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 static bool vgic_present;
56 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
57 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
59 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
61 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
64 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
65 struct kvm_enable_cap *cap)
73 case KVM_CAP_ARM_NISV_TO_USER:
75 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
79 mutex_lock(&kvm->lock);
80 if (!system_supports_mte() || kvm->created_vcpus) {
84 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
86 mutex_unlock(&kvm->lock);
88 case KVM_CAP_ARM_SYSTEM_SUSPEND:
90 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
100 static int kvm_arm_default_max_vcpus(void)
102 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
105 static void set_default_spectre(struct kvm *kvm)
108 * The default is to expose CSV2 == 1 if the HW isn't affected.
109 * Although this is a per-CPU feature, we make it global because
110 * asymmetric systems are just a nuisance.
112 * Userspace can override this as long as it doesn't promise
115 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
116 kvm->arch.pfr0_csv2 = 1;
117 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
118 kvm->arch.pfr0_csv3 = 1;
122 * kvm_arch_init_vm - initializes a VM data structure
123 * @kvm: pointer to the KVM struct
125 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
129 mutex_init(&kvm->arch.config_lock);
131 #ifdef CONFIG_LOCKDEP
132 /* Clue in lockdep that the config_lock must be taken inside kvm->lock */
133 mutex_lock(&kvm->lock);
134 mutex_lock(&kvm->arch.config_lock);
135 mutex_unlock(&kvm->arch.config_lock);
136 mutex_unlock(&kvm->lock);
139 ret = kvm_share_hyp(kvm, kvm + 1);
143 ret = pkvm_init_host_vm(kvm);
145 goto err_unshare_kvm;
147 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
149 goto err_unshare_kvm;
151 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
153 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
155 goto err_free_cpumask;
157 kvm_vgic_early_init(kvm);
159 kvm_timer_init_vm(kvm);
161 /* The maximum number of VCPUs is limited by the host's GIC model */
162 kvm->max_vcpus = kvm_arm_default_max_vcpus();
164 set_default_spectre(kvm);
165 kvm_arm_init_hypercalls(kvm);
168 * Initialise the default PMUver before there is a chance to
169 * create an actual PMU.
171 kvm->arch.dfr0_pmuver.imp = kvm_arm_pmu_get_pmuver_limit();
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();
312 long kvm_arch_dev_ioctl(struct file *filp,
313 unsigned int ioctl, unsigned long arg)
318 struct kvm *kvm_arch_alloc_vm(void)
320 size_t sz = sizeof(struct kvm);
323 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
325 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
328 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
330 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
333 if (id >= kvm->max_vcpus)
339 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
343 spin_lock_init(&vcpu->arch.mp_state_lock);
345 #ifdef CONFIG_LOCKDEP
346 /* Inform lockdep that the config_lock is acquired after vcpu->mutex */
347 mutex_lock(&vcpu->mutex);
348 mutex_lock(&vcpu->kvm->arch.config_lock);
349 mutex_unlock(&vcpu->kvm->arch.config_lock);
350 mutex_unlock(&vcpu->mutex);
353 /* Force users to call KVM_ARM_VCPU_INIT */
354 vcpu->arch.target = -1;
355 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
357 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
360 * Default value for the FP state, will be overloaded at load
361 * time if we support FP (pretty likely)
363 vcpu->arch.fp_state = FP_STATE_FREE;
365 /* Set up the timer */
366 kvm_timer_vcpu_init(vcpu);
368 kvm_pmu_vcpu_init(vcpu);
370 kvm_arm_reset_debug_ptr(vcpu);
372 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
374 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
376 err = kvm_vgic_vcpu_init(vcpu);
380 return kvm_share_hyp(vcpu, vcpu + 1);
383 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
387 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
389 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
390 static_branch_dec(&userspace_irqchip_in_use);
392 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
393 kvm_timer_vcpu_terminate(vcpu);
394 kvm_pmu_vcpu_destroy(vcpu);
396 kvm_arm_vcpu_destroy(vcpu);
399 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
404 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
409 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
411 struct kvm_s2_mmu *mmu;
414 mmu = vcpu->arch.hw_mmu;
415 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
418 * We guarantee that both TLBs and I-cache are private to each
419 * vcpu. If detecting that a vcpu from the same VM has
420 * previously run on the same physical CPU, call into the
421 * hypervisor code to nuke the relevant contexts.
423 * We might get preempted before the vCPU actually runs, but
424 * over-invalidation doesn't affect correctness.
