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/kmemleak.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_irqfd.h>
22 #include <linux/irqbypass.h>
23 #include <linux/sched/stat.h>
24 #include <linux/psci.h>
25 #include <trace/events/kvm.h>
27 #define CREATE_TRACE_POINTS
28 #include "trace_arm.h"
30 #include <linux/uaccess.h>
31 #include <asm/ptrace.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cacheflush.h>
35 #include <asm/cpufeature.h>
37 #include <asm/kvm_arm.h>
38 #include <asm/kvm_asm.h>
39 #include <asm/kvm_mmu.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;
48 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
52 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
56 static bool vgic_present;
58 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
59 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
61 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
63 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
66 int kvm_arch_hardware_setup(void *opaque)
71 int kvm_arch_check_processor_compat(void *opaque)
76 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
77 struct kvm_enable_cap *cap)
85 case KVM_CAP_ARM_NISV_TO_USER:
87 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
91 mutex_lock(&kvm->lock);
92 if (!system_supports_mte() || kvm->created_vcpus) {
96 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
98 mutex_unlock(&kvm->lock);
100 case KVM_CAP_ARM_SYSTEM_SUSPEND:
102 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
112 static int kvm_arm_default_max_vcpus(void)
114 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
117 static void set_default_spectre(struct kvm *kvm)
120 * The default is to expose CSV2 == 1 if the HW isn't affected.
121 * Although this is a per-CPU feature, we make it global because
122 * asymmetric systems are just a nuisance.
124 * Userspace can override this as long as it doesn't promise
127 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
128 kvm->arch.pfr0_csv2 = 1;
129 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
130 kvm->arch.pfr0_csv3 = 1;
134 * kvm_arch_init_vm - initializes a VM data structure
135 * @kvm: pointer to the KVM struct
137 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
141 ret = kvm_arm_setup_stage2(kvm, type);
145 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
149 ret = kvm_share_hyp(kvm, kvm + 1);
151 goto out_free_stage2_pgd;
153 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL)) {
155 goto out_free_stage2_pgd;
157 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
159 kvm_vgic_early_init(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);
169 kvm_free_stage2_pgd(&kvm->arch.mmu);
173 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
175 return VM_FAULT_SIGBUS;
180 * kvm_arch_destroy_vm - destroy the VM data structure
181 * @kvm: pointer to the KVM struct
183 void kvm_arch_destroy_vm(struct kvm *kvm)
185 bitmap_free(kvm->arch.pmu_filter);
186 free_cpumask_var(kvm->arch.supported_cpus);
188 kvm_vgic_destroy(kvm);
190 kvm_destroy_vcpus(kvm);
192 kvm_unshare_hyp(kvm, kvm + 1);
195 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
199 case KVM_CAP_IRQCHIP:
202 case KVM_CAP_IOEVENTFD:
203 case KVM_CAP_DEVICE_CTRL:
204 case KVM_CAP_USER_MEMORY:
205 case KVM_CAP_SYNC_MMU:
206 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
207 case KVM_CAP_ONE_REG:
208 case KVM_CAP_ARM_PSCI:
209 case KVM_CAP_ARM_PSCI_0_2:
210 case KVM_CAP_READONLY_MEM:
211 case KVM_CAP_MP_STATE:
212 case KVM_CAP_IMMEDIATE_EXIT:
213 case KVM_CAP_VCPU_EVENTS:
214 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
215 case KVM_CAP_ARM_NISV_TO_USER:
216 case KVM_CAP_ARM_INJECT_EXT_DABT:
217 case KVM_CAP_SET_GUEST_DEBUG:
218 case KVM_CAP_VCPU_ATTRIBUTES:
219 case KVM_CAP_PTP_KVM:
220 case KVM_CAP_ARM_SYSTEM_SUSPEND:
223 case KVM_CAP_SET_GUEST_DEBUG2:
224 return KVM_GUESTDBG_VALID_MASK;
225 case KVM_CAP_ARM_SET_DEVICE_ADDR:
228 case KVM_CAP_NR_VCPUS:
230 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
231 * architectures, as it does not always bound it to
232 * KVM_CAP_MAX_VCPUS. It should not matter much because
233 * this is just an advisory value.
