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/errno.h>
10 #include <linux/err.h>
11 #include <linux/kvm_host.h>
12 #include <linux/list.h>
13 #include <linux/module.h>
14 #include <linux/vmalloc.h>
16 #include <linux/mman.h>
17 #include <linux/sched.h>
18 #include <linux/kvm.h>
19 #include <linux/kvm_irqfd.h>
20 #include <linux/irqbypass.h>
21 #include <linux/sched/stat.h>
22 #include <linux/psci.h>
23 #include <trace/events/kvm.h>
25 #define CREATE_TRACE_POINTS
26 #include "trace_arm.h"
28 #include <linux/uaccess.h>
29 #include <asm/ptrace.h>
31 #include <asm/tlbflush.h>
32 #include <asm/cacheflush.h>
33 #include <asm/cpufeature.h>
35 #include <asm/kvm_arm.h>
36 #include <asm/kvm_asm.h>
37 #include <asm/kvm_mmu.h>
38 #include <asm/kvm_emulate.h>
39 #include <asm/sections.h>
41 #include <kvm/arm_hypercalls.h>
42 #include <kvm/arm_pmu.h>
43 #include <kvm/arm_psci.h>
46 __asm__(".arch_extension virt");
49 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
50 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
52 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
55 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
56 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
58 /* The VMID used in the VTTBR */
59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
60 static u32 kvm_next_vmid;
61 static DEFINE_SPINLOCK(kvm_vmid_lock);
63 static bool vgic_present;
65 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
66 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
68 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
70 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
73 int kvm_arch_hardware_setup(void *opaque)
78 int kvm_arch_check_processor_compat(void *opaque)
83 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
84 struct kvm_enable_cap *cap)
92 case KVM_CAP_ARM_NISV_TO_USER:
94 kvm->arch.return_nisv_io_abort_to_user = true;
104 static int kvm_arm_default_max_vcpus(void)
106 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
109 static void set_default_spectre(struct kvm *kvm)
112 * The default is to expose CSV2 == 1 if the HW isn't affected.
113 * Although this is a per-CPU feature, we make it global because
114 * asymmetric systems are just a nuisance.
116 * Userspace can override this as long as it doesn't promise
119 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
120 kvm->arch.pfr0_csv2 = 1;
121 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
122 kvm->arch.pfr0_csv3 = 1;
126 * kvm_arch_init_vm - initializes a VM data structure
127 * @kvm: pointer to the KVM struct
129 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
133 ret = kvm_arm_setup_stage2(kvm, type);
137 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
141 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
143 goto out_free_stage2_pgd;
145 kvm_vgic_early_init(kvm);
147 /* The maximum number of VCPUs is limited by the host's GIC model */
148 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
150 set_default_spectre(kvm);
154 kvm_free_stage2_pgd(&kvm->arch.mmu);
158 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
160 return VM_FAULT_SIGBUS;
165 * kvm_arch_destroy_vm - destroy the VM data structure
166 * @kvm: pointer to the KVM struct
168 void kvm_arch_destroy_vm(struct kvm *kvm)
172 bitmap_free(kvm->arch.pmu_filter);
174 kvm_vgic_destroy(kvm);
176 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
178 kvm_vcpu_destroy(kvm->vcpus[i]);
179 kvm->vcpus[i] = NULL;
182 atomic_set(&kvm->online_vcpus, 0);
185 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
189 case KVM_CAP_IRQCHIP:
192 case KVM_CAP_IOEVENTFD:
193 case KVM_CAP_DEVICE_CTRL:
194 case KVM_CAP_USER_MEMORY:
195 case KVM_CAP_SYNC_MMU:
196 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
197 case KVM_CAP_ONE_REG:
198 case KVM_CAP_ARM_PSCI:
199 case KVM_CAP_ARM_PSCI_0_2:
200 case KVM_CAP_READONLY_MEM:
201 case KVM_CAP_MP_STATE:
202 case KVM_CAP_IMMEDIATE_EXIT:
203 case KVM_CAP_VCPU_EVENTS:
204 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
205 case KVM_CAP_ARM_NISV_TO_USER:
206 case KVM_CAP_ARM_INJECT_EXT_DABT:
207 case KVM_CAP_SET_GUEST_DEBUG:
208 case KVM_CAP_VCPU_ATTRIBUTES:
209 case KVM_CAP_PTP_KVM:
212 case KVM_CAP_ARM_SET_DEVICE_ADDR:
215 case KVM_CAP_NR_VCPUS:
216 r = num_online_cpus();
218 case KVM_CAP_MAX_VCPUS:
219 case KVM_CAP_MAX_VCPU_ID:
221 r = kvm->arch.max_vcpus;
223 r = kvm_arm_default_max_vcpus();
225 case KVM_CAP_MSI_DEVID:
229 r = kvm->arch.vgic.msis_require_devid;
231 case KVM_CAP_ARM_USER_IRQ:
233 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
234 * (bump this number if adding more devices)
238 case KVM_CAP_STEAL_TIME:
239 r = kvm_arm_pvtime_supported();
241 case KVM_CAP_ARM_EL1_32BIT:
242 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
