va_bits)
2211 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2215 cpu_hyp_init_context();
2216 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2217 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2219 cpu_hyp_init_features();
2222 * The stub hypercalls are now disabled, so set our local flag to
2223 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2225 __this_cpu_write(kvm_hyp_initialized, 1);
2231 static u64 get_hyp_id_aa64pfr0_el1(void)
2234 * Track whether the system isn't affected by spectre/meltdown in the
2235 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2236 * Although this is per-CPU, we make it global for simplicity, e.g., not
2237 * to have to worry about vcpu migration.
2239 * Unlike for non-protected VMs, userspace cannot override this for
2242 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2244 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2245 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2247 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2248 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2249 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2250 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2255 static void kvm_hyp_init_symbols(void)
2257 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2258 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2259 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2260 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2261 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2262 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2263 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2264 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2265 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2266 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2267 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2270 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2272 void *addr = phys_to_virt(hyp_mem_base);
2275 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2279 ret = do_pkvm_init(hyp_va_bits);
2288 static void pkvm_hyp_init_ptrauth(void)
2290 struct kvm_cpu_context *hyp_ctxt;
2293 for_each_possible_cpu(cpu) {
2294 hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2295 hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2296 hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2297 hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2298 hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2299 hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2300 hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2301 hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2302 hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2303 hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2304 hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2308 /* Inits Hyp-mode on all online CPUs */
2309 static int __init init_hyp_mode(void)
2316 * The protected Hyp-mode cannot be initialized if the memory pool
2317 * allocation has failed.
2319 if (is_protected_kvm_enabled() && !hyp_mem_base)
2323 * Allocate Hyp PGD and setup Hyp identity mapping
2325 err = kvm_mmu_init(&hyp_va_bits);
2330 * Allocate stack pages for Hypervisor-mode
2332 for_each_possible_cpu(cpu) {
2333 unsigned long stack_page;
2335 stack_page = __get_free_page(GFP_KERNEL);
2341 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2345 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2347 for_each_possible_cpu(cpu) {
2351 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2357 page_addr = page_address(page);
2358 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2359 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2363 * Map the Hyp-code called directly from the host
2365 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2366 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2368 kvm_err("Cannot map world-switch code\n");
2372 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2373 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2375 kvm_err("Cannot map .hyp.rodata section\n");
2379 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2380 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2382 kvm_err("Cannot map rodata section\n");
2387 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2388 * section thanks to an assertion in the linker script. Map it RW and
2389 * the rest of .bss RO.
2391 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2392 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2394 kvm_err("Cannot map hyp bss section: %d\n", err);
2398 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2399 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2401 kvm_err("Cannot map bss section\n");
2406 * Map the Hyp stack pages
2408 for_each_possible_cpu(cpu) {
2409 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2410 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2412 err = create_hyp_stack(__pa(stack_page), ¶ms->stack_hyp_va);
2414 kvm_err("Cannot map hyp stack\n");
2419 * Save the stack PA in nvhe_init_params. This will be needed
2420 * to recreate the stack mapping in protected nVHE mode.
2421 * __hyp_pa() won't do the right thing there, since the stack
2422 * has been mapped in the flexible private VA space.