235 r = min_t(unsigned int, num_online_cpus(),
236 kvm_arm_default_max_vcpus());
238 case KVM_CAP_MAX_VCPUS:
239 case KVM_CAP_MAX_VCPU_ID:
243 r = kvm_arm_default_max_vcpus();
245 case KVM_CAP_MSI_DEVID:
249 r = kvm->arch.vgic.msis_require_devid;
251 case KVM_CAP_ARM_USER_IRQ:
253 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
254 * (bump this number if adding more devices)
258 case KVM_CAP_ARM_MTE:
259 r = system_supports_mte();
261 case KVM_CAP_STEAL_TIME:
262 r = kvm_arm_pvtime_supported();
264 case KVM_CAP_ARM_EL1_32BIT:
265 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
267 case KVM_CAP_GUEST_DEBUG_HW_BPS:
270 case KVM_CAP_GUEST_DEBUG_HW_WPS:
273 case KVM_CAP_ARM_PMU_V3:
274 r = kvm_arm_support_pmu_v3();
276 case KVM_CAP_ARM_INJECT_SERROR_ESR:
277 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
279 case KVM_CAP_ARM_VM_IPA_SIZE:
280 r = get_kvm_ipa_limit();
282 case KVM_CAP_ARM_SVE:
283 r = system_supports_sve();
285 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
286 case KVM_CAP_ARM_PTRAUTH_GENERIC:
287 r = system_has_full_ptr_auth();
296 long kvm_arch_dev_ioctl(struct file *filp,
297 unsigned int ioctl, unsigned long arg)
302 struct kvm *kvm_arch_alloc_vm(void)
304 size_t sz = sizeof(struct kvm);
307 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
309 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
312 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
314 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
317 if (id >= kvm->max_vcpus)
323 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
327 /* Force users to call KVM_ARM_VCPU_INIT */
328 vcpu->arch.target = -1;
329 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
331 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
334 * Default value for the FP state, will be overloaded at load
335 * time if we support FP (pretty likely)
337 vcpu->arch.fp_state = FP_STATE_FREE;
339 /* Set up the timer */
340 kvm_timer_vcpu_init(vcpu);
342 kvm_pmu_vcpu_init(vcpu);
344 kvm_arm_reset_debug_ptr(vcpu);
346 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
348 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
350 err = kvm_vgic_vcpu_init(vcpu);
354 return kvm_share_hyp(vcpu, vcpu + 1);
357 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
361 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
363 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
364 static_branch_dec(&userspace_irqchip_in_use);
366 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
367 kvm_timer_vcpu_terminate(vcpu);
368 kvm_pmu_vcpu_destroy(vcpu);
370 kvm_arm_vcpu_destroy(vcpu);
373 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
378 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
383 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
385 struct kvm_s2_mmu *mmu;
388 mmu = vcpu->arch.hw_mmu;
389 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
392 * We guarantee that both TLBs and I-cache are private to each
393 * vcpu. If detecting that a vcpu from the same VM has
394 * previously run on the same physical CPU, call into the
395 * hypervisor code to nuke the relevant contexts.
397 * We might get preempted before the vCPU actually runs, but
398 * over-invalidation doesn't affect correctness.
400 if (*last_ran != vcpu->vcpu_id) {
401 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
402 *last_ran = vcpu->vcpu_id;
408 kvm_timer_vcpu_load(vcpu);
410 kvm_vcpu_load_sysregs_vhe(vcpu);
411 kvm_arch_vcpu_load_fp(vcpu);
412 kvm_vcpu_pmu_restore_guest(vcpu);
413 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
414 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
416 if (single_task_running())
417 vcpu_clear_wfx_traps(vcpu);
419 vcpu_set_wfx_traps(vcpu);
421 if (vcpu_has_ptrauth(vcpu))
422 vcpu_ptrauth_disable(vcpu);
423 kvm_arch_vcpu_load_debug_state_flags(vcpu);
425 if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus))
426 vcpu_set_on_unsupported_cpu(vcpu);
429 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
431 kvm_arch_vcpu_put_debug_state_flags(vcpu);
432 kvm_arch_vcpu_put_fp(vcpu);
434 kvm_vcpu_put_sysregs_vhe(vcpu);
435 kvm_timer_vcpu_put(vcpu);
437 kvm_vcpu_pmu_restore_host(vcpu);
438 kvm_arm_vmid_clear_active();
440 vcpu_clear_on_unsupported_cpu(vcpu);
444 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
446 vcpu->arch.mp_state.mp_state = KVM_MP_STATE_STOPPED;
447 kvm_make_request(KVM_REQ_SLEEP, vcpu);
451 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
453 return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_STOPPED;
456 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
458 vcpu->arch.mp_state.mp_state = KVM_MP_STATE_SUSPENDED;
459 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
463 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
465 return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_SUSPENDED;
468 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
469 struct kvm_mp_state *mp_state)
471 *mp_state = vcpu->arch.mp_state;
476 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
477 struct kvm_mp_state *mp_state)
481 switch (mp_state->mp_state) {
482 case KVM_MP_STATE_RUNNABLE:
483 vcpu->arch.mp_state = *mp_state;
485 case KVM_MP_STATE_STOPPED:
486 kvm_arm_vcpu_power_off(vcpu);
488 case KVM_MP_STATE_SUSPENDED:
489 kvm_arm_vcpu_suspend(vcpu);
499 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
500 * @v: The VCPU pointer
502 * If the guest CPU is not waiting for interrupts or an interrupt line is
503 * asserted, the CPU is by definition runnable.
505 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
507 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
508 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
509 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
512 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
514 return vcpu_mode_priv(vcpu);
517 #ifdef CONFIG_GUEST_PERF_EVENTS
518 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
520 return *vcpu_pc(vcpu);
524 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
526 return vcpu->arch.target >= 0;
530 * Handle both the initialisation that is being done when the vcpu is
531 * run for the first time, as well as the updates that must be
532 * performed each time we get a new thread dealing with this vcpu.
534 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
536 struct kvm *kvm = vcpu->kvm;
539 if (!kvm_vcpu_initialized(vcpu))
542 if (!kvm_arm_vcpu_is_finalized(vcpu))
545 ret = kvm_arch_vcpu_run_map_fp(vcpu);
549 if (likely(vcpu_has_run_once(vcpu)))
552 kvm_arm_vcpu_init_debug(vcpu);
554 if (likely(irqchip_in_kernel(kvm))) {
556 * Map the VGIC hardware resources before running a vcpu the
557 * first time on this VM.
559 ret = kvm_vgic_map_resources(kvm);
564 ret = kvm_timer_enable(vcpu);
568 ret = kvm_arm_pmu_v3_enable(vcpu);
572 if (!irqchip_in_kernel(kvm)) {
574 * Tell the rest of the code that there are userspace irqchip
577 static_branch_inc(&userspace_irqchip_in_use);
581 * Initialize traps for protected VMs.