244 case KVM_CAP_GUEST_DEBUG_HW_BPS:
247 case KVM_CAP_GUEST_DEBUG_HW_WPS:
250 case KVM_CAP_ARM_PMU_V3:
251 r = kvm_arm_support_pmu_v3();
253 case KVM_CAP_ARM_INJECT_SERROR_ESR:
254 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
256 case KVM_CAP_ARM_VM_IPA_SIZE:
257 r = get_kvm_ipa_limit();
259 case KVM_CAP_ARM_SVE:
260 r = system_supports_sve();
262 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
263 case KVM_CAP_ARM_PTRAUTH_GENERIC:
264 r = system_has_full_ptr_auth();
273 long kvm_arch_dev_ioctl(struct file *filp,
274 unsigned int ioctl, unsigned long arg)
279 struct kvm *kvm_arch_alloc_vm(void)
282 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
284 return vzalloc(sizeof(struct kvm));
287 void kvm_arch_free_vm(struct kvm *kvm)
295 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
297 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
300 if (id >= kvm->arch.max_vcpus)
306 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
310 /* Force users to call KVM_ARM_VCPU_INIT */
311 vcpu->arch.target = -1;
312 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
314 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
316 /* Set up the timer */
317 kvm_timer_vcpu_init(vcpu);
319 kvm_pmu_vcpu_init(vcpu);
321 kvm_arm_reset_debug_ptr(vcpu);
323 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
325 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
327 err = kvm_vgic_vcpu_init(vcpu);
331 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
334 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
338 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
340 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
341 static_branch_dec(&userspace_irqchip_in_use);
343 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
344 kvm_timer_vcpu_terminate(vcpu);
345 kvm_pmu_vcpu_destroy(vcpu);
347 kvm_arm_vcpu_destroy(vcpu);
350 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
352 return kvm_timer_is_pending(vcpu);
355 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
358 * If we're about to block (most likely because we've just hit a
359 * WFI), we need to sync back the state of the GIC CPU interface
360 * so that we have the latest PMR and group enables. This ensures
361 * that kvm_arch_vcpu_runnable has up-to-date data to decide
362 * whether we have pending interrupts.
364 * For the same reason, we want to tell GICv4 that we need
365 * doorbells to be signalled, should an interrupt become pending.
368 kvm_vgic_vmcr_sync(vcpu);
369 vgic_v4_put(vcpu, true);
373 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
380 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
382 struct kvm_s2_mmu *mmu;
385 mmu = vcpu->arch.hw_mmu;
386 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
389 * We guarantee that both TLBs and I-cache are private to each
390 * vcpu. If detecting that a vcpu from the same VM has
391 * previously run on the same physical CPU, call into the
392 * hypervisor code to nuke the relevant contexts.
394 * We might get preempted before the vCPU actually runs, but
395 * over-invalidation doesn't affect correctness.
397 if (*last_ran != vcpu->vcpu_id) {
398 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
399 *last_ran = vcpu->vcpu_id;
405 kvm_timer_vcpu_load(vcpu);
407 kvm_vcpu_load_sysregs_vhe(vcpu);
408 kvm_arch_vcpu_load_fp(vcpu);
409 kvm_vcpu_pmu_restore_guest(vcpu);
410 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
411 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
413 if (single_task_running())
414 vcpu_clear_wfx_traps(vcpu);
416 vcpu_set_wfx_traps(vcpu);
418 if (vcpu_has_ptrauth(vcpu))
419 vcpu_ptrauth_disable(vcpu);
420 kvm_arch_vcpu_load_debug_state_flags(vcpu);
423 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
425 kvm_arch_vcpu_put_debug_state_flags(vcpu);
426 kvm_arch_vcpu_put_fp(vcpu);
428 kvm_vcpu_put_sysregs_vhe(vcpu);
429 kvm_timer_vcpu_put(vcpu);
431 kvm_vcpu_pmu_restore_host(vcpu);
436 static void vcpu_power_off(struct kvm_vcpu *vcpu)
438 vcpu->arch.power_off = true;
439 kvm_make_request(KVM_REQ_SLEEP, vcpu);
443 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
444 struct kvm_mp_state *mp_state)
446 if (vcpu->arch.power_off)
447 mp_state->mp_state = KVM_MP_STATE_STOPPED;
449 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
454 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
455 struct kvm_mp_state *mp_state)
459 switch (mp_state->mp_state) {
460 case KVM_MP_STATE_RUNNABLE:
461 vcpu->arch.power_off = false;
463 case KVM_MP_STATE_STOPPED:
464 vcpu_power_off(vcpu);
474 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
475 * @v: The VCPU pointer
477 * If the guest CPU is not waiting for interrupts or an interrupt line is
478 * asserted, the CPU is by definition runnable.
