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_SET_GUEST_DEBUG2:
213 return KVM_GUESTDBG_VALID_MASK;
214 case KVM_CAP_ARM_SET_DEVICE_ADDR:
217 case KVM_CAP_NR_VCPUS:
218 r = num_online_cpus();
220 case KVM_CAP_MAX_VCPUS:
221 case KVM_CAP_MAX_VCPU_ID:
223 r = kvm->arch.max_vcpus;
225 r = kvm_arm_default_max_vcpus();
227 case KVM_CAP_MSI_DEVID:
231 r = kvm->arch.vgic.msis_require_devid;
233 case KVM_CAP_ARM_USER_IRQ:
235 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
236 * (bump this number if adding more devices)
240 case KVM_CAP_STEAL_TIME:
241 r = kvm_arm_pvtime_supported();
243 case KVM_CAP_ARM_EL1_32BIT:
244 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
246 case KVM_CAP_GUEST_DEBUG_HW_BPS:
249 case KVM_CAP_GUEST_DEBUG_HW_WPS:
252 case KVM_CAP_ARM_PMU_V3:
253 r = kvm_arm_support_pmu_v3();
255 case KVM_CAP_ARM_INJECT_SERROR_ESR:
256 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
258 case KVM_CAP_ARM_VM_IPA_SIZE:
259 r = get_kvm_ipa_limit();
261 case KVM_CAP_ARM_SVE:
262 r = system_supports_sve();
264 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
265 case KVM_CAP_ARM_PTRAUTH_GENERIC:
266 r = system_has_full_ptr_auth();
275 long kvm_arch_dev_ioctl(struct file *filp,
276 unsigned int ioctl, unsigned long arg)
281 struct kvm *kvm_arch_alloc_vm(void)
284 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
286 return vzalloc(sizeof(struct kvm));
289 void kvm_arch_free_vm(struct kvm *kvm)
297 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
299 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
302 if (id >= kvm->arch.max_vcpus)
308 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
312 /* Force users to call KVM_ARM_VCPU_INIT */
313 vcpu->arch.target = -1;
314 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
316 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
318 /* Set up the timer */
319 kvm_timer_vcpu_init(vcpu);
321 kvm_pmu_vcpu_init(vcpu);
323 kvm_arm_reset_debug_ptr(vcpu);
325 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
327 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
329 err = kvm_vgic_vcpu_init(vcpu);
333 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
336 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
340 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
342 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
343 static_branch_dec(&userspace_irqchip_in_use);
345 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
346 kvm_timer_vcpu_terminate(vcpu);
347 kvm_pmu_vcpu_destroy(vcpu);
349 kvm_arm_vcpu_destroy(vcpu);
352 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
354 return kvm_timer_is_pending(vcpu);
357 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
360 * If we're about to block (most likely because we've just hit a
361 * WFI), we need to sync back the state of the GIC CPU interface
362 * so that we have the latest PMR and group enables. This ensures
363 * that kvm_arch_vcpu_runnable has up-to-date data to decide
364 * whether we have pending interrupts.
366 * For the same reason, we want to tell GICv4 that we need
367 * doorbells to be signalled, should an interrupt become pending.
370 kvm_vgic_vmcr_sync(vcpu);
371 vgic_v4_put(vcpu, true);
375 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
382 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
384 struct kvm_s2_mmu *mmu;
387 mmu = vcpu->arch.hw_mmu;
388 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
391 * We guarantee that both TLBs and I-cache are private to each
392 * vcpu. If detecting that a vcpu from the same VM has
393 * previously run on the same physical CPU, call into the
394 * hypervisor code to nuke the relevant contexts.
396 * We might get preempted before the vCPU actually runs, but
397 * over-invalidation doesn't affect correctness.