582 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
583 * the code is in place for first run initialization at EL2.
585 if (kvm_vm_is_protected(kvm))
586 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
588 mutex_lock(&kvm->lock);
589 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
590 mutex_unlock(&kvm->lock);
595 bool kvm_arch_intc_initialized(struct kvm *kvm)
597 return vgic_initialized(kvm);
600 void kvm_arm_halt_guest(struct kvm *kvm)
603 struct kvm_vcpu *vcpu;
605 kvm_for_each_vcpu(i, vcpu, kvm)
606 vcpu->arch.pause = true;
607 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
610 void kvm_arm_resume_guest(struct kvm *kvm)
613 struct kvm_vcpu *vcpu;
615 kvm_for_each_vcpu(i, vcpu, kvm) {
616 vcpu->arch.pause = false;
617 __kvm_vcpu_wake_up(vcpu);
621 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
623 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
625 rcuwait_wait_event(wait,
626 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
629 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
630 /* Awaken to handle a signal, request we sleep again later. */
631 kvm_make_request(KVM_REQ_SLEEP, vcpu);
635 * Make sure we will observe a potential reset request if we've
636 * observed a change to the power state. Pairs with the smp_wmb() in
637 * kvm_psci_vcpu_on().
643 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
644 * @vcpu: The VCPU pointer
646 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
647 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
648 * on when a wake event arrives, e.g. there may already be a pending wake event.
650 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
653 * Sync back the state of the GIC CPU interface so that we have
654 * the latest PMR and group enables. This ensures that
655 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
656 * we have pending interrupts, e.g. when determining if the
659 * For the same reason, we want to tell GICv4 that we need
660 * doorbells to be signalled, should an interrupt become pending.
663 kvm_vgic_vmcr_sync(vcpu);
664 vgic_v4_put(vcpu, true);
668 vcpu_clear_flag(vcpu, IN_WFIT);
675 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
677 if (!kvm_arm_vcpu_suspended(vcpu))
683 * The suspend state is sticky; we do not leave it until userspace
684 * explicitly marks the vCPU as runnable. Request that we suspend again
687 kvm_make_request(KVM_REQ_SUSPEND, vcpu);
690 * Check to make sure the vCPU is actually runnable. If so, exit to
691 * userspace informing it of the wakeup condition.
693 if (kvm_arch_vcpu_runnable(vcpu)) {
694 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
695 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
696 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
701 * Otherwise, we were unblocked to process a different event, such as a
702 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
709 * check_vcpu_requests - check and handle pending vCPU requests
710 * @vcpu: the VCPU pointer
712 * Return: 1 if we should enter the guest
713 * 0 if we should exit to userspace
714 * < 0 if we should exit to userspace, where the return value indicates
717 static int check_vcpu_requests(struct kvm_vcpu *vcpu)
719 if (kvm_request_pending(vcpu)) {
720 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
721 kvm_vcpu_sleep(vcpu);
723 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
724 kvm_reset_vcpu(vcpu);
727 * Clear IRQ_PENDING requests that were made to guarantee
728 * that a VCPU sees new virtual interrupts.
730 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
732 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
733 kvm_update_stolen_time(vcpu);
735 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
736 /* The distributor enable bits were changed */
738 vgic_v4_put(vcpu, false);
743 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
744 kvm_pmu_handle_pmcr(vcpu,
745 __vcpu_sys_reg(vcpu, PMCR_EL0));
747 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
748 return kvm_vcpu_suspend(vcpu);
754 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
756 if (likely(!vcpu_mode_is_32bit(vcpu)))
759 return !kvm_supports_32bit_el0();
763 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
764 * @vcpu: The VCPU pointer
765 * @ret: Pointer to write optional return code
767 * Returns: true if the VCPU needs to return to a preemptible + interruptible
768 * and skip guest entry.
770 * This function disambiguates between two different types of exits: exits to a
771 * preemptible + interruptible kernel context and exits to userspace. For an
772 * exit to userspace, this function will write the return code to ret and return
773 * true. For an exit to preemptible + interruptible kernel context (i.e. check
774 * for pending work and re-enter), return true without writing to ret.
776 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
778 struct kvm_run *run = vcpu->run;
781 * If we're using a userspace irqchip, then check if we need
782 * to tell a userspace irqchip about timer or PMU level
783 * changes and if so, exit to userspace (the actual level
784 * state gets updated in kvm_timer_update_run and
785 * kvm_pmu_update_run below).
787 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
788 if (kvm_timer_should_notify_user(vcpu) ||
789 kvm_pmu_should_notify_user(vcpu)) {
791 run->exit_reason = KVM_EXIT_INTR;
796 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
797 run->exit_reason = KVM_EXIT_FAIL_ENTRY;
798 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
799 run->fail_entry.cpu = smp_processor_id();
804 return kvm_request_pending(vcpu) ||
805 xfer_to_guest_mode_work_pending();
809 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
810 * the vCPU is running.
812 * This must be noinstr as instrumentation may make use of RCU, and this is not
813 * safe during the EQS.
815 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
819 guest_state_enter_irqoff();
820 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
821 guest_state_exit_irqoff();
827 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
828 * @vcpu: The VCPU pointer
830 * This function is called through the VCPU_RUN ioctl called from user space. It
831 * will execute VM code in a loop until the time slice for the process is used
832 * or some emulation is needed from user space in which case the function will
833 * return with return value 0 and with the kvm_run structure filled in with the
834 * required data for the requested emulation.