426 if (*last_ran != vcpu->vcpu_id) {
427 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
428 *last_ran = vcpu->vcpu_id;
434 kvm_timer_vcpu_load(vcpu);
436 kvm_vcpu_load_sysregs_vhe(vcpu);
437 kvm_arch_vcpu_load_fp(vcpu);
438 kvm_vcpu_pmu_restore_guest(vcpu);
439 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
440 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
442 if (single_task_running())
443 vcpu_clear_wfx_traps(vcpu);
445 vcpu_set_wfx_traps(vcpu);
447 if (vcpu_has_ptrauth(vcpu))
448 vcpu_ptrauth_disable(vcpu);
449 kvm_arch_vcpu_load_debug_state_flags(vcpu);
451 if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
452 vcpu_set_on_unsupported_cpu(vcpu);
455 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
457 kvm_arch_vcpu_put_debug_state_flags(vcpu);
458 kvm_arch_vcpu_put_fp(vcpu);
460 kvm_vcpu_put_sysregs_vhe(vcpu);
461 kvm_timer_vcpu_put(vcpu);
463 kvm_vcpu_pmu_restore_host(vcpu);
464 kvm_arm_vmid_clear_active();
466 vcpu_clear_on_unsupported_cpu(vcpu);
470 static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
472 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
473 kvm_make_request(KVM_REQ_SLEEP, vcpu);
477 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
479 spin_lock(&vcpu->arch.mp_state_lock);
480 __kvm_arm_vcpu_power_off(vcpu);
481 spin_unlock(&vcpu->arch.mp_state_lock);
484 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
486 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
489 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
491 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
492 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
496 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
498 return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
501 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
502 struct kvm_mp_state *mp_state)
504 *mp_state = READ_ONCE(vcpu->arch.mp_state);
509 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
510 struct kvm_mp_state *mp_state)
514 spin_lock(&vcpu->arch.mp_state_lock);
516 switch (mp_state->mp_state) {
517 case KVM_MP_STATE_RUNNABLE:
518 WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
520 case KVM_MP_STATE_STOPPED:
521 __kvm_arm_vcpu_power_off(vcpu);
523 case KVM_MP_STATE_SUSPENDED:
524 kvm_arm_vcpu_suspend(vcpu);
530 spin_unlock(&vcpu->arch.mp_state_lock);
536 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
537 * @v: The VCPU pointer
539 * If the guest CPU is not waiting for interrupts or an interrupt line is
540 * asserted, the CPU is by definition runnable.
542 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
544 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
545 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
546 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
549 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
551 return vcpu_mode_priv(vcpu);
554 #ifdef CONFIG_GUEST_PERF_EVENTS
555 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
557 return *vcpu_pc(vcpu);
561 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
563 return vcpu->arch.target >= 0;
567 * Handle both the initialisation that is being done when the vcpu is
568 * run for the first time, as well as the updates that must be
569 * performed each time we get a new thread dealing with this vcpu.
571 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
573 struct kvm *kvm = vcpu->kvm;
576 if (!kvm_vcpu_initialized(vcpu))
579 if (!kvm_arm_vcpu_is_finalized(vcpu))
582 ret = kvm_arch_vcpu_run_map_fp(vcpu);
586 if (likely(vcpu_has_run_once(vcpu)))
589 kvm_arm_vcpu_init_debug(vcpu);
591 if (likely(irqchip_in_kernel(kvm))) {
593 * Map the VGIC hardware resources before running a vcpu the
594 * first time on this VM.
596 ret = kvm_vgic_map_resources(kvm);
601 ret = kvm_timer_enable(vcpu);
605 ret = kvm_arm_pmu_v3_enable(vcpu);
609 if (is_protected_kvm_enabled()) {
610 ret = pkvm_create_hyp_vm(kvm);
615 if (!irqchip_in_kernel(kvm)) {
617 * Tell the rest of the code that there are userspace irqchip
620 static_branch_inc(&userspace_irqchip_in_use);
624 * Initialize traps for protected VMs.
625 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
626 * the code is in place for first run initialization at EL2.
628 if (kvm_vm_is_protected(kvm))
629 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
631 mutex_lock(&kvm->arch.config_lock);
632 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
633 mutex_unlock(&kvm->arch.config_lock);
638 bool kvm_arch_intc_initialized(struct kvm *kvm)
640 return vgic_initialized(kvm);
643 void kvm_arm_halt_guest(struct kvm *kvm)
646 struct kvm_vcpu *vcpu;
648 kvm_for_each_vcpu(i, vcpu, kvm)
649 vcpu->arch.pause = true;
650 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
653 void kvm_arm_resume_guest(struct kvm *kvm)
656 struct kvm_vcpu *vcpu;
658 kvm_for_each_vcpu(i, vcpu, kvm) {
659 vcpu->arch.pause = false;
660 __kvm_vcpu_wake_up(vcpu);
664 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
666 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
668 rcuwait_wait_event(wait,
669 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
672 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
673 /* Awaken to handle a signal, request we sleep again later. */
674 kvm_make_request(KVM_REQ_SLEEP, vcpu);
678 * Make sure we will observe a potential reset request if we've
679 * observed a change to the power state. Pairs with the smp_wmb() in
680 * kvm_psci_vcpu_on().
686 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
687 * @vcpu: The VCPU pointer
689 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
690 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
691 * on when a wake event arrives, e.g. there may already be a pending wake event.
693 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
696 * Sync back the state of the GIC CPU interface so that we have
697 * the latest PMR and group enables. This ensures that
698 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
699 * we have pending interrupts, e.g. when determining if the
702 * For the same reason, we want to tell GICv4 that we need
703 * doorbells to be signalled, should an interrupt become pending.
706 kvm_vgic_vmcr_sync(vcpu);
707 vgic_v4_put(vcpu, true);
711 vcpu_clear_flag(vcpu, IN_WFIT);
718 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
720 if (!kvm_arm_vcpu_suspended(vcpu))
726 * The suspend state is sticky; we do not leave it until userspace
727 * explicitly marks the vCPU as runnable. Request that we suspend again
730 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
733 * Check to make sure the vCPU is actually runnable. If so, exit to
734 * userspace informing it of the wakeup condition.