480 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
482 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
483 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
484 && !v->arch.power_off && !v->arch.pause);
487 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
489 return vcpu_mode_priv(vcpu);
492 /* Just ensure a guest exit from a particular CPU */
493 static void exit_vm_noop(void *info)
497 void force_vm_exit(const cpumask_t *mask)
500 smp_call_function_many(mask, exit_vm_noop, NULL, true);
505 * need_new_vmid_gen - check that the VMID is still valid
506 * @vmid: The VMID to check
508 * return true if there is a new generation of VMIDs being used
510 * The hardware supports a limited set of values with the value zero reserved
511 * for the host, so we check if an assigned value belongs to a previous
512 * generation, which requires us to assign a new value. If we're the first to
513 * use a VMID for the new generation, we must flush necessary caches and TLBs
516 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
518 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
519 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
520 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
524 * update_vmid - Update the vmid with a valid VMID for the current generation
525 * @vmid: The stage-2 VMID information struct
527 static void update_vmid(struct kvm_vmid *vmid)
529 if (!need_new_vmid_gen(vmid))
532 spin_lock(&kvm_vmid_lock);
535 * We need to re-check the vmid_gen here to ensure that if another vcpu
536 * already allocated a valid vmid for this vm, then this vcpu should
539 if (!need_new_vmid_gen(vmid)) {
540 spin_unlock(&kvm_vmid_lock);
544 /* First user of a new VMID generation? */
545 if (unlikely(kvm_next_vmid == 0)) {
546 atomic64_inc(&kvm_vmid_gen);
550 * On SMP we know no other CPUs can use this CPU's or each
551 * other's VMID after force_vm_exit returns since the
552 * kvm_vmid_lock blocks them from reentry to the guest.
554 force_vm_exit(cpu_all_mask);
556 * Now broadcast TLB + ICACHE invalidation over the inner
557 * shareable domain to make sure all data structures are
560 kvm_call_hyp(__kvm_flush_vm_context);
563 vmid->vmid = kvm_next_vmid;
565 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
568 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
570 spin_unlock(&kvm_vmid_lock);
573 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
575 struct kvm *kvm = vcpu->kvm;
578 if (likely(vcpu->arch.has_run_once))
581 if (!kvm_arm_vcpu_is_finalized(vcpu))
584 vcpu->arch.has_run_once = true;
586 kvm_arm_vcpu_init_debug(vcpu);
588 if (likely(irqchip_in_kernel(kvm))) {
590 * Map the VGIC hardware resources before running a vcpu the
591 * first time on this VM.
593 ret = kvm_vgic_map_resources(kvm);
598 * Tell the rest of the code that there are userspace irqchip
601 static_branch_inc(&userspace_irqchip_in_use);
604 ret = kvm_timer_enable(vcpu);
608 ret = kvm_arm_pmu_v3_enable(vcpu);
613 bool kvm_arch_intc_initialized(struct kvm *kvm)
615 return vgic_initialized(kvm);
618 void kvm_arm_halt_guest(struct kvm *kvm)
621 struct kvm_vcpu *vcpu;
623 kvm_for_each_vcpu(i, vcpu, kvm)
624 vcpu->arch.pause = true;
625 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
628 void kvm_arm_resume_guest(struct kvm *kvm)
631 struct kvm_vcpu *vcpu;
633 kvm_for_each_vcpu(i, vcpu, kvm) {
634 vcpu->arch.pause = false;
635 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
639 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
641 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
643 rcuwait_wait_event(wait,
644 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
647 if (vcpu->arch.power_off || vcpu->arch.pause) {
648 /* Awaken to handle a signal, request we sleep again later. */
649 kvm_make_request(KVM_REQ_SLEEP, vcpu);
653 * Make sure we will observe a potential reset request if we've
654 * observed a change to the power state. Pairs with the smp_wmb() in
655 * kvm_psci_vcpu_on().
660 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
662 return vcpu->arch.target >= 0;
665 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
667 if (kvm_request_pending(vcpu)) {
668 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
669 vcpu_req_sleep(vcpu);
671 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
672 kvm_reset_vcpu(vcpu);
675 * Clear IRQ_PENDING requests that were made to guarantee
676 * that a VCPU sees new virtual interrupts.
678 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
680 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
681 kvm_update_stolen_time(vcpu);
683 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
684 /* The distributor enable bits were changed */
686 vgic_v4_put(vcpu, false);
694 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
695 * @vcpu: The VCPU pointer
697 * This function is called through the VCPU_RUN ioctl called from user space. It
698 * will execute VM code in a loop until the time slice for the process is used
699 * or some emulation is needed from user space in which case the function will
700 * return with return value 0 and with the kvm_run structure filled in with the
701 * required data for the requested emulation.
703 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
705 struct kvm_run *run = vcpu->run;
708 if (unlikely(!kvm_vcpu_initialized(vcpu)))
711 ret = kvm_vcpu_first_run_init(vcpu);
715 if (run->exit_reason == KVM_EXIT_MMIO) {
716 ret = kvm_handle_mmio_return(vcpu);
721 if (run->immediate_exit)
726 kvm_sigset_activate(vcpu);
729 run->exit_reason = KVM_EXIT_UNKNOWN;
732 * Check conditions before entering the guest
736 update_vmid(&vcpu->arch.hw_mmu->vmid);
738 check_vcpu_requests(vcpu);
741 * Preparing the interrupts to be injected also
742 * involves poking the GIC, which must be done in a
743 * non-preemptible context.