399 if (*last_ran != vcpu->vcpu_id) {
400 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
401 *last_ran = vcpu->vcpu_id;
407 kvm_timer_vcpu_load(vcpu);
409 kvm_vcpu_load_sysregs_vhe(vcpu);
410 kvm_arch_vcpu_load_fp(vcpu);
411 kvm_vcpu_pmu_restore_guest(vcpu);
412 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
413 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
415 if (single_task_running())
416 vcpu_clear_wfx_traps(vcpu);
418 vcpu_set_wfx_traps(vcpu);
420 if (vcpu_has_ptrauth(vcpu))
421 vcpu_ptrauth_disable(vcpu);
422 kvm_arch_vcpu_load_debug_state_flags(vcpu);
425 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
427 kvm_arch_vcpu_put_debug_state_flags(vcpu);
428 kvm_arch_vcpu_put_fp(vcpu);
430 kvm_vcpu_put_sysregs_vhe(vcpu);
431 kvm_timer_vcpu_put(vcpu);
433 kvm_vcpu_pmu_restore_host(vcpu);
438 static void vcpu_power_off(struct kvm_vcpu *vcpu)
440 vcpu->arch.power_off = true;
441 kvm_make_request(KVM_REQ_SLEEP, vcpu);
445 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
446 struct kvm_mp_state *mp_state)
448 if (vcpu->arch.power_off)
449 mp_state->mp_state = KVM_MP_STATE_STOPPED;
451 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
456 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
457 struct kvm_mp_state *mp_state)
461 switch (mp_state->mp_state) {
462 case KVM_MP_STATE_RUNNABLE:
463 vcpu->arch.power_off = false;
465 case KVM_MP_STATE_STOPPED:
466 vcpu_power_off(vcpu);
476 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
477 * @v: The VCPU pointer
479 * If the guest CPU is not waiting for interrupts or an interrupt line is
480 * asserted, the CPU is by definition runnable.
482 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
484 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
485 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
486 && !v->arch.power_off && !v->arch.pause);
489 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
491 return vcpu_mode_priv(vcpu);
494 /* Just ensure a guest exit from a particular CPU */
495 static void exit_vm_noop(void *info)
499 void force_vm_exit(const cpumask_t *mask)
502 smp_call_function_many(mask, exit_vm_noop, NULL, true);
507 * need_new_vmid_gen - check that the VMID is still valid
508 * @vmid: The VMID to check
510 * return true if there is a new generation of VMIDs being used
512 * The hardware supports a limited set of values with the value zero reserved
513 * for the host, so we check if an assigned value belongs to a previous
514 * generation, which requires us to assign a new value. If we're the first to
515 * use a VMID for the new generation, we must flush necessary caches and TLBs
518 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
520 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
521 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
522 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
526 * update_vmid - Update the vmid with a valid VMID for the current generation
527 * @vmid: The stage-2 VMID information struct
529 static void update_vmid(struct kvm_vmid *vmid)
531 if (!need_new_vmid_gen(vmid))
534 spin_lock(&kvm_vmid_lock);
537 * We need to re-check the vmid_gen here to ensure that if another vcpu
538 * already allocated a valid vmid for this vm, then this vcpu should
541 if (!need_new_vmid_gen(vmid)) {
542 spin_unlock(&kvm_vmid_lock);
546 /* First user of a new VMID generation? */
547 if (unlikely(kvm_next_vmid == 0)) {
548 atomic64_inc(&kvm_vmid_gen);
552 * On SMP we know no other CPUs can use this CPU's or each
553 * other's VMID after force_vm_exit returns since the
554 * kvm_vmid_lock blocks them from reentry to the guest.
556 force_vm_exit(cpu_all_mask);
558 * Now broadcast TLB + ICACHE invalidation over the inner
559 * shareable domain to make sure all data structures are
562 kvm_call_hyp(__kvm_flush_vm_context);
565 vmid->vmid = kvm_next_vmid;
567 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
570 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
572 spin_unlock(&kvm_vmid_lock);
575 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
577 struct kvm *kvm = vcpu->kvm;
580 if (likely(vcpu->arch.has_run_once))
583 if (!kvm_arm_vcpu_is_finalized(vcpu))
586 vcpu->arch.has_run_once = true;
588 kvm_arm_vcpu_init_debug(vcpu);
590 if (likely(irqchip_in_kernel(kvm))) {
592 * Map the VGIC hardware resources before running a vcpu the
593 * first time on this VM.