836 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
838 struct kvm_run *run = vcpu->run;
841 if (run->exit_reason == KVM_EXIT_MMIO) {
842 ret = kvm_handle_mmio_return(vcpu);
849 if (run->immediate_exit) {
854 kvm_sigset_activate(vcpu);
857 run->exit_reason = KVM_EXIT_UNKNOWN;
861 * Check conditions before entering the guest
863 ret = xfer_to_guest_mode_handle_work(vcpu);
868 ret = check_vcpu_requests(vcpu);
871 * Preparing the interrupts to be injected also
872 * involves poking the GIC, which must be done in a
873 * non-preemptible context.
878 * The VMID allocator only tracks active VMIDs per
879 * physical CPU, and therefore the VMID allocated may not be
880 * preserved on VMID roll-over if the task was preempted,
881 * making a thread's VMID inactive. So we need to call
882 * kvm_arm_vmid_update() in non-premptible context.
884 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid);
886 kvm_pmu_flush_hwstate(vcpu);
890 kvm_vgic_flush_hwstate(vcpu);
892 kvm_pmu_update_vcpu_events(vcpu);
895 * Ensure we set mode to IN_GUEST_MODE after we disable
896 * interrupts and before the final VCPU requests check.
897 * See the comment in kvm_vcpu_exiting_guest_mode() and
898 * Documentation/virt/kvm/vcpu-requests.rst
900 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
902 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
903 vcpu->mode = OUTSIDE_GUEST_MODE;
904 isb(); /* Ensure work in x_flush_hwstate is committed */
905 kvm_pmu_sync_hwstate(vcpu);
906 if (static_branch_unlikely(&userspace_irqchip_in_use))
907 kvm_timer_sync_user(vcpu);
908 kvm_vgic_sync_hwstate(vcpu);
914 kvm_arm_setup_debug(vcpu);
915 kvm_arch_vcpu_ctxflush_fp(vcpu);
917 /**************************************************************
920 trace_kvm_entry(*vcpu_pc(vcpu));
921 guest_timing_enter_irqoff();
923 ret = kvm_arm_vcpu_enter_exit(vcpu);
925 vcpu->mode = OUTSIDE_GUEST_MODE;
929 *************************************************************/
931 kvm_arm_clear_debug(vcpu);
934 * We must sync the PMU state before the vgic state so
935 * that the vgic can properly sample the updated state of the
938 kvm_pmu_sync_hwstate(vcpu);
941 * Sync the vgic state before syncing the timer state because
942 * the timer code needs to know if the virtual timer
943 * interrupts are active.
945 kvm_vgic_sync_hwstate(vcpu);
948 * Sync the timer hardware state before enabling interrupts as
949 * we don't want vtimer interrupts to race with syncing the
950 * timer virtual interrupt state.
952 if (static_branch_unlikely(&userspace_irqchip_in_use))
953 kvm_timer_sync_user(vcpu);
955 kvm_arch_vcpu_ctxsync_fp(vcpu);
958 * We must ensure that any pending interrupts are taken before
959 * we exit guest timing so that timer ticks are accounted as
960 * guest time. Transiently unmask interrupts so that any
961 * pending interrupts are taken.
963 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
964 * context synchronization event) is necessary to ensure that
965 * pending interrupts are taken.
967 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
973 guest_timing_exit_irqoff();
977 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
979 /* Exit types that need handling before we can be preempted */
980 handle_exit_early(vcpu, ret);
985 * The ARMv8 architecture doesn't give the hypervisor
986 * a mechanism to prevent a guest from dropping to AArch32 EL0
987 * if implemented by the CPU. If we spot the guest in such
988 * state and that we decided it wasn't supposed to do so (like
989 * with the asymmetric AArch32 case), return to userspace with
992 if (vcpu_mode_is_bad_32bit(vcpu)) {
994 * As we have caught the guest red-handed, decide that
995 * it isn't fit for purpose anymore by making the vcpu
996 * invalid. The VMM can try and fix it by issuing a
997 * KVM_ARM_VCPU_INIT if it really wants to.
999 vcpu->arch.target = -1;
1000 ret = ARM_EXCEPTION_IL;
1003 ret = handle_exit(vcpu, ret);
1006 /* Tell userspace about in-kernel device output levels */
1007 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1008 kvm_timer_update_run(vcpu);
1009 kvm_pmu_update_run(vcpu);
1012 kvm_sigset_deactivate(vcpu);
1016 * In the unlikely event that we are returning to userspace
1017 * with pending exceptions or PC adjustment, commit these
1018 * adjustments in order to give userspace a consistent view of
1019 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1020 * being preempt-safe on VHE.
1022 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1023 vcpu_get_flag(vcpu, INCREMENT_PC)))
1024 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1030 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1036 if (number == KVM_ARM_IRQ_CPU_IRQ)
1037 bit_index = __ffs(HCR_VI);
1038 else /* KVM_ARM_IRQ_CPU_FIQ */
1039 bit_index = __ffs(HCR_VF);
1041 hcr = vcpu_hcr(vcpu);
1043 set = test_and_set_bit(bit_index, hcr);
1045 set = test_and_clear_bit(bit_index, hcr);
1048 * If we didn't change anything, no need to wake up or kick other CPUs
1054 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1055 * trigger a world-switch round on the running physical CPU to set the
1056 * virtual IRQ/FIQ fields in the HCR appropriately.