736 if (kvm_arch_vcpu_runnable(vcpu)) {
737 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
738 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
739 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
744 * Otherwise, we were unblocked to process a different event, such as a
745 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
752 * check_vcpu_requests - check and handle pending vCPU requests
753 * @vcpu: the VCPU pointer
755 * Return: 1 if we should enter the guest
756 * 0 if we should exit to userspace
757 * < 0 if we should exit to userspace, where the return value indicates
760 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
762 if (kvm_request_pending(vcpu)) {
763 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
764 kvm_vcpu_sleep(vcpu);
766 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
767 kvm_reset_vcpu(vcpu);
770 * Clear IRQ_PENDING requests that were made to guarantee
771 * that a VCPU sees new virtual interrupts.
773 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
775 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
776 kvm_update_stolen_time(vcpu);
778 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
779 /* The distributor enable bits were changed */
781 vgic_v4_put(vcpu, false);
786 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
787 kvm_pmu_handle_pmcr(vcpu,
788 __vcpu_sys_reg(vcpu, PMCR_EL0));
790 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
791 return kvm_vcpu_suspend(vcpu);
793 if (kvm_dirty_ring_check_request(vcpu))
800 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
802 if (likely(!vcpu_mode_is_32bit(vcpu)))
805 return !kvm_supports_32bit_el0();
809 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
810 * @vcpu: The VCPU pointer
811 * @ret: Pointer to write optional return code
813 * Returns: true if the VCPU needs to return to a preemptible + interruptible
814 * and skip guest entry.
816 * This function disambiguates between two different types of exits: exits to a
817 * preemptible + interruptible kernel context and exits to userspace. For an
818 * exit to userspace, this function will write the return code to ret and return
819 * true. For an exit to preemptible + interruptible kernel context (i.e. check
820 * for pending work and re-enter), return true without writing to ret.
822 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
824 struct kvm_run *run = vcpu->run;
827 * If we're using a userspace irqchip, then check if we need
828 * to tell a userspace irqchip about timer or PMU level
829 * changes and if so, exit to userspace (the actual level
830 * state gets updated in kvm_timer_update_run and
831 * kvm_pmu_update_run below).
833 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
834 if (kvm_timer_should_notify_user(vcpu) ||
835 kvm_pmu_should_notify_user(vcpu)) {
837 run->exit_reason = KVM_EXIT_INTR;
842 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
843 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
844 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
845 run->fail_entry.cpu = smp_processor_id();
850 return kvm_request_pending(vcpu) ||
851 xfer_to_guest_mode_work_pending();
855 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
856 * the vCPU is running.
858 * This must be noinstr as instrumentation may make use of RCU, and this is not
859 * safe during the EQS.
861 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
865 guest_state_enter_irqoff();
866 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
867 guest_state_exit_irqoff();
873 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
874 * @vcpu: The VCPU pointer
876 * This function is called through the VCPU_RUN ioctl called from user space. It
877 * will execute VM code in a loop until the time slice for the process is used
878 * or some emulation is needed from user space in which case the function will
879 * return with return value 0 and with the kvm_run structure filled in with the
880 * required data for the requested emulation.
882 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
884 struct kvm_run *run = vcpu->run;
887 if (run->exit_reason == KVM_EXIT_MMIO) {
888 ret = kvm_handle_mmio_return(vcpu);
895 if (run->immediate_exit) {
900 kvm_sigset_activate(vcpu);
903 run->exit_reason = KVM_EXIT_UNKNOWN;
907 * Check conditions before entering the guest
909 ret = xfer_to_guest_mode_handle_work(vcpu);
914 ret = check_vcpu_requests(vcpu);
917 * Preparing the interrupts to be injected also
918 * involves poking the GIC, which must be done in a
919 * non-preemptible context.
924 * The VMID allocator only tracks active VMIDs per
925 * physical CPU, and therefore the VMID allocated may not be
926 * preserved on VMID roll-over if the task was preempted,
927 * making a thread's VMID inactive. So we need to call
928 * kvm_arm_vmid_update() in non-premptible context.
930 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
932 kvm_pmu_flush_hwstate(vcpu);
936 kvm_vgic_flush_hwstate(vcpu);
938 kvm_pmu_update_vcpu_events(vcpu);
941 * Ensure we set mode to IN_GUEST_MODE after we disable
942 * interrupts and before the final VCPU requests check.
943 * See the comment in kvm_vcpu_exiting_guest_mode() and
944 * Documentation/virt/kvm/vcpu-requests.rst
946 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
948 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
949 vcpu->mode = OUTSIDE_GUEST_MODE;
950 isb(); /* Ensure work in x_flush_hwstate is committed */
951 kvm_pmu_sync_hwstate(vcpu);
952 if (static_branch_unlikely(&userspace_irqchip_in_use))
953 kvm_timer_sync_user(vcpu);
954 kvm_vgic_sync_hwstate(vcpu);
960 kvm_arm_setup_debug(vcpu);
961 kvm_arch_vcpu_ctxflush_fp(vcpu);
963 /**************************************************************
966 trace_kvm_entry(*vcpu_pc(vcpu));
967 guest_timing_enter_irqoff();
969 ret = kvm_arm_vcpu_enter_exit(vcpu);
971 vcpu->mode = OUTSIDE_GUEST_MODE;
975 *************************************************************/
977 kvm_arm_clear_debug(vcpu);
980 * We must sync the PMU state before the vgic state so
981 * that the vgic can properly sample the updated state of the
984 kvm_pmu_sync_hwstate(vcpu);
987 * Sync the vgic state before syncing the timer state because
988 * the timer code needs to know if the virtual timer
989 * interrupts are active.
991 kvm_vgic_sync_hwstate(vcpu);
994 * Sync the timer hardware state before enabling interrupts as
995 * we don't want vtimer interrupts to race with syncing the
996 * timer virtual interrupt state.