747 kvm_pmu_flush_hwstate(vcpu);
751 kvm_vgic_flush_hwstate(vcpu);
754 * Exit if we have a signal pending so that we can deliver the
755 * signal to user space.
757 if (signal_pending(current)) {
759 run->exit_reason = KVM_EXIT_INTR;
763 * If we're using a userspace irqchip, then check if we need
764 * to tell a userspace irqchip about timer or PMU level
765 * changes and if so, exit to userspace (the actual level
766 * state gets updated in kvm_timer_update_run and
767 * kvm_pmu_update_run below).
769 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
770 if (kvm_timer_should_notify_user(vcpu) ||
771 kvm_pmu_should_notify_user(vcpu)) {
773 run->exit_reason = KVM_EXIT_INTR;
778 * Ensure we set mode to IN_GUEST_MODE after we disable
779 * interrupts and before the final VCPU requests check.
780 * See the comment in kvm_vcpu_exiting_guest_mode() and
781 * Documentation/virt/kvm/vcpu-requests.rst
783 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
785 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
786 kvm_request_pending(vcpu)) {
787 vcpu->mode = OUTSIDE_GUEST_MODE;
788 isb(); /* Ensure work in x_flush_hwstate is committed */
789 kvm_pmu_sync_hwstate(vcpu);
790 if (static_branch_unlikely(&userspace_irqchip_in_use))
791 kvm_timer_sync_user(vcpu);
792 kvm_vgic_sync_hwstate(vcpu);
798 kvm_arm_setup_debug(vcpu);
800 /**************************************************************
803 trace_kvm_entry(*vcpu_pc(vcpu));
804 guest_enter_irqoff();
806 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
808 vcpu->mode = OUTSIDE_GUEST_MODE;
812 *************************************************************/
814 kvm_arm_clear_debug(vcpu);
817 * We must sync the PMU state before the vgic state so
818 * that the vgic can properly sample the updated state of the
821 kvm_pmu_sync_hwstate(vcpu);
824 * Sync the vgic state before syncing the timer state because
825 * the timer code needs to know if the virtual timer
826 * interrupts are active.
828 kvm_vgic_sync_hwstate(vcpu);
831 * Sync the timer hardware state before enabling interrupts as
832 * we don't want vtimer interrupts to race with syncing the
833 * timer virtual interrupt state.
835 if (static_branch_unlikely(&userspace_irqchip_in_use))
836 kvm_timer_sync_user(vcpu);
838 kvm_arch_vcpu_ctxsync_fp(vcpu);
841 * We may have taken a host interrupt in HYP mode (ie
842 * while executing the guest). This interrupt is still
843 * pending, as we haven't serviced it yet!
845 * We're now back in SVC mode, with interrupts
846 * disabled. Enabling the interrupts now will have
847 * the effect of taking the interrupt again, in SVC
853 * We do local_irq_enable() before calling guest_exit() so
854 * that if a timer interrupt hits while running the guest we
855 * account that tick as being spent in the guest. We enable
856 * preemption after calling guest_exit() so that if we get
857 * preempted we make sure ticks after that is not counted as
861 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
863 /* Exit types that need handling before we can be preempted */
864 handle_exit_early(vcpu, ret);
869 * The ARMv8 architecture doesn't give the hypervisor
870 * a mechanism to prevent a guest from dropping to AArch32 EL0
871 * if implemented by the CPU. If we spot the guest in such
872 * state and that we decided it wasn't supposed to do so (like
873 * with the asymmetric AArch32 case), return to userspace with
876 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
878 * As we have caught the guest red-handed, decide that
879 * it isn't fit for purpose anymore by making the vcpu
880 * invalid. The VMM can try and fix it by issuing a
881 * KVM_ARM_VCPU_INIT if it really wants to.
883 vcpu->arch.target = -1;
884 ret = ARM_EXCEPTION_IL;
887 ret = handle_exit(vcpu, ret);
890 /* Tell userspace about in-kernel device output levels */
891 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
892 kvm_timer_update_run(vcpu);
893 kvm_pmu_update_run(vcpu);
896 kvm_sigset_deactivate(vcpu);
902 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
908 if (number == KVM_ARM_IRQ_CPU_IRQ)
909 bit_index = __ffs(HCR_VI);
910 else /* KVM_ARM_IRQ_CPU_FIQ */
911 bit_index = __ffs(HCR_VF);
913 hcr = vcpu_hcr(vcpu);
915 set = test_and_set_bit(bit_index, hcr);
917 set = test_and_clear_bit(bit_index, hcr);
920 * If we didn't change anything, no need to wake up or kick other CPUs
926 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
927 * trigger a world-switch round on the running physical CPU to set the
928 * virtual IRQ/FIQ fields in the HCR appropriately.