1058 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1059 kvm_vcpu_kick(vcpu);
1064 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1067 u32 irq = irq_level->irq;
1068 unsigned int irq_type, vcpu_idx, irq_num;
1069 int nrcpus = atomic_read(&kvm->online_vcpus);
1070 struct kvm_vcpu *vcpu = NULL;
1071 bool level = irq_level->level;
1073 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1074 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1075 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1076 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1078 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1081 case KVM_ARM_IRQ_TYPE_CPU:
1082 if (irqchip_in_kernel(kvm))
1085 if (vcpu_idx >= nrcpus)
1088 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1092 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1095 return vcpu_interrupt_line(vcpu, irq_num, level);
1096 case KVM_ARM_IRQ_TYPE_PPI:
1097 if (!irqchip_in_kernel(kvm))
1100 if (vcpu_idx >= nrcpus)
1103 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1107 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1110 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1111 case KVM_ARM_IRQ_TYPE_SPI:
1112 if (!irqchip_in_kernel(kvm))
1115 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1118 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1124 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1125 const struct kvm_vcpu_init *init)
1127 unsigned int i, ret;
1128 u32 phys_target = kvm_target_cpu();
1130 if (init->target != phys_target)
1134 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1135 * use the same target.
1137 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1140 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1141 for (i = 0; i < sizeof(init->features) * 8; i++) {
1142 bool set = (init->features[i / 32] & (1 << (i % 32)));
1144 if (set && i >= KVM_VCPU_MAX_FEATURES)
1148 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1149 * use the same feature set.
1151 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1152 test_bit(i, vcpu->arch.features) != set)
1156 set_bit(i, vcpu->arch.features);
1159 vcpu->arch.target = phys_target;
1161 /* Now we know what it is, we can reset it. */
1162 ret = kvm_reset_vcpu(vcpu);
1164 vcpu->arch.target = -1;
1165 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1171 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1172 struct kvm_vcpu_init *init)
1176 ret = kvm_vcpu_set_target(vcpu, init);
1181 * Ensure a rebooted VM will fault in RAM pages and detect if the
1182 * guest MMU is turned off and flush the caches as needed.
1184 * S2FWB enforces all memory accesses to RAM being cacheable,
1185 * ensuring that the data side is always coherent. We still
1186 * need to invalidate the I-cache though, as FWB does *not*
1187 * imply CTR_EL0.DIC.
1189 if (vcpu_has_run_once(vcpu)) {
1190 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1191 stage2_unmap_vm(vcpu->kvm);
1193 icache_inval_all_pou();
1196 vcpu_reset_hcr(vcpu);
1197 vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1200 * Handle the "start in power-off" case.
1202 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1203 kvm_arm_vcpu_power_off(vcpu);
1205 vcpu->arch.mp_state.mp_state = KVM_MP_STATE_RUNNABLE;
1210 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1211 struct kvm_device_attr *attr)
1215 switch (attr->group) {
1217 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1224 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1225 struct kvm_device_attr *attr)
1229 switch (attr->group) {
1231 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1238 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1239 struct kvm_device_attr *attr)
1243 switch (attr->group) {
1245 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1252 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1253 struct kvm_vcpu_events *events)
1255 memset(events, 0, sizeof(*events));
1257 return __kvm_arm_vcpu_get_events(vcpu, events);
1260 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1261 struct kvm_vcpu_events *events)
1265 /* check whether the reserved field is zero */
1266 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1267 if (events->reserved[i])
1270 /* check whether the pad field is zero */
1271 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1272 if (events->exception.pad[i])
1275 return __kvm_arm_vcpu_set_events(vcpu, events);
1278 long kvm_arch_vcpu_ioctl(struct file *filp,
1279 unsigned int ioctl, unsigned long arg)
1281 struct kvm_vcpu *vcpu = filp->private_data;
1282 void __user *argp = (void __user *)arg;
1283 struct kvm_device_attr attr;
1287 case KVM_ARM_VCPU_INIT: {
1288 struct kvm_vcpu_init init;
1291 if (copy_from_user(&init, argp, sizeof(init)))
1294 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1297 case KVM_SET_ONE_REG:
1298 case KVM_GET_ONE_REG: {
1299 struct kvm_one_reg reg;
1302 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1306 if (copy_from_user(®, argp, sizeof(reg)))
1310 * We could owe a reset due to PSCI. Handle the pending reset
1311 * here to ensure userspace register accesses are ordered after
1314 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1315 kvm_reset_vcpu(vcpu);
1317 if (ioctl == KVM_SET_ONE_REG)
1318 r = kvm_arm_set_reg(vcpu, ®);
1320 r = kvm_arm_get_reg(vcpu, ®);
1323 case KVM_GET_REG_LIST: {
1324 struct kvm_reg_list __user *user_list = argp;
1325 struct kvm_reg_list reg_list;
1329 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1333 if (!kvm_arm_vcpu_is_finalized(vcpu))
1337 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1340 reg_list.n = kvm_arm_num_regs(vcpu);
1341 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1346 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1349 case KVM_SET_DEVICE_ATTR: {
1351 if (copy_from_user(&attr, argp, sizeof(attr)))
1353 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1356 case KVM_GET_DEVICE_ATTR: {
1358 if (copy_from_user(&attr, argp, sizeof(attr)))
1360 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1363 case KVM_HAS_DEVICE_ATTR: {
1365 if (copy_from_user(&attr, argp, sizeof(attr)))
1367 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1370 case KVM_GET_VCPU_EVENTS: {
1371 struct kvm_vcpu_events events;
1373 if (kvm_arm_vcpu_get_events(vcpu, &events))
1376 if (copy_to_user(argp, &events, sizeof(events)))
1381 case KVM_SET_VCPU_EVENTS: {