998 if (static_branch_unlikely(&userspace_irqchip_in_use))
999 kvm_timer_sync_user(vcpu);
1001 kvm_arch_vcpu_ctxsync_fp(vcpu);
1004 * We must ensure that any pending interrupts are taken before
1005 * we exit guest timing so that timer ticks are accounted as
1006 * guest time. Transiently unmask interrupts so that any
1007 * pending interrupts are taken.
1009 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1010 * context synchronization event) is necessary to ensure that
1011 * pending interrupts are taken.
1013 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1016 local_irq_disable();
1019 guest_timing_exit_irqoff();
1023 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1025 /* Exit types that need handling before we can be preempted */
1026 handle_exit_early(vcpu, ret);
1031 * The ARMv8 architecture doesn't give the hypervisor
1032 * a mechanism to prevent a guest from dropping to AArch32 EL0
1033 * if implemented by the CPU. If we spot the guest in such
1034 * state and that we decided it wasn't supposed to do so (like
1035 * with the asymmetric AArch32 case), return to userspace with
1038 if (vcpu_mode_is_bad_32bit(vcpu)) {
1040 * As we have caught the guest red-handed, decide that
1041 * it isn't fit for purpose anymore by making the vcpu
1042 * invalid. The VMM can try and fix it by issuing a
1043 * KVM_ARM_VCPU_INIT if it really wants to.
1045 vcpu->arch.target = -1;
1046 ret = ARM_EXCEPTION_IL;
1049 ret = handle_exit(vcpu, ret);
1052 /* Tell userspace about in-kernel device output levels */
1053 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1054 kvm_timer_update_run(vcpu);
1055 kvm_pmu_update_run(vcpu);
1058 kvm_sigset_deactivate(vcpu);
1062 * In the unlikely event that we are returning to userspace
1063 * with pending exceptions or PC adjustment, commit these
1064 * adjustments in order to give userspace a consistent view of
1065 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1066 * being preempt-safe on VHE.
1068 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1069 vcpu_get_flag(vcpu, INCREMENT_PC)))
1070 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1076 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1082 if (number == KVM_ARM_IRQ_CPU_IRQ)
1083 bit_index = __ffs(HCR_VI);
1084 else /* KVM_ARM_IRQ_CPU_FIQ */
1085 bit_index = __ffs(HCR_VF);
1087 hcr = vcpu_hcr(vcpu);
1089 set = test_and_set_bit(bit_index, hcr);
1091 set = test_and_clear_bit(bit_index, hcr);
1094 * If we didn't change anything, no need to wake up or kick other CPUs
1100 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1101 * trigger a world-switch round on the running physical CPU to set the
1102 * virtual IRQ/FIQ fields in the HCR appropriately.
1104 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1105 kvm_vcpu_kick(vcpu);
1110 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1113 u32 irq = irq_level->irq;
1114 unsigned int irq_type, vcpu_idx, irq_num;
1115 int nrcpus = atomic_read(&kvm->online_vcpus);
1116 struct kvm_vcpu *vcpu = NULL;
1117 bool level = irq_level->level;
1119 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1120 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1121 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1122 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1124 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1127 case KVM_ARM_IRQ_TYPE_CPU:
1128 if (irqchip_in_kernel(kvm))
1131 if (vcpu_idx >= nrcpus)
1134 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1138 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1141 return vcpu_interrupt_line(vcpu, irq_num, level);
1142 case KVM_ARM_IRQ_TYPE_PPI:
1143 if (!irqchip_in_kernel(kvm))
1146 if (vcpu_idx >= nrcpus)
1149 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1153 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1156 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1157 case KVM_ARM_IRQ_TYPE_SPI:
1158 if (!irqchip_in_kernel(kvm))
1161 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1164 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1170 static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1171 const struct kvm_vcpu_init *init)
1173 unsigned long features = init->features[0];
1176 if (features & ~KVM_VCPU_VALID_FEATURES)
1179 for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1180 if (init->features[i])
1187 static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1188 const struct kvm_vcpu_init *init)
1190 unsigned long features = init->features[0];
1192 return !bitmap_equal(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES) ||
1193 vcpu->arch.target != init->target;
1196 static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1197 const struct kvm_vcpu_init *init)
1199 unsigned long features = init->features[0];
1202 vcpu->arch.target = init->target;
1203 bitmap_copy(vcpu->arch.features, &features, KVM_VCPU_MAX_FEATURES);
1205 /* Now we know what it is, we can reset it. */
1206 ret = kvm_reset_vcpu(vcpu);
1208 vcpu->arch.target = -1;
1209 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1215 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1216 const struct kvm_vcpu_init *init)
1220 if (init->target != kvm_target_cpu())
1223 ret = kvm_vcpu_init_check_features(vcpu, init);
1227 if (vcpu->arch.target == -1)
1228 return __kvm_vcpu_set_target(vcpu, init);
1230 if (kvm_vcpu_init_changed(vcpu, init))
1233 return kvm_reset_vcpu(vcpu);
1236 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1237 struct kvm_vcpu_init *init)
1241 ret = kvm_vcpu_set_target(vcpu, init);
1246 * Ensure a rebooted VM will fault in RAM pages and detect if the
1247 * guest MMU is turned off and flush the caches as needed.
1249 * S2FWB enforces all memory accesses to RAM being cacheable,
1250 * ensuring that the data side is always coherent. We still
1251 * need to invalidate the I-cache though, as FWB does *not*
1252 * imply CTR_EL0.DIC.