595 ret = kvm_vgic_map_resources(kvm);
600 * Tell the rest of the code that there are userspace irqchip
603 static_branch_inc(&userspace_irqchip_in_use);
606 ret = kvm_timer_enable(vcpu);
610 ret = kvm_arm_pmu_v3_enable(vcpu);
615 bool kvm_arch_intc_initialized(struct kvm *kvm)
617 return vgic_initialized(kvm);
620 void kvm_arm_halt_guest(struct kvm *kvm)
623 struct kvm_vcpu *vcpu;
625 kvm_for_each_vcpu(i, vcpu, kvm)
626 vcpu->arch.pause = true;
627 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
630 void kvm_arm_resume_guest(struct kvm *kvm)
633 struct kvm_vcpu *vcpu;
635 kvm_for_each_vcpu(i, vcpu, kvm) {
636 vcpu->arch.pause = false;
637 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
641 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
643 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
645 rcuwait_wait_event(wait,
646 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
649 if (vcpu->arch.power_off || vcpu->arch.pause) {
650 /* Awaken to handle a signal, request we sleep again later. */
651 kvm_make_request(KVM_REQ_SLEEP, vcpu);
655 * Make sure we will observe a potential reset request if we've
656 * observed a change to the power state. Pairs with the smp_wmb() in
657 * kvm_psci_vcpu_on().
662 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
664 return vcpu->arch.target >= 0;
667 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
669 if (kvm_request_pending(vcpu)) {
670 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
671 vcpu_req_sleep(vcpu);
673 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
674 kvm_reset_vcpu(vcpu);
677 * Clear IRQ_PENDING requests that were made to guarantee
678 * that a VCPU sees new virtual interrupts.
680 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
682 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
683 kvm_update_stolen_time(vcpu);
685 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
686 /* The distributor enable bits were changed */
688 vgic_v4_put(vcpu, false);
696 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
697 * @vcpu: The VCPU pointer
699 * This function is called through the VCPU_RUN ioctl called from user space. It
700 * will execute VM code in a loop until the time slice for the process is used
701 * or some emulation is needed from user space in which case the function will
702 * return with return value 0 and with the kvm_run structure filled in with the
703 * required data for the requested emulation.
705 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
707 struct kvm_run *run = vcpu->run;
710 if (unlikely(!kvm_vcpu_initialized(vcpu)))
713 ret = kvm_vcpu_first_run_init(vcpu);
717 if (run->exit_reason == KVM_EXIT_MMIO) {
718 ret = kvm_handle_mmio_return(vcpu);
723 if (run->immediate_exit)
728 kvm_sigset_activate(vcpu);
731 run->exit_reason = KVM_EXIT_UNKNOWN;
734 * Check conditions before entering the guest
738 update_vmid(&vcpu->arch.hw_mmu->vmid);
740 check_vcpu_requests(vcpu);
743 * Preparing the interrupts to be injected also
744 * involves poking the GIC, which must be done in a
745 * non-preemptible context.
749 kvm_pmu_flush_hwstate(vcpu);
753 kvm_vgic_flush_hwstate(vcpu);
756 * Exit if we have a signal pending so that we can deliver the
757 * signal to user space.
759 if (signal_pending(current)) {
761 run->exit_reason = KVM_EXIT_INTR;
765 * If we're using a userspace irqchip, then check if we need
766 * to tell a userspace irqchip about timer or PMU level
767 * changes and if so, exit to userspace (the actual level
768 * state gets updated in kvm_timer_update_run and
769 * kvm_pmu_update_run below).