1382 struct kvm_vcpu_events events;
1384 if (copy_from_user(&events, argp, sizeof(events)))
1387 return kvm_arm_vcpu_set_events(vcpu, &events);
1389 case KVM_ARM_VCPU_FINALIZE: {
1392 if (!kvm_vcpu_initialized(vcpu))
1395 if (get_user(what, (const int __user *)argp))
1398 return kvm_arm_vcpu_finalize(vcpu, what);
1407 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1412 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1413 const struct kvm_memory_slot *memslot)
1415 kvm_flush_remote_tlbs(kvm);
1418 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1419 struct kvm_arm_device_addr *dev_addr)
1421 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1422 case KVM_ARM_DEVICE_VGIC_V2:
1425 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1431 long kvm_arch_vm_ioctl(struct file *filp,
1432 unsigned int ioctl, unsigned long arg)
1434 struct kvm *kvm = filp->private_data;
1435 void __user *argp = (void __user *)arg;
1438 case KVM_CREATE_IRQCHIP: {
1442 mutex_lock(&kvm->lock);
1443 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1444 mutex_unlock(&kvm->lock);
1447 case KVM_ARM_SET_DEVICE_ADDR: {
1448 struct kvm_arm_device_addr dev_addr;
1450 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1452 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1454 case KVM_ARM_PREFERRED_TARGET: {
1455 struct kvm_vcpu_init init;
1457 kvm_vcpu_preferred_target(&init);
1459 if (copy_to_user(argp, &init, sizeof(init)))
1464 case KVM_ARM_MTE_COPY_TAGS: {
1465 struct kvm_arm_copy_mte_tags copy_tags;
1467 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1469 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1476 static unsigned long nvhe_percpu_size(void)
1478 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1479 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1482 static unsigned long nvhe_percpu_order(void)
1484 unsigned long size = nvhe_percpu_size();
1486 return size ? get_order(size) : 0;
1489 /* A lookup table holding the hypervisor VA for each vector slot */
1490 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1492 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1494 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1497 static int kvm_init_vector_slots(void)
1502 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1503 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1505 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1506 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1508 if (kvm_system_needs_idmapped_vectors() &&
1509 !is_protected_kvm_enabled()) {
1510 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1511 __BP_HARDEN_HYP_VECS_SZ, &base);
1516 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1517 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1521 static void cpu_prepare_hyp_mode(int cpu)
1523 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1527 * Calculate the raw per-cpu offset without a translation from the
1528 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1529 * so that we can use adr_l to access per-cpu variables in EL2.
1530 * Also drop the KASAN tag which gets in the way...
1532 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1533 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1535 params->mair_el2 = read_sysreg(mair_el1);
1538 * The ID map may be configured to use an extended virtual address
1539 * range. This is only the case if system RAM is out of range for the
1540 * currently configured page size and VA_BITS, in which case we will
1541 * also need the extended virtual range for the HYP ID map, or we won't
1542 * be able to enable the EL2 MMU.
1544 * However, at EL2, there is only one TTBR register, and we can't switch
1545 * between translation tables *and* update TCR_EL2.T0SZ at the same
1546 * time. Bottom line: we need to use the extended range with *both* our
1547 * translation tables.
1549 * So use the same T0SZ value we use for the ID map.
1551 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1552 tcr &= ~TCR_T0SZ_MASK;
1553 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1554 params->tcr_el2 = tcr;
1556 params->pgd_pa = kvm_mmu_get_httbr();
1557 if (is_protected_kvm_enabled())
1558 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1560 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1561 params->vttbr = params->vtcr = 0;
1564 * Flush the init params from the data cache because the struct will
1565 * be read while the MMU is off.
1567 kvm_flush_dcache_to_poc(params, sizeof(*params));
1570 static void hyp_install_host_vector(void)
1572 struct kvm_nvhe_init_params *params;
1573 struct arm_smccc_res res;
1575 /* Switch from the HYP stub to our own HYP init vector */
1576 __hyp_set_vectors(kvm_get_idmap_vector());
1579 * Call initialization code, and switch to the full blown HYP code.
1580 * If the cpucaps haven't been finalized yet, something has gone very
1581 * wrong, and hyp will crash and burn when it uses any
1582 * cpus_have_const_cap() wrapper.
1584 BUG_ON(!system_capabilities_finalized());
1585 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1586 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1587 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1590 static void cpu_init_hyp_mode(void)
1592 hyp_install_host_vector();
1595 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1598 if (this_cpu_has_cap(ARM64_SSBS) &&
1599 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1600 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1604 static void cpu_hyp_reset(void)
1606 if (!is_kernel_in_hyp_mode())
1607 __hyp_reset_vectors();
1611 * EL2 vectors can be mapped and rerouted in a number of ways,
1612 * depending on the kernel configuration and CPU present:
1614 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1615 * placed in one of the vector slots, which is executed before jumping
1616 * to the real vectors.
1618 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1619 * containing the hardening sequence is mapped next to the idmap page,
1620 * and executed before jumping to the real vectors.
1622 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1623 * empty slot is selected, mapped next to the idmap page, and
1624 * executed before jumping to the real vectors.
1626 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1627 * VHE, as we don't have hypervisor-specific mappings. If the system
1628 * is VHE and yet selects this capability, it will be ignored.