1254 if (vcpu_has_run_once(vcpu)) {
1255 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1256 stage2_unmap_vm(vcpu->kvm);
1258 icache_inval_all_pou();
1261 vcpu_reset_hcr(vcpu);
1262 vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1265 * Handle the "start in power-off" case.
1267 spin_lock(&vcpu->arch.mp_state_lock);
1269 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1270 __kvm_arm_vcpu_power_off(vcpu);
1272 WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1274 spin_unlock(&vcpu->arch.mp_state_lock);
1279 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1280 struct kvm_device_attr *attr)
1284 switch (attr->group) {
1286 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1293 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1294 struct kvm_device_attr *attr)
1298 switch (attr->group) {
1300 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1307 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1308 struct kvm_device_attr *attr)
1312 switch (attr->group) {
1314 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1321 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1322 struct kvm_vcpu_events *events)
1324 memset(events, 0, sizeof(*events));
1326 return __kvm_arm_vcpu_get_events(vcpu, events);
1329 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1330 struct kvm_vcpu_events *events)
1334 /* check whether the reserved field is zero */
1335 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1336 if (events->reserved[i])
1339 /* check whether the pad field is zero */
1340 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1341 if (events->exception.pad[i])
1344 return __kvm_arm_vcpu_set_events(vcpu, events);
1347 long kvm_arch_vcpu_ioctl(struct file *filp,
1348 unsigned int ioctl, unsigned long arg)
1350 struct kvm_vcpu *vcpu = filp->private_data;
1351 void __user *argp = (void __user *)arg;
1352 struct kvm_device_attr attr;
1356 case KVM_ARM_VCPU_INIT: {
1357 struct kvm_vcpu_init init;
1360 if (copy_from_user(&init, argp, sizeof(init)))
1363 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1366 case KVM_SET_ONE_REG:
1367 case KVM_GET_ONE_REG: {
1368 struct kvm_one_reg reg;
1371 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1375 if (copy_from_user(®, argp, sizeof(reg)))
1379 * We could owe a reset due to PSCI. Handle the pending reset
1380 * here to ensure userspace register accesses are ordered after
1383 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1384 kvm_reset_vcpu(vcpu);
1386 if (ioctl == KVM_SET_ONE_REG)
1387 r = kvm_arm_set_reg(vcpu, ®);
1389 r = kvm_arm_get_reg(vcpu, ®);
1392 case KVM_GET_REG_LIST: {
1393 struct kvm_reg_list __user *user_list = argp;
1394 struct kvm_reg_list reg_list;
1398 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1402 if (!kvm_arm_vcpu_is_finalized(vcpu))
1406 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1409 reg_list.n = kvm_arm_num_regs(vcpu);
1410 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1415 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1418 case KVM_SET_DEVICE_ATTR: {
1420 if (copy_from_user(&attr, argp, sizeof(attr)))
1422 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1425 case KVM_GET_DEVICE_ATTR: {
1427 if (copy_from_user(&attr, argp, sizeof(attr)))
1429 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1432 case KVM_HAS_DEVICE_ATTR: {
1434 if (copy_from_user(&attr, argp, sizeof(attr)))
1436 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1439 case KVM_GET_VCPU_EVENTS: {
1440 struct kvm_vcpu_events events;
1442 if (kvm_arm_vcpu_get_events(vcpu, &events))
1445 if (copy_to_user(argp, &events, sizeof(events)))
1450 case KVM_SET_VCPU_EVENTS: {
1451 struct kvm_vcpu_events events;
1453 if (copy_from_user(&events, argp, sizeof(events)))
1456 return kvm_arm_vcpu_set_events(vcpu, &events);
1458 case KVM_ARM_VCPU_FINALIZE: {
1461 if (!kvm_vcpu_initialized(vcpu))
1464 if (get_user(what, (const int __user *)argp))
1467 return kvm_arm_vcpu_finalize(vcpu, what);
1476 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1481 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1482 const struct kvm_memory_slot *memslot)
1484 kvm_flush_remote_tlbs(kvm);
1487 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1488 struct kvm_arm_device_addr *dev_addr)
1490 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1491 case KVM_ARM_DEVICE_VGIC_V2:
1494 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1500 static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1502 switch (attr->group) {
1503 case KVM_ARM_VM_SMCCC_CTRL:
1504 return kvm_vm_smccc_has_attr(kvm, attr);
1510 static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1512 switch (attr->group) {
1513 case KVM_ARM_VM_SMCCC_CTRL:
1514 return kvm_vm_smccc_set_attr(kvm, attr);
1520 int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1522 struct kvm *kvm = filp->private_data;
1523 void __user *argp = (void __user *)arg;
1524 struct kvm_device_attr attr;
1527 case KVM_CREATE_IRQCHIP: {
1531 mutex_lock(&kvm->lock);
1532 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1533 mutex_unlock(&kvm->lock);
1536 case KVM_ARM_SET_DEVICE_ADDR: {
1537 struct kvm_arm_device_addr dev_addr;
1539 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1541 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1543 case KVM_ARM_PREFERRED_TARGET: {
1544 struct kvm_vcpu_init init;
1546 kvm_vcpu_preferred_target(&init);
1548 if (copy_to_user(argp, &init, sizeof(init)))
1553 case KVM_ARM_MTE_COPY_TAGS: {
1554 struct kvm_arm_copy_mte_tags copy_tags;
1556 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1558 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1560 case KVM_ARM_SET_COUNTER_OFFSET: {
1561 struct kvm_arm_counter_offset offset;
1563 if (copy_from_user(&offset, argp, sizeof(offset)))
1565 return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1567 case KVM_HAS_DEVICE_ATTR: {
1568 if (copy_from_user(&attr, argp, sizeof(attr)))
1571 return kvm_vm_has_attr(kvm, &attr);
1573 case KVM_SET_DEVICE_ATTR: {
1574 if (copy_from_user(&attr, argp, sizeof(attr)))
1577 return kvm_vm_set_attr(kvm, &attr);
1584 /* unlocks vcpus from @vcpu_lock_idx and smaller */
1585 static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1587 struct kvm_vcpu *tmp_vcpu;
1589 for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1590 tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1591 mutex_unlock(&tmp_vcpu->mutex);
1595 void unlock_all_vcpus(struct kvm *kvm)
1597 lockdep_assert_held(&kvm->lock);
1599 unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1602 /* Returns true if all vcpus were locked, false otherwise */
1603 bool lock_all_vcpus(struct kvm *kvm)
1605 struct kvm_vcpu *tmp_vcpu;
1608 lockdep_assert_held(&kvm->lock);
1611 * Any time a vcpu is in an ioctl (including running), the
1612 * core KVM code tries to grab the vcpu->mutex.