771 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
772 if (kvm_timer_should_notify_user(vcpu) ||
773 kvm_pmu_should_notify_user(vcpu)) {
775 run->exit_reason = KVM_EXIT_INTR;
780 * Ensure we set mode to IN_GUEST_MODE after we disable
781 * interrupts and before the final VCPU requests check.
782 * See the comment in kvm_vcpu_exiting_guest_mode() and
783 * Documentation/virt/kvm/vcpu-requests.rst
785 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
787 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
788 kvm_request_pending(vcpu)) {
789 vcpu->mode = OUTSIDE_GUEST_MODE;
790 isb(); /* Ensure work in x_flush_hwstate is committed */
791 kvm_pmu_sync_hwstate(vcpu);
792 if (static_branch_unlikely(&userspace_irqchip_in_use))
793 kvm_timer_sync_user(vcpu);
794 kvm_vgic_sync_hwstate(vcpu);
800 kvm_arm_setup_debug(vcpu);
802 /**************************************************************
805 trace_kvm_entry(*vcpu_pc(vcpu));
806 guest_enter_irqoff();
808 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
810 vcpu->mode = OUTSIDE_GUEST_MODE;
814 *************************************************************/
816 kvm_arm_clear_debug(vcpu);
819 * We must sync the PMU state before the vgic state so
820 * that the vgic can properly sample the updated state of the
823 kvm_pmu_sync_hwstate(vcpu);
826 * Sync the vgic state before syncing the timer state because
827 * the timer code needs to know if the virtual timer
828 * interrupts are active.
830 kvm_vgic_sync_hwstate(vcpu);
833 * Sync the timer hardware state before enabling interrupts as
834 * we don't want vtimer interrupts to race with syncing the
835 * timer virtual interrupt state.
837 if (static_branch_unlikely(&userspace_irqchip_in_use))
838 kvm_timer_sync_user(vcpu);
840 kvm_arch_vcpu_ctxsync_fp(vcpu);
843 * We may have taken a host interrupt in HYP mode (ie
844 * while executing the guest). This interrupt is still
845 * pending, as we haven't serviced it yet!
847 * We're now back in SVC mode, with interrupts
848 * disabled. Enabling the interrupts now will have
849 * the effect of taking the interrupt again, in SVC
855 * We do local_irq_enable() before calling guest_exit() so
856 * that if a timer interrupt hits while running the guest we
857 * account that tick as being spent in the guest. We enable
858 * preemption after calling guest_exit() so that if we get
859 * preempted we make sure ticks after that is not counted as
863 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
865 /* Exit types that need handling before we can be preempted */
866 handle_exit_early(vcpu, ret);
871 * The ARMv8 architecture doesn't give the hypervisor
872 * a mechanism to prevent a guest from dropping to AArch32 EL0
873 * if implemented by the CPU. If we spot the guest in such
874 * state and that we decided it wasn't supposed to do so (like
875 * with the asymmetric AArch32 case), return to userspace with
878 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
880 * As we have caught the guest red-handed, decide that
881 * it isn't fit for purpose anymore by making the vcpu
882 * invalid. The VMM can try and fix it by issuing a
883 * KVM_ARM_VCPU_INIT if it really wants to.
885 vcpu->arch.target = -1;
886 ret = ARM_EXCEPTION_IL;
889 ret = handle_exit(vcpu, ret);
892 /* Tell userspace about in-kernel device output levels */
893 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
894 kvm_timer_update_run(vcpu);
895 kvm_pmu_update_run(vcpu);
898 kvm_sigset_deactivate(vcpu);
904 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
910 if (number == KVM_ARM_IRQ_CPU_IRQ)
911 bit_index = __ffs(HCR_VI);
912 else /* KVM_ARM_IRQ_CPU_FIQ */
913 bit_index = __ffs(HCR_VF);
915 hcr = vcpu_hcr(vcpu);
917 set = test_and_set_bit(bit_index, hcr);
919 set = test_and_clear_bit(bit_index, hcr);
922 * If we didn't change anything, no need to wake up or kick other CPUs
928 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
929 * trigger a world-switch round on the running physical CPU to set the
930 * virtual IRQ/FIQ fields in the HCR appropriately.