930 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
936 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
939 u32 irq = irq_level->irq;
940 unsigned int irq_type, vcpu_idx, irq_num;
941 int nrcpus = atomic_read(&kvm->online_vcpus);
942 struct kvm_vcpu *vcpu = NULL;
943 bool level = irq_level->level;
945 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
946 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
947 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
948 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
950 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
953 case KVM_ARM_IRQ_TYPE_CPU:
954 if (irqchip_in_kernel(kvm))
957 if (vcpu_idx >= nrcpus)
960 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
964 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
967 return vcpu_interrupt_line(vcpu, irq_num, level);
968 case KVM_ARM_IRQ_TYPE_PPI:
969 if (!irqchip_in_kernel(kvm))
972 if (vcpu_idx >= nrcpus)
975 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
979 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
982 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
983 case KVM_ARM_IRQ_TYPE_SPI:
984 if (!irqchip_in_kernel(kvm))
987 if (irq_num < VGIC_NR_PRIVATE_IRQS)
990 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
996 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
997 const struct kvm_vcpu_init *init)
1000 int phys_target = kvm_target_cpu();
1002 if (init->target != phys_target)
1006 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1007 * use the same target.
1009 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1012 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1013 for (i = 0; i < sizeof(init->features) * 8; i++) {
1014 bool set = (init->features[i / 32] & (1 << (i % 32)));
1016 if (set && i >= KVM_VCPU_MAX_FEATURES)
1020 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1021 * use the same feature set.
1023 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1024 test_bit(i, vcpu->arch.features) != set)
1028 set_bit(i, vcpu->arch.features);
1031 vcpu->arch.target = phys_target;
1033 /* Now we know what it is, we can reset it. */
1034 ret = kvm_reset_vcpu(vcpu);
1036 vcpu->arch.target = -1;
1037 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1043 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1044 struct kvm_vcpu_init *init)
1048 ret = kvm_vcpu_set_target(vcpu, init);
1053 * Ensure a rebooted VM will fault in RAM pages and detect if the
1054 * guest MMU is turned off and flush the caches as needed.
1056 * S2FWB enforces all memory accesses to RAM being cacheable,
1057 * ensuring that the data side is always coherent. We still
1058 * need to invalidate the I-cache though, as FWB does *not*
1059 * imply CTR_EL0.DIC.
1061 if (vcpu->arch.has_run_once) {
1062 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1063 stage2_unmap_vm(vcpu->kvm);
1065 __flush_icache_all();
1068 vcpu_reset_hcr(vcpu);
1071 * Handle the "start in power-off" case.
1073 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1074 vcpu_power_off(vcpu);
1076 vcpu->arch.power_off = false;
1081 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1082 struct kvm_device_attr *attr)
1086 switch (attr->group) {
1088 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1095 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1096 struct kvm_device_attr *attr)
1100 switch (attr->group) {
1102 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1109 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1110 struct kvm_device_attr *attr)
1114 switch (attr->group) {
1116 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1123 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1124 struct kvm_vcpu_events *events)
1126 memset(events, 0, sizeof(*events));
1128 return __kvm_arm_vcpu_get_events(vcpu, events);
1131 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1132 struct kvm_vcpu_events *events)
1136 /* check whether the reserved field is zero */
1137 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1138 if (events->reserved[i])
1141 /* check whether the pad field is zero */
1142 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1143 if (events->exception.pad[i])
1146 return __kvm_arm_vcpu_set_events(vcpu, events);
1149 long kvm_arch_vcpu_ioctl(struct file *filp,
1150 unsigned int ioctl, unsigned long arg)
1152 struct kvm_vcpu *vcpu = filp->private_data;
1153 void __user *argp = (void __user *)arg;
1154 struct kvm_device_attr attr;
1158 case KVM_ARM_VCPU_INIT: {
1159 struct kvm_vcpu_init init;
1162 if (copy_from_user(&init, argp, sizeof(init)))
1165 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1168 case KVM_SET_ONE_REG:
1169 case KVM_GET_ONE_REG: {
1170 struct kvm_one_reg reg;
1173 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1177 if (copy_from_user(®, argp, sizeof(reg)))
1180 if (ioctl == KVM_SET_ONE_REG)
1181 r = kvm_arm_set_reg(vcpu, ®);
1183 r = kvm_arm_get_reg(vcpu, ®);
1186 case KVM_GET_REG_LIST: {
1187 struct kvm_reg_list __user *user_list = argp;
1188 struct kvm_reg_list reg_list;
1192 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1196 if (!kvm_arm_vcpu_is_finalized(vcpu))
1200 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1203 reg_list.n = kvm_arm_num_regs(vcpu);
1204 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1209 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1212 case KVM_SET_DEVICE_ATTR: {
1214 if (copy_from_user(&attr, argp, sizeof(attr)))
1216 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1219 case KVM_GET_DEVICE_ATTR: {
1221 if (copy_from_user(&attr, argp, sizeof(attr)))
1223 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1226 case KVM_HAS_DEVICE_ATTR: {
1228 if (copy_from_user(&attr, argp, sizeof(attr)))
1230 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1233 case KVM_GET_VCPU_EVENTS: {
1234 struct kvm_vcpu_events events;
1236 if (kvm_arm_vcpu_get_events(vcpu, &events))
1239 if (copy_to_user(argp, &events, sizeof(events)))
1244 case KVM_SET_VCPU_EVENTS: {
1245 struct kvm_vcpu_events events;
1247 if (copy_from_user(&events, argp, sizeof(events)))