1614 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1615 * other VCPUs can fiddle with the state while we access it.
1617 kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1618 if (!mutex_trylock(&tmp_vcpu->mutex)) {
1619 unlock_vcpus(kvm, c - 1);
1627 static unsigned long nvhe_percpu_size(void)
1629 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1630 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1633 static unsigned long nvhe_percpu_order(void)
1635 unsigned long size = nvhe_percpu_size();
1637 return size ? get_order(size) : 0;
1640 /* A lookup table holding the hypervisor VA for each vector slot */
1641 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1643 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1645 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1648 static int kvm_init_vector_slots(void)
1653 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1654 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1656 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1657 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1659 if (kvm_system_needs_idmapped_vectors() &&
1660 !is_protected_kvm_enabled()) {
1661 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1662 __BP_HARDEN_HYP_VECS_SZ, &base);
1667 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1668 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1672 static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1674 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1678 * Calculate the raw per-cpu offset without a translation from the
1679 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1680 * so that we can use adr_l to access per-cpu variables in EL2.
1681 * Also drop the KASAN tag which gets in the way...
1683 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1684 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1686 params->mair_el2 = read_sysreg(mair_el1);
1688 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1689 tcr &= ~TCR_T0SZ_MASK;
1690 tcr |= TCR_T0SZ(hyp_va_bits);
1691 params->tcr_el2 = tcr;
1693 params->pgd_pa = kvm_mmu_get_httbr();
1694 if (is_protected_kvm_enabled())
1695 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1697 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1698 params->vttbr = params->vtcr = 0;
1701 * Flush the init params from the data cache because the struct will
1702 * be read while the MMU is off.
1704 kvm_flush_dcache_to_poc(params, sizeof(*params));
1707 static void hyp_install_host_vector(void)
1709 struct kvm_nvhe_init_params *params;
1710 struct arm_smccc_res res;
1712 /* Switch from the HYP stub to our own HYP init vector */
1713 __hyp_set_vectors(kvm_get_idmap_vector());
1716 * Call initialization code, and switch to the full blown HYP code.
1717 * If the cpucaps haven't been finalized yet, something has gone very
1718 * wrong, and hyp will crash and burn when it uses any
1719 * cpus_have_const_cap() wrapper.
1721 BUG_ON(!system_capabilities_finalized());
1722 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1723 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1724 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1727 static void cpu_init_hyp_mode(void)
1729 hyp_install_host_vector();
1732 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1735 if (this_cpu_has_cap(ARM64_SSBS) &&
1736 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1737 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1741 static void cpu_hyp_reset(void)
1743 if (!is_kernel_in_hyp_mode())
1744 __hyp_reset_vectors();
1748 * EL2 vectors can be mapped and rerouted in a number of ways,
1749 * depending on the kernel configuration and CPU present:
1751 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1752 * placed in one of the vector slots, which is executed before jumping
1753 * to the real vectors.
1755 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1756 * containing the hardening sequence is mapped next to the idmap page,
1757 * and executed before jumping to the real vectors.
1759 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1760 * empty slot is selected, mapped next to the idmap page, and
1761 * executed before jumping to the real vectors.
1763 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1764 * VHE, as we don't have hypervisor-specific mappings. If the system
1765 * is VHE and yet selects this capability, it will be ignored.