932 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
938 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
941 u32 irq = irq_level->irq;
942 unsigned int irq_type, vcpu_idx, irq_num;
943 int nrcpus = atomic_read(&kvm->online_vcpus);
944 struct kvm_vcpu *vcpu = NULL;
945 bool level = irq_level->level;
947 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
948 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
949 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
950 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
952 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
955 case KVM_ARM_IRQ_TYPE_CPU:
956 if (irqchip_in_kernel(kvm))
959 if (vcpu_idx >= nrcpus)
962 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
966 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
969 return vcpu_interrupt_line(vcpu, irq_num, level);
970 case KVM_ARM_IRQ_TYPE_PPI:
971 if (!irqchip_in_kernel(kvm))
974 if (vcpu_idx >= nrcpus)
977 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
981 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
984 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
985 case KVM_ARM_IRQ_TYPE_SPI:
986 if (!irqchip_in_kernel(kvm))
989 if (irq_num < VGIC_NR_PRIVATE_IRQS)
992 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
998 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
999 const struct kvm_vcpu_init *init)
1001 unsigned int i, ret;
1002 int phys_target = kvm_target_cpu();
1004 if (init->target != phys_target)
1008 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1009 * use the same target.
1011 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1014 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1015 for (i = 0; i < sizeof(init->features) * 8; i++) {
1016 bool set = (init->features[i / 32] & (1 << (i % 32)));
1018 if (set && i >= KVM_VCPU_MAX_FEATURES)
1022 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1023 * use the same feature set.
1025 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1026 test_bit(i, vcpu->arch.features) != set)
1030 set_bit(i, vcpu->arch.features);
1033 vcpu->arch.target = phys_target;
1035 /* Now we know what it is, we can reset it. */
1036 ret = kvm_reset_vcpu(vcpu);
1038 vcpu->arch.target = -1;
1039 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1045 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1046 struct kvm_vcpu_init *init)
1050 ret = kvm_vcpu_set_target(vcpu, init);
1055 * Ensure a rebooted VM will fault in RAM pages and detect if the
1056 * guest MMU is turned off and flush the caches as needed.
1058 * S2FWB enforces all memory accesses to RAM being cacheable,
1059 * ensuring that the data side is always coherent. We still
1060 * need to invalidate the I-cache though, as FWB does *not*
1061 * imply CTR_EL0.DIC.
1063 if (vcpu->arch.has_run_once) {
1064 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1065 stage2_unmap_vm(vcpu->kvm);
1067 __flush_icache_all();
1070 vcpu_reset_hcr(vcpu);
1073 * Handle the "start in power-off" case.