1250 return kvm_arm_vcpu_set_events(vcpu, &events);
1252 case KVM_ARM_VCPU_FINALIZE: {
1255 if (!kvm_vcpu_initialized(vcpu))
1258 if (get_user(what, (const int __user *)argp))
1261 return kvm_arm_vcpu_finalize(vcpu, what);
1270 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1275 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1276 struct kvm_memory_slot *memslot)
1278 kvm_flush_remote_tlbs(kvm);
1281 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1282 struct kvm_arm_device_addr *dev_addr)
1284 unsigned long dev_id, type;
1286 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1287 KVM_ARM_DEVICE_ID_SHIFT;
1288 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1289 KVM_ARM_DEVICE_TYPE_SHIFT;
1292 case KVM_ARM_DEVICE_VGIC_V2:
1295 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1301 long kvm_arch_vm_ioctl(struct file *filp,
1302 unsigned int ioctl, unsigned long arg)
1304 struct kvm *kvm = filp->private_data;
1305 void __user *argp = (void __user *)arg;
1308 case KVM_CREATE_IRQCHIP: {
1312 mutex_lock(&kvm->lock);
1313 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1314 mutex_unlock(&kvm->lock);
1317 case KVM_ARM_SET_DEVICE_ADDR: {
1318 struct kvm_arm_device_addr dev_addr;
1320 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1322 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1324 case KVM_ARM_PREFERRED_TARGET: {
1326 struct kvm_vcpu_init init;
1328 err = kvm_vcpu_preferred_target(&init);
1332 if (copy_to_user(argp, &init, sizeof(init)))
1342 static unsigned long nvhe_percpu_size(void)
1344 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1345 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1348 static unsigned long nvhe_percpu_order(void)
1350 unsigned long size = nvhe_percpu_size();
1352 return size ? get_order(size) : 0;
1355 /* A lookup table holding the hypervisor VA for each vector slot */
1356 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1358 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1360 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1363 static int kvm_init_vector_slots(void)
1368 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1369 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1371 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1372 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1374 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1378 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1379 __BP_HARDEN_HYP_VECS_SZ, &base);
1384 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1385 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1389 static void cpu_prepare_hyp_mode(int cpu)
1391 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1395 * Calculate the raw per-cpu offset without a translation from the
1396 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1397 * so that we can use adr_l to access per-cpu variables in EL2.
1398 * Also drop the KASAN tag which gets in the way...
1400 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1401 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1403 params->mair_el2 = read_sysreg(mair_el1);
1406 * The ID map may be configured to use an extended virtual address
1407 * range. This is only the case if system RAM is out of range for the
1408 * currently configured page size and VA_BITS, in which case we will
1409 * also need the extended virtual range for the HYP ID map, or we won't
1410 * be able to enable the EL2 MMU.
1412 * However, at EL2, there is only one TTBR register, and we can't switch
1413 * between translation tables *and* update TCR_EL2.T0SZ at the same
1414 * time. Bottom line: we need to use the extended range with *both* our
1415 * translation tables.
1417 * So use the same T0SZ value we use for the ID map.
1419 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1420 tcr &= ~TCR_T0SZ_MASK;
1421 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1422 params->tcr_el2 = tcr;
1424 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1425 params->pgd_pa = kvm_mmu_get_httbr();
1426 if (is_protected_kvm_enabled())
1427 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1429 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1430 params->vttbr = params->vtcr = 0;
1433 * Flush the init params from the data cache because the struct will
1434 * be read while the MMU is off.
1436 kvm_flush_dcache_to_poc(params, sizeof(*params));
1439 static void hyp_install_host_vector(void)
1441 struct kvm_nvhe_init_params *params;
1442 struct arm_smccc_res res;
1444 /* Switch from the HYP stub to our own HYP init vector */
1445 __hyp_set_vectors(kvm_get_idmap_vector());
1448 * Call initialization code, and switch to the full blown HYP code.
1449 * If the cpucaps haven't been finalized yet, something has gone very
1450 * wrong, and hyp will crash and burn when it uses any
1451 * cpus_have_const_cap() wrapper.
1453 BUG_ON(!system_capabilities_finalized());
1454 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1455 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1456 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1459 static void cpu_init_hyp_mode(void)
1461 hyp_install_host_vector();
1464 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1467 if (this_cpu_has_cap(ARM64_SSBS) &&
1468 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1469 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1473 static void cpu_hyp_reset(void)
1475 if (!is_kernel_in_hyp_mode())
1476 __hyp_reset_vectors();
1480 * EL2 vectors can be mapped and rerouted in a number of ways,
1481 * depending on the kernel configuration and CPU present:
1483 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1484 * placed in one of the vector slots, which is executed before jumping
1485 * to the real vectors.
1487 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1488 * containing the hardening sequence is mapped next to the idmap page,
1489 * and executed before jumping to the real vectors.
1491 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1492 * empty slot is selected, mapped next to the idmap page, and
1493 * executed before jumping to the real vectors.