1767 static void cpu_set_hyp_vector(void)
1769 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1770 void *vector = hyp_spectre_vector_selector[data->slot];
1772 if (!is_protected_kvm_enabled())
1773 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1775 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1778 static void cpu_hyp_init_context(void)
1780 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1782 if (!is_kernel_in_hyp_mode())
1783 cpu_init_hyp_mode();
1786 static void cpu_hyp_init_features(void)
1788 cpu_set_hyp_vector();
1789 kvm_arm_init_debug();
1791 if (is_kernel_in_hyp_mode())
1792 kvm_timer_init_vhe();
1795 kvm_vgic_init_cpu_hardware();
1798 static void cpu_hyp_reinit(void)
1801 cpu_hyp_init_context();
1802 cpu_hyp_init_features();
1805 static void _kvm_arch_hardware_enable(void *discard)
1807 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1809 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1813 int kvm_arch_hardware_enable(void)
1815 int was_enabled = __this_cpu_read(kvm_arm_hardware_enabled);
1817 _kvm_arch_hardware_enable(NULL);
1827 static void _kvm_arch_hardware_disable(void *discard)
1829 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1831 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1835 void kvm_arch_hardware_disable(void)
1837 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1838 kvm_timer_cpu_down();
1839 kvm_vgic_cpu_down();
1842 if (!is_protected_kvm_enabled())
1843 _kvm_arch_hardware_disable(NULL);
1846 #ifdef CONFIG_CPU_PM
1847 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1852 * kvm_arm_hardware_enabled is left with its old value over
1853 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1858 if (__this_cpu_read(kvm_arm_hardware_enabled))
1860 * don't update kvm_arm_hardware_enabled here
1861 * so that the hardware will be re-enabled
1862 * when we resume. See below.
1867 case CPU_PM_ENTER_FAILED:
1869 if (__this_cpu_read(kvm_arm_hardware_enabled))
1870 /* The hardware was enabled before suspend. */
1880 static struct notifier_block hyp_init_cpu_pm_nb = {
1881 .notifier_call = hyp_init_cpu_pm_notifier,
1884 static void __init hyp_cpu_pm_init(void)
1886 if (!is_protected_kvm_enabled())
1887 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1889 static void __init hyp_cpu_pm_exit(void)
1891 if (!is_protected_kvm_enabled())
1892 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1895 static inline void __init hyp_cpu_pm_init(void)
1898 static inline void __init hyp_cpu_pm_exit(void)
1903 static void __init init_cpu_logical_map(void)
1908 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1909 * Only copy the set of online CPUs whose features have been checked
1910 * against the finalized system capabilities. The hypervisor will not
1911 * allow any other CPUs from the `possible` set to boot.
1913 for_each_online_cpu(cpu)
1914 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1917 #define init_psci_0_1_impl_state(config, what) \
1918 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1920 static bool __init init_psci_relay(void)
1923 * If PSCI has not been initialized, protected KVM cannot install
1924 * itself on newly booted CPUs.
1926 if (!psci_ops.get_version) {
1927 kvm_err("Cannot initialize protected mode without PSCI\n");
1931 kvm_host_psci_config.version = psci_ops.get_version();
1933 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1934 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1935 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1936 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1937 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1938 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1943 static int __init init_subsystems(void)
1948 * Enable hardware so that subsystem initialisation can access EL2.
1950 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1953 * Register CPU lower-power notifier
1958 * Init HYP view of VGIC
1960 err = kvm_vgic_hyp_init();
1963 vgic_present = true;
1967 vgic_present = false;
1975 * Init HYP architected timer support
1977 err = kvm_timer_hyp_init(vgic_present);
1981 kvm_register_perf_callbacks(NULL);
1987 if (err || !is_protected_kvm_enabled())
1988 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1993 static void __init teardown_subsystems(void)
1995 kvm_unregister_perf_callbacks();
1999 static void __init teardown_hyp_mode(void)
2004 for_each_possible_cpu(cpu) {
2005 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2006 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2010 static int __init do_pkvm_init(u32 hyp_va_bits)
2012 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2016 cpu_hyp_init_context();
2017 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2018 num_possible_cpus(), kern_hyp_va(per_cpu_base),
2020 cpu_hyp_init_features();
2023 * The stub hypercalls are now disabled, so set our local flag to
2024 * prevent a later re-init attempt in kvm_arch_hardware_enable().
2026 __this_cpu_write(kvm_arm_hardware_enabled, 1);
2032 static u64 get_hyp_id_aa64pfr0_el1(void)
2035 * Track whether the system isn't affected by spectre/meltdown in the
2036 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2037 * Although this is per-CPU, we make it global for simplicity, e.g., not
2038 * to have to worry about vcpu migration.
2040 * Unlike for non-protected VMs, userspace cannot override this for
2043 u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2045 val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2046 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2048 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2049 arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2050 val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2051 arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2056 static void kvm_hyp_init_symbols(void)
2058 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2059 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2060 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2061 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2062 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2063 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2064 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2065 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2066 kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2067 kvm_nvhe_sym(__icache_flags) = __icache_flags;
2068 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2071 static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2073 void *addr = phys_to_virt(hyp_mem_base);
2076 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2080 ret = do_pkvm_init(hyp_va_bits);
2089 /* Inits Hyp-mode on all online CPUs */
2090 static int __init init_hyp_mode(void)
2097 * The protected Hyp-mode cannot be initialized if the memory pool
2098 * allocation has failed.
2100 if (is_protected_kvm_enabled() && !hyp_mem_base)
2104 * Allocate Hyp PGD and setup Hyp identity mapping
2106 err = kvm_mmu_init(&hyp_va_bits);
2111 * Allocate stack pages for Hypervisor-mode
2113 for_each_possible_cpu(cpu) {
2114 unsigned long stack_page;
2116 stack_page = __get_free_page(GFP_KERNEL);
2122 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2126 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2128 for_each_possible_cpu(cpu) {
2132 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2138 page_addr = page_address(page);
2139 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2140 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2144 * Map the Hyp-code called directly from the host
2146 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2147 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2149 kvm_err("Cannot map world-switch code\n");
2153 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2154 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2156 kvm_err("Cannot map .hyp.rodata section\n");
2160 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2161 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2163 kvm_err("Cannot map rodata section\n");
2168 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2169 * section thanks to an assertion in the linker script. Map it RW and
2170 * the rest of .bss RO.