1075 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1076 vcpu_power_off(vcpu);
1078 vcpu->arch.power_off = false;
1083 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1084 struct kvm_device_attr *attr)
1088 switch (attr->group) {
1090 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1097 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1098 struct kvm_device_attr *attr)
1102 switch (attr->group) {
1104 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1111 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1112 struct kvm_device_attr *attr)
1116 switch (attr->group) {
1118 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1125 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1126 struct kvm_vcpu_events *events)
1128 memset(events, 0, sizeof(*events));
1130 return __kvm_arm_vcpu_get_events(vcpu, events);
1133 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1134 struct kvm_vcpu_events *events)
1138 /* check whether the reserved field is zero */
1139 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1140 if (events->reserved[i])
1143 /* check whether the pad field is zero */
1144 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1145 if (events->exception.pad[i])
1148 return __kvm_arm_vcpu_set_events(vcpu, events);
1151 long kvm_arch_vcpu_ioctl(struct file *filp,
1152 unsigned int ioctl, unsigned long arg)
1154 struct kvm_vcpu *vcpu = filp->private_data;
1155 void __user *argp = (void __user *)arg;
1156 struct kvm_device_attr attr;
1160 case KVM_ARM_VCPU_INIT: {
1161 struct kvm_vcpu_init init;
1164 if (copy_from_user(&init, argp, sizeof(init)))
1167 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1170 case KVM_SET_ONE_REG:
1171 case KVM_GET_ONE_REG: {
1172 struct kvm_one_reg reg;
1175 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1179 if (copy_from_user(®, argp, sizeof(reg)))
1182 if (ioctl == KVM_SET_ONE_REG)
1183 r = kvm_arm_set_reg(vcpu, ®);
1185 r = kvm_arm_get_reg(vcpu, ®);
1188 case KVM_GET_REG_LIST: {
1189 struct kvm_reg_list __user *user_list = argp;
1190 struct kvm_reg_list reg_list;
1194 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1198 if (!kvm_arm_vcpu_is_finalized(vcpu))
1202 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1205 reg_list.n = kvm_arm_num_regs(vcpu);
1206 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1211 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1214 case KVM_SET_DEVICE_ATTR: {
1216 if (copy_from_user(&attr, argp, sizeof(attr)))
1218 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1221 case KVM_GET_DEVICE_ATTR: {
1223 if (copy_from_user(&attr, argp, sizeof(attr)))
1225 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1228 case KVM_HAS_DEVICE_ATTR: {
1230 if (copy_from_user(&attr, argp, sizeof(attr)))
1232 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1235 case KVM_GET_VCPU_EVENTS: {
1236 struct kvm_vcpu_events events;
1238 if (kvm_arm_vcpu_get_events(vcpu, &events))
1241 if (copy_to_user(argp, &events, sizeof(events)))
1246 case KVM_SET_VCPU_EVENTS: {
1247 struct kvm_vcpu_events events;
1249 if (copy_from_user(&events, argp, sizeof(events)))
1252 return kvm_arm_vcpu_set_events(vcpu, &events);
1254 case KVM_ARM_VCPU_FINALIZE: {
1257 if (!kvm_vcpu_initialized(vcpu))
1260 if (get_user(what, (const int __user *)argp))
1263 return kvm_arm_vcpu_finalize(vcpu, what);
1272 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1277 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1278 const struct kvm_memory_slot *memslot)
1280 kvm_flush_remote_tlbs(kvm);
1283 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1284 struct kvm_arm_device_addr *dev_addr)
1286 unsigned long dev_id, type;
1288 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1289 KVM_ARM_DEVICE_ID_SHIFT;
1290 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1291 KVM_ARM_DEVICE_TYPE_SHIFT;
1294 case KVM_ARM_DEVICE_VGIC_V2:
1297 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1303 long kvm_arch_vm_ioctl(struct file *filp,
1304 unsigned int ioctl, unsigned long arg)
1306 struct kvm *kvm = filp->private_data;
1307 void __user *argp = (void __user *)arg;
1310 case KVM_CREATE_IRQCHIP: {
1314 mutex_lock(&kvm->lock);
1315 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1316 mutex_unlock(&kvm->lock);
1319 case KVM_ARM_SET_DEVICE_ADDR: {
1320 struct kvm_arm_device_addr dev_addr;
1322 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1324 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1326 case KVM_ARM_PREFERRED_TARGET: {
1328 struct kvm_vcpu_init init;
1330 err = kvm_vcpu_preferred_target(&init);
1334 if (copy_to_user(argp, &init, sizeof(init)))
1344 static unsigned long nvhe_percpu_size(void)
1346 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1347 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1350 static unsigned long nvhe_percpu_order(void)
1352 unsigned long size = nvhe_percpu_size();
1354 return size ? get_order(size) : 0;
1357 /* A lookup table holding the hypervisor VA for each vector slot */
1358 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1360 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1362 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1365 static int kvm_init_vector_slots(void)
1370 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1371 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1373 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1374 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1376 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1380 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1381 __BP_HARDEN_HYP_VECS_SZ, &base);
1386 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1387 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1391 static void cpu_prepare_hyp_mode(int cpu)
1393 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1397 * Calculate the raw per-cpu offset without a translation from the
1398 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1399 * so that we can use adr_l to access per-cpu variables in EL2.