1630 static void cpu_set_hyp_vector(void)
1632 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1633 void *vector = hyp_spectre_vector_selector[data->slot];
1635 if (!is_protected_kvm_enabled())
1636 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1638 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1641 static void cpu_hyp_init_context(void)
1643 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1645 if (!is_kernel_in_hyp_mode())
1646 cpu_init_hyp_mode();
1649 static void cpu_hyp_init_features(void)
1651 cpu_set_hyp_vector();
1652 kvm_arm_init_debug();
1654 if (is_kernel_in_hyp_mode())
1655 kvm_timer_init_vhe();
1658 kvm_vgic_init_cpu_hardware();
1661 static void cpu_hyp_reinit(void)
1664 cpu_hyp_init_context();
1665 cpu_hyp_init_features();
1668 static void _kvm_arch_hardware_enable(void *discard)
1670 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1672 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1676 int kvm_arch_hardware_enable(void)
1678 _kvm_arch_hardware_enable(NULL);
1682 static void _kvm_arch_hardware_disable(void *discard)
1684 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1686 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1690 void kvm_arch_hardware_disable(void)
1692 if (!is_protected_kvm_enabled())
1693 _kvm_arch_hardware_disable(NULL);
1696 #ifdef CONFIG_CPU_PM
1697 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1702 * kvm_arm_hardware_enabled is left with its old value over
1703 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1708 if (__this_cpu_read(kvm_arm_hardware_enabled))
1710 * don't update kvm_arm_hardware_enabled here
1711 * so that the hardware will be re-enabled
1712 * when we resume. See below.
1717 case CPU_PM_ENTER_FAILED:
1719 if (__this_cpu_read(kvm_arm_hardware_enabled))
1720 /* The hardware was enabled before suspend. */
1730 static struct notifier_block hyp_init_cpu_pm_nb = {
1731 .notifier_call = hyp_init_cpu_pm_notifier,
1734 static void hyp_cpu_pm_init(void)
1736 if (!is_protected_kvm_enabled())
1737 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1739 static void hyp_cpu_pm_exit(void)
1741 if (!is_protected_kvm_enabled())
1742 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1745 static inline void hyp_cpu_pm_init(void)
1748 static inline void hyp_cpu_pm_exit(void)
1753 static void init_cpu_logical_map(void)
1758 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1759 * Only copy the set of online CPUs whose features have been checked
1760 * against the finalized system capabilities. The hypervisor will not
1761 * allow any other CPUs from the `possible` set to boot.
1763 for_each_online_cpu(cpu)
1764 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1767 #define init_psci_0_1_impl_state(config, what) \
1768 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1770 static bool init_psci_relay(void)
1773 * If PSCI has not been initialized, protected KVM cannot install
1774 * itself on newly booted CPUs.
1776 if (!psci_ops.get_version) {
1777 kvm_err("Cannot initialize protected mode without PSCI\n");
1781 kvm_host_psci_config.version = psci_ops.get_version();
1783 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1784 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1785 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1786 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1787 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1788 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1793 static int init_subsystems(void)
1798 * Enable hardware so that subsystem initialisation can access EL2.
1800 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1803 * Register CPU lower-power notifier
1808 * Init HYP view of VGIC
1810 err = kvm_vgic_hyp_init();
1813 vgic_present = true;
1817 vgic_present = false;
1825 * Init HYP architected timer support
1827 err = kvm_timer_hyp_init(vgic_present);
1831 kvm_register_perf_callbacks(NULL);
1834 if (err || !is_protected_kvm_enabled())
1835 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1840 static void teardown_hyp_mode(void)
1845 for_each_possible_cpu(cpu) {
1846 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1847 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1851 static int do_pkvm_init(u32 hyp_va_bits)
1853 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1857 cpu_hyp_init_context();
1858 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1859 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1861 cpu_hyp_init_features();
1864 * The stub hypercalls are now disabled, so set our local flag to
1865 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1867 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1873 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1875 void *addr = phys_to_virt(hyp_mem_base);
1878 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1879 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1880 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1881 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1882 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
1883 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1884 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1885 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1887 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1891 ret = do_pkvm_init(hyp_va_bits);
1901 * Inits Hyp-mode on all online CPUs
1903 static int init_hyp_mode(void)
1910 * The protected Hyp-mode cannot be initialized if the memory pool
1911 * allocation has failed.
1913 if (is_protected_kvm_enabled() && !hyp_mem_base)
1917 * Allocate Hyp PGD and setup Hyp identity mapping
1919 err = kvm_mmu_init(&hyp_va_bits);
1924 * Allocate stack pages for Hypervisor-mode
1926 for_each_possible_cpu(cpu) {
1927 unsigned long stack_page;
1929 stack_page = __get_free_page(GFP_KERNEL);
1935 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1939 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1941 for_each_possible_cpu(cpu) {
1945 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1951 page_addr = page_address(page);
1952 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1953 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1957 * Map the Hyp-code called directly from the host
1959 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1960 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1962 kvm_err("Cannot map world-switch code\n");
1966 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1967 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1969 kvm_err("Cannot map .hyp.rodata section\n");
1973 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1974 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1976 kvm_err("Cannot map rodata section\n");
1981 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1982 * section thanks to an assertion in the linker script. Map it RW and
1983 * the rest of .bss RO.
1985 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1986 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1988 kvm_err("Cannot map hyp bss section: %d\n", err);
1992 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1993 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1995 kvm_err("Cannot map bss section\n");
2000 * Map the Hyp stack pages
2002 for_each_possible_cpu(cpu) {
2003 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2004 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2005 unsigned long hyp_addr;
2008 * Allocate a contiguous HYP private VA range for the stack
2009 * and guard page. The allocation is also aligned based on
2010 * the order of its size.
2012 err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr);
2014 kvm_err("Cannot allocate hyp stack guard page\n");
2019 * Since the stack grows downwards, map the stack to the page
2020 * at the higher address and leave the lower guard page
2023 * Any valid stack address now has the PAGE_SHIFT bit as 1
2024 * and addresses corresponding to the guard page have the
2025 * PAGE_SHIFT bit as 0 - this is used for overflow detection.