1495 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1496 * VHE, as we don't have hypervisor-specific mappings. If the system
1497 * is VHE and yet selects this capability, it will be ignored.
1499 static void cpu_set_hyp_vector(void)
1501 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1502 void *vector = hyp_spectre_vector_selector[data->slot];
1504 if (!is_protected_kvm_enabled())
1505 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1507 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1510 static void cpu_hyp_reinit(void)
1512 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1516 if (is_kernel_in_hyp_mode())
1517 kvm_timer_init_vhe();
1519 cpu_init_hyp_mode();
1521 cpu_set_hyp_vector();
1523 kvm_arm_init_debug();
1526 kvm_vgic_init_cpu_hardware();
1529 static void _kvm_arch_hardware_enable(void *discard)
1531 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1533 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1537 int kvm_arch_hardware_enable(void)
1539 _kvm_arch_hardware_enable(NULL);
1543 static void _kvm_arch_hardware_disable(void *discard)
1545 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1547 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1551 void kvm_arch_hardware_disable(void)
1553 if (!is_protected_kvm_enabled())
1554 _kvm_arch_hardware_disable(NULL);
1557 #ifdef CONFIG_CPU_PM
1558 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1563 * kvm_arm_hardware_enabled is left with its old value over
1564 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1569 if (__this_cpu_read(kvm_arm_hardware_enabled))
1571 * don't update kvm_arm_hardware_enabled here
1572 * so that the hardware will be re-enabled
1573 * when we resume. See below.
1578 case CPU_PM_ENTER_FAILED:
1580 if (__this_cpu_read(kvm_arm_hardware_enabled))
1581 /* The hardware was enabled before suspend. */
1591 static struct notifier_block hyp_init_cpu_pm_nb = {
1592 .notifier_call = hyp_init_cpu_pm_notifier,
1595 static void hyp_cpu_pm_init(void)
1597 if (!is_protected_kvm_enabled())
1598 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1600 static void hyp_cpu_pm_exit(void)
1602 if (!is_protected_kvm_enabled())
1603 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1606 static inline void hyp_cpu_pm_init(void)
1609 static inline void hyp_cpu_pm_exit(void)
1614 static void init_cpu_logical_map(void)
1619 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1620 * Only copy the set of online CPUs whose features have been chacked
1621 * against the finalized system capabilities. The hypervisor will not
1622 * allow any other CPUs from the `possible` set to boot.
1624 for_each_online_cpu(cpu)
1625 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1628 #define init_psci_0_1_impl_state(config, what) \
1629 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1631 static bool init_psci_relay(void)
1634 * If PSCI has not been initialized, protected KVM cannot install
1635 * itself on newly booted CPUs.
1637 if (!psci_ops.get_version) {
1638 kvm_err("Cannot initialize protected mode without PSCI\n");
1642 kvm_host_psci_config.version = psci_ops.get_version();
1644 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1645 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1646 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1647 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1648 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1649 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1654 static int init_common_resources(void)
1656 return kvm_set_ipa_limit();
1659 static int init_subsystems(void)
1664 * Enable hardware so that subsystem initialisation can access EL2.
1666 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1669 * Register CPU lower-power notifier
1674 * Init HYP view of VGIC
1676 err = kvm_vgic_hyp_init();
1679 vgic_present = true;
1683 vgic_present = false;
1691 * Init HYP architected timer support
1693 err = kvm_timer_hyp_init(vgic_present);
1698 kvm_sys_reg_table_init();
1701 if (err || !is_protected_kvm_enabled())
1702 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1707 static void teardown_hyp_mode(void)
1712 for_each_possible_cpu(cpu) {
1713 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1714 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1718 static int do_pkvm_init(u32 hyp_va_bits)
1720 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1724 hyp_install_host_vector();
1725 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1726 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1733 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1735 void *addr = phys_to_virt(hyp_mem_base);
1738 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1739 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1741 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1745 ret = do_pkvm_init(hyp_va_bits);
1755 * Inits Hyp-mode on all online CPUs
1757 static int init_hyp_mode(void)
1764 * The protected Hyp-mode cannot be initialized if the memory pool
1765 * allocation has failed.
1767 if (is_protected_kvm_enabled() && !hyp_mem_base)
1771 * Allocate Hyp PGD and setup Hyp identity mapping
1773 err = kvm_mmu_init(&hyp_va_bits);
1778 * Allocate stack pages for Hypervisor-mode
1780 for_each_possible_cpu(cpu) {
1781 unsigned long stack_page;
1783 stack_page = __get_free_page(GFP_KERNEL);
1789 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1793 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1795 for_each_possible_cpu(cpu) {
1799 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1805 page_addr = page_address(page);
1806 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1807 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1811 * Map the Hyp-code called directly from the host
1813 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1814 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1816 kvm_err("Cannot map world-switch code\n");
1820 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1821 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1823 kvm_err("Cannot map .hyp.rodata section\n");
1827 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1828 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1830 kvm_err("Cannot map rodata section\n");
1835 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1836 * section thanks to an assertion in the linker script. Map it RW and
1837 * the rest of .bss RO.