2424 params->stack_pa = __pa(stack_page);
2427 for_each_possible_cpu(cpu) {
2428 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2429 char *percpu_end = percpu_begin + nvhe_percpu_size();
2431 /* Map Hyp percpu pages */
2432 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2434 kvm_err("Cannot map hyp percpu region\n");
2438 /* Prepare the CPU initialization parameters */
2439 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2442 kvm_hyp_init_symbols();
2444 if (is_protected_kvm_enabled()) {
2445 if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2446 cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2447 pkvm_hyp_init_ptrauth();
2449 init_cpu_logical_map();
2451 if (!init_psci_relay()) {
2456 err = kvm_hyp_init_protection(hyp_va_bits);
2458 kvm_err("Failed to init hyp memory protection\n");
2466 teardown_hyp_mode();
2467 kvm_err("error initializing Hyp mode: %d\n", err);
2471 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2473 struct kvm_vcpu *vcpu;
2476 mpidr &= MPIDR_HWID_BITMASK;
2478 if (kvm->arch.mpidr_data) {
2479 u16 idx = kvm_mpidr_index(kvm->arch.mpidr_data, mpidr);
2481 vcpu = kvm_get_vcpu(kvm,
2482 kvm->arch.mpidr_data->cmpidr_to_idx[idx]);
2483 if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2489 kvm_for_each_vcpu(i, vcpu, kvm) {
2490 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2496 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2498 return irqchip_in_kernel(kvm);
2501 bool kvm_arch_has_irq_bypass(void)
2506 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2507 struct irq_bypass_producer *prod)
2509 struct kvm_kernel_irqfd *irqfd =
2510 container_of(cons, struct kvm_kernel_irqfd, consumer);
2512 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2515 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2516 struct irq_bypass_producer *prod)
2518 struct kvm_kernel_irqfd *irqfd =
2519 container_of(cons, struct kvm_kernel_irqfd, consumer);
2521 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2525 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2527 struct kvm_kernel_irqfd *irqfd =
2528 container_of(cons, struct kvm_kernel_irqfd, consumer);
2530 kvm_arm_halt_guest(irqfd->kvm);
2533 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2535 struct kvm_kernel_irqfd *irqfd =
2536 container_of(cons, struct kvm_kernel_irqfd, consumer);
2538 kvm_arm_resume_guest(irqfd->kvm);
2541 /* Initialize Hyp-mode and memory mappings on all CPUs */
2542 static __init int kvm_arm_init(void)
2547 if (!is_hyp_mode_available()) {
2548 kvm_info("HYP mode not available\n");
2552 if (kvm_get_mode() == KVM_MODE_NONE) {
2553 kvm_info("KVM disabled from command line\n");
2557 err = kvm_sys_reg_table_init();
2559 kvm_info("Error initializing system register tables");
2563 in_hyp_mode = is_kernel_in_hyp_mode();
2565 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2566 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2567 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2568 "Only trusted guests should be used on this system.\n");
2570 err = kvm_set_ipa_limit();
2574 err = kvm_arm_init_sve();
2578 err = kvm_arm_vmid_alloc_init();
2580 kvm_err("Failed to initialize VMID allocator.\n");
2585 err = init_hyp_mode();
2590 err = kvm_init_vector_slots();
2592 kvm_err("Cannot initialise vector slots\n");
2596 err = init_subsystems();
2600 if (is_protected_kvm_enabled()) {
2601 kvm_info("Protected nVHE mode initialized successfully\n");
2602 } else if (in_hyp_mode) {
2603 kvm_info("VHE mode initialized successfully\n");
2605 char mode = cpus_have_final_cap(ARM64_KVM_HVHE) ? 'h' : 'n';
2606 kvm_info("Hyp mode (%cVHE) initialized successfully\n", mode);
2610 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2611 * hypervisor protection is finalized.
2613 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2617 kvm_arm_initialised = true;
2622 teardown_subsystems();
2625 teardown_hyp_mode();
2627 kvm_arm_vmid_alloc_free();
2631 static int __init early_kvm_mode_cfg(char *arg)
2636 if (strcmp(arg, "none") == 0) {
2637 kvm_mode = KVM_MODE_NONE;
2641 if (!is_hyp_mode_available()) {
2642 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2646 if (strcmp(arg, "protected") == 0) {
2647 if (!is_kernel_in_hyp_mode())
2648 kvm_mode = KVM_MODE_PROTECTED;
2650 pr_warn_once("Protected KVM not available with VHE\n");
2655 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2656 kvm_mode = KVM_MODE_DEFAULT;
2660 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2661 kvm_mode = KVM_MODE_NV;
2667 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2669 enum kvm_mode kvm_get_mode(void)
2674 module_init(kvm_arm_init);