2027 err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE,
2028 __pa(stack_page), PAGE_HYP);
2030 kvm_err("Cannot map hyp stack\n");
2035 * Save the stack PA in nvhe_init_params. This will be needed
2036 * to recreate the stack mapping in protected nVHE mode.
2037 * __hyp_pa() won't do the right thing there, since the stack
2038 * has been mapped in the flexible private VA space.
2040 params->stack_pa = __pa(stack_page);
2042 params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE);
2045 for_each_possible_cpu(cpu) {
2046 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
2047 char *percpu_end = percpu_begin + nvhe_percpu_size();
2049 /* Map Hyp percpu pages */
2050 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2052 kvm_err("Cannot map hyp percpu region\n");
2056 /* Prepare the CPU initialization parameters */
2057 cpu_prepare_hyp_mode(cpu);
2060 if (is_protected_kvm_enabled()) {
2061 init_cpu_logical_map();
2063 if (!init_psci_relay()) {
2069 if (is_protected_kvm_enabled()) {
2070 err = kvm_hyp_init_protection(hyp_va_bits);
2072 kvm_err("Failed to init hyp memory protection\n");
2080 teardown_hyp_mode();
2081 kvm_err("error initializing Hyp mode: %d\n", err);
2085 static void _kvm_host_prot_finalize(void *arg)
2089 if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
2090 WRITE_ONCE(*err, -EINVAL);
2093 static int pkvm_drop_host_privileges(void)
2098 * Flip the static key upfront as that may no longer be possible
2099 * once the host stage 2 is installed.
2101 static_branch_enable(&kvm_protected_mode_initialized);
2102 on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2106 static int finalize_hyp_mode(void)
2108 if (!is_protected_kvm_enabled())
2112 * Exclude HYP sections from kmemleak so that they don't get peeked
2113 * at, which would end badly once inaccessible.
2115 kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2116 kmemleak_free_part_phys(hyp_mem_base, hyp_mem_size);
2117 return pkvm_drop_host_privileges();
2120 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2122 struct kvm_vcpu *vcpu;
2125 mpidr &= MPIDR_HWID_BITMASK;
2126 kvm_for_each_vcpu(i, vcpu, kvm) {
2127 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2133 bool kvm_arch_has_irq_bypass(void)
2138 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2139 struct irq_bypass_producer *prod)
2141 struct kvm_kernel_irqfd *irqfd =
2142 container_of(cons, struct kvm_kernel_irqfd, consumer);
2144 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2147 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2148 struct irq_bypass_producer *prod)
2150 struct kvm_kernel_irqfd *irqfd =
2151 container_of(cons, struct kvm_kernel_irqfd, consumer);
2153 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2157 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2159 struct kvm_kernel_irqfd *irqfd =
2160 container_of(cons, struct kvm_kernel_irqfd, consumer);
2162 kvm_arm_halt_guest(irqfd->kvm);
2165 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2167 struct kvm_kernel_irqfd *irqfd =
2168 container_of(cons, struct kvm_kernel_irqfd, consumer);
2170 kvm_arm_resume_guest(irqfd->kvm);
2174 * Initialize Hyp-mode and memory mappings on all CPUs.
2176 int kvm_arch_init(void *opaque)
2181 if (!is_hyp_mode_available()) {
2182 kvm_info("HYP mode not available\n");
2186 if (kvm_get_mode() == KVM_MODE_NONE) {
2187 kvm_info("KVM disabled from command line\n");
2191 err = kvm_sys_reg_table_init();
2193 kvm_info("Error initializing system register tables");
2197 in_hyp_mode = is_kernel_in_hyp_mode();
2199 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2200 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2201 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2202 "Only trusted guests should be used on this system.\n");
2204 err = kvm_set_ipa_limit();
2208 err = kvm_arm_init_sve();
2212 err = kvm_arm_vmid_alloc_init();
2214 kvm_err("Failed to initialize VMID allocator.\n");
2219 err = init_hyp_mode();
2224 err = kvm_init_vector_slots();
2226 kvm_err("Cannot initialise vector slots\n");
2230 err = init_subsystems();
2235 err = finalize_hyp_mode();
2237 kvm_err("Failed to finalize Hyp protection\n");
2242 if (is_protected_kvm_enabled()) {
2243 kvm_info("Protected nVHE mode initialized successfully\n");
2244 } else if (in_hyp_mode) {
2245 kvm_info("VHE mode initialized successfully\n");
2247 kvm_info("Hyp mode initialized successfully\n");
2255 teardown_hyp_mode();
2257 kvm_arm_vmid_alloc_free();
2261 /* NOP: Compiling as a module not supported */
2262 void kvm_arch_exit(void)
2264 kvm_unregister_perf_callbacks();
2267 static int __init early_kvm_mode_cfg(char *arg)
2272 if (strcmp(arg, "none") == 0) {
2273 kvm_mode = KVM_MODE_NONE;
2277 if (!is_hyp_mode_available()) {
2278 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2282 if (strcmp(arg, "protected") == 0) {
2283 if (!is_kernel_in_hyp_mode())
2284 kvm_mode = KVM_MODE_PROTECTED;
2286 pr_warn_once("Protected KVM not available with VHE\n");
2291 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2292 kvm_mode = KVM_MODE_DEFAULT;
2298 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2300 enum kvm_mode kvm_get_mode(void)
2305 static int arm_init(void)
2307 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2311 module_init(arm_init);