1839 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1840 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1842 kvm_err("Cannot map hyp bss section: %d\n", err);
1846 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1847 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1849 kvm_err("Cannot map bss section\n");
1854 * Map the Hyp stack pages
1856 for_each_possible_cpu(cpu) {
1857 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1858 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1862 kvm_err("Cannot map hyp stack\n");
1867 for_each_possible_cpu(cpu) {
1868 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1869 char *percpu_end = percpu_begin + nvhe_percpu_size();
1871 /* Map Hyp percpu pages */
1872 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1874 kvm_err("Cannot map hyp percpu region\n");
1878 /* Prepare the CPU initialization parameters */
1879 cpu_prepare_hyp_mode(cpu);
1882 if (is_protected_kvm_enabled()) {
1883 init_cpu_logical_map();
1885 if (!init_psci_relay()) {
1891 if (is_protected_kvm_enabled()) {
1892 err = kvm_hyp_init_protection(hyp_va_bits);
1894 kvm_err("Failed to init hyp memory protection\n");
1902 teardown_hyp_mode();
1903 kvm_err("error initializing Hyp mode: %d\n", err);
1907 static void _kvm_host_prot_finalize(void *discard)
1909 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1912 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1914 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1917 #define pkvm_mark_hyp_section(__section) \
1918 pkvm_mark_hyp(__pa_symbol(__section##_start), \
1919 __pa_symbol(__section##_end))
1921 static int finalize_hyp_mode(void)
1925 if (!is_protected_kvm_enabled())
1928 ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1932 ret = pkvm_mark_hyp_section(__hyp_text);
1936 ret = pkvm_mark_hyp_section(__hyp_rodata);
1940 ret = pkvm_mark_hyp_section(__hyp_bss);
1944 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1948 for_each_possible_cpu(cpu) {
1949 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1950 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
1952 ret = pkvm_mark_hyp(start, end);
1956 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
1957 end = start + PAGE_SIZE;
1958 ret = pkvm_mark_hyp(start, end);
1964 * Flip the static key upfront as that may no longer be possible
1965 * once the host stage 2 is installed.
1967 static_branch_enable(&kvm_protected_mode_initialized);
1968 on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
1973 static void check_kvm_target_cpu(void *ret)
1975 *(int *)ret = kvm_target_cpu();
1978 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1980 struct kvm_vcpu *vcpu;
1983 mpidr &= MPIDR_HWID_BITMASK;
1984 kvm_for_each_vcpu(i, vcpu, kvm) {
1985 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1991 bool kvm_arch_has_irq_bypass(void)
1996 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1997 struct irq_bypass_producer *prod)
1999 struct kvm_kernel_irqfd *irqfd =
2000 container_of(cons, struct kvm_kernel_irqfd, consumer);
2002 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2005 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2006 struct irq_bypass_producer *prod)
2008 struct kvm_kernel_irqfd *irqfd =
2009 container_of(cons, struct kvm_kernel_irqfd, consumer);
2011 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2015 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2017 struct kvm_kernel_irqfd *irqfd =
2018 container_of(cons, struct kvm_kernel_irqfd, consumer);
2020 kvm_arm_halt_guest(irqfd->kvm);
2023 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2025 struct kvm_kernel_irqfd *irqfd =
2026 container_of(cons, struct kvm_kernel_irqfd, consumer);
2028 kvm_arm_resume_guest(irqfd->kvm);
2032 * Initialize Hyp-mode and memory mappings on all CPUs.
2034 int kvm_arch_init(void *opaque)
2040 if (!is_hyp_mode_available()) {
2041 kvm_info("HYP mode not available\n");
2045 in_hyp_mode = is_kernel_in_hyp_mode();
2047 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2048 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2049 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2050 "Only trusted guests should be used on this system.\n");
2052 for_each_online_cpu(cpu) {
2053 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
2055 kvm_err("Error, CPU %d not supported!\n", cpu);
2060 err = init_common_resources();
2064 err = kvm_arm_init_sve();
2069 err = init_hyp_mode();
2074 err = kvm_init_vector_slots();
2076 kvm_err("Cannot initialise vector slots\n");
2080 err = init_subsystems();
2085 err = finalize_hyp_mode();
2087 kvm_err("Failed to finalize Hyp protection\n");
2092 if (is_protected_kvm_enabled()) {
2093 kvm_info("Protected nVHE mode initialized successfully\n");
2094 } else if (in_hyp_mode) {
2095 kvm_info("VHE mode initialized successfully\n");
2097 kvm_info("Hyp mode initialized successfully\n");
2105 teardown_hyp_mode();
2110 /* NOP: Compiling as a module not supported */
2111 void kvm_arch_exit(void)
2113 kvm_perf_teardown();
2116 static int __init early_kvm_mode_cfg(char *arg)
2121 if (strcmp(arg, "protected") == 0) {
2122 kvm_mode = KVM_MODE_PROTECTED;
2126 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2131 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2133 enum kvm_mode kvm_get_mode(void)
2138 static int arm_init(void)
2140 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2144 module_init(arm_init);