2172 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2173 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2175 kvm_err("Cannot map hyp bss section: %d\n", err);
2179 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2180 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2182 kvm_err("Cannot map bss section\n");
2187 * Map the Hyp stack pages
2189 for_each_possible_cpu(cpu) {
2190 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2191 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2192 unsigned long hyp_addr;
2195 * Allocate a contiguous HYP private VA range for the stack
2196 * and guard page. The allocation is also aligned based on
2197 * the order of its size.
2199 err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr);
2201 kvm_err("Cannot allocate hyp stack guard page\n");
2206 * Since the stack grows downwards, map the stack to the page
2207 * at the higher address and leave the lower guard page
2210 * Any valid stack address now has the PAGE_SHIFT bit as 1
2211 * and addresses corresponding to the guard page have the
2212 * PAGE_SHIFT bit as 0 - this is used for overflow detection.
2214 err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE,
2215 __pa(stack_page), PAGE_HYP);
2217 kvm_err("Cannot map hyp stack\n");
2222 * Save the stack PA in nvhe_init_params. This will be needed
2223 * to recreate the stack mapping in protected nVHE mode.
2224 * __hyp_pa() won't do the right thing there, since the stack
2225 * has been mapped in the flexible private VA space.
2227 params->stack_pa = __pa(stack_page);
2229 params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE);
2232 for_each_possible_cpu(cpu) {
2233 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2234 char *percpu_end = percpu_begin + nvhe_percpu_size();
2236 /* Map Hyp percpu pages */
2237 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2239 kvm_err("Cannot map hyp percpu region\n");
2243 /* Prepare the CPU initialization parameters */
2244 cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2247 kvm_hyp_init_symbols();
2249 if (is_protected_kvm_enabled()) {
2250 init_cpu_logical_map();
2252 if (!init_psci_relay()) {
2257 err = kvm_hyp_init_protection(hyp_va_bits);
2259 kvm_err("Failed to init hyp memory protection\n");
2267 teardown_hyp_mode();
2268 kvm_err("error initializing Hyp mode: %d\n", err);
2272 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2274 struct kvm_vcpu *vcpu;
2277 mpidr &= MPIDR_HWID_BITMASK;
2278 kvm_for_each_vcpu(i, vcpu, kvm) {
2279 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2285 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2287 return irqchip_in_kernel(kvm);
2290 bool kvm_arch_has_irq_bypass(void)
2295 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2296 struct irq_bypass_producer *prod)
2298 struct kvm_kernel_irqfd *irqfd =
2299 container_of(cons, struct kvm_kernel_irqfd, consumer);
2301 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2304 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2305 struct irq_bypass_producer *prod)
2307 struct kvm_kernel_irqfd *irqfd =
2308 container_of(cons, struct kvm_kernel_irqfd, consumer);
2310 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2314 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2316 struct kvm_kernel_irqfd *irqfd =
2317 container_of(cons, struct kvm_kernel_irqfd, consumer);
2319 kvm_arm_halt_guest(irqfd->kvm);
2322 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2324 struct kvm_kernel_irqfd *irqfd =
2325 container_of(cons, struct kvm_kernel_irqfd, consumer);
2327 kvm_arm_resume_guest(irqfd->kvm);
2330 /* Initialize Hyp-mode and memory mappings on all CPUs */
2331 static __init int kvm_arm_init(void)
2336 if (!is_hyp_mode_available()) {
2337 kvm_info("HYP mode not available\n");
2341 if (kvm_get_mode() == KVM_MODE_NONE) {
2342 kvm_info("KVM disabled from command line\n");
2346 err = kvm_sys_reg_table_init();
2348 kvm_info("Error initializing system register tables");
2352 in_hyp_mode = is_kernel_in_hyp_mode();
2354 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2355 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2356 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2357 "Only trusted guests should be used on this system.\n");
2359 err = kvm_set_ipa_limit();
2363 err = kvm_arm_init_sve();
2367 err = kvm_arm_vmid_alloc_init();
2369 kvm_err("Failed to initialize VMID allocator.\n");
2374 err = init_hyp_mode();
2379 err = kvm_init_vector_slots();
2381 kvm_err("Cannot initialise vector slots\n");
2385 err = init_subsystems();
2389 if (is_protected_kvm_enabled()) {
2390 kvm_info("Protected nVHE mode initialized successfully\n");
2391 } else if (in_hyp_mode) {
2392 kvm_info("VHE mode initialized successfully\n");
2394 kvm_info("Hyp mode initialized successfully\n");
2398 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2399 * hypervisor protection is finalized.
2401 err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2408 teardown_subsystems();
2411 teardown_hyp_mode();
2413 kvm_arm_vmid_alloc_free();
2417 static int __init early_kvm_mode_cfg(char *arg)
2422 if (strcmp(arg, "none") == 0) {
2423 kvm_mode = KVM_MODE_NONE;
2427 if (!is_hyp_mode_available()) {
2428 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2432 if (strcmp(arg, "protected") == 0) {
2433 if (!is_kernel_in_hyp_mode())
2434 kvm_mode = KVM_MODE_PROTECTED;
2436 pr_warn_once("Protected KVM not available with VHE\n");
2441 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2442 kvm_mode = KVM_MODE_DEFAULT;
2446 if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2447 kvm_mode = KVM_MODE_NV;
2453 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2455 enum kvm_mode kvm_get_mode(void)
2460 module_init(kvm_arm_init);