1400 * Also drop the KASAN tag which gets in the way...
1402 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1403 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1405 params->mair_el2 = read_sysreg(mair_el1);
1408 * The ID map may be configured to use an extended virtual address
1409 * range. This is only the case if system RAM is out of range for the
1410 * currently configured page size and VA_BITS, in which case we will
1411 * also need the extended virtual range for the HYP ID map, or we won't
1412 * be able to enable the EL2 MMU.
1414 * However, at EL2, there is only one TTBR register, and we can't switch
1415 * between translation tables *and* update TCR_EL2.T0SZ at the same
1416 * time. Bottom line: we need to use the extended range with *both* our
1417 * translation tables.
1419 * So use the same T0SZ value we use for the ID map.
1421 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1422 tcr &= ~TCR_T0SZ_MASK;
1423 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1424 params->tcr_el2 = tcr;
1426 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1427 params->pgd_pa = kvm_mmu_get_httbr();
1428 if (is_protected_kvm_enabled())
1429 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1431 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1432 params->vttbr = params->vtcr = 0;
1435 * Flush the init params from the data cache because the struct will
1436 * be read while the MMU is off.
1438 kvm_flush_dcache_to_poc(params, sizeof(*params));
1441 static void hyp_install_host_vector(void)
1443 struct kvm_nvhe_init_params *params;
1444 struct arm_smccc_res res;
1446 /* Switch from the HYP stub to our own HYP init vector */
1447 __hyp_set_vectors(kvm_get_idmap_vector());
1450 * Call initialization code, and switch to the full blown HYP code.
1451 * If the cpucaps haven't been finalized yet, something has gone very
1452 * wrong, and hyp will crash and burn when it uses any
1453 * cpus_have_const_cap() wrapper.
1455 BUG_ON(!system_capabilities_finalized());
1456 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1457 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1458 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1461 static void cpu_init_hyp_mode(void)
1463 hyp_install_host_vector();
1466 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1469 if (this_cpu_has_cap(ARM64_SSBS) &&
1470 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1471 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1475 static void cpu_hyp_reset(void)
1477 if (!is_kernel_in_hyp_mode())
1478 __hyp_reset_vectors();
1482 * EL2 vectors can be mapped and rerouted in a number of ways,
1483 * depending on the kernel configuration and CPU present:
1485 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1486 * placed in one of the vector slots, which is executed before jumping
1487 * to the real vectors.
1489 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1490 * containing the hardening sequence is mapped next to the idmap page,
1491 * and executed before jumping to the real vectors.
1493 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1494 * empty slot is selected, mapped next to the idmap page, and
1495 * executed before jumping to the real vectors.
1497 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1498 * VHE, as we don't have hypervisor-specific mappings. If the system
1499 * is VHE and yet selects this capability, it will be ignored.
1501 static void cpu_set_hyp_vector(void)
1503 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1504 void *vector = hyp_spectre_vector_selector[data->slot];
1506 if (!is_protected_kvm_enabled())
1507 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1509 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1512 static void cpu_hyp_reinit(void)
1514 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1518 if (is_kernel_in_hyp_mode())
1519 kvm_timer_init_vhe();
1521 cpu_init_hyp_mode();
1523 cpu_set_hyp_vector();
1525 kvm_arm_init_debug();
1528 kvm_vgic_init_cpu_hardware();
1531 static void _kvm_arch_hardware_enable(void *discard)
1533 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1535 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1539 int kvm_arch_hardware_enable(void)
1541 _kvm_arch_hardware_enable(NULL);
1545 static void _kvm_arch_hardware_disable(void *discard)
1547 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1549 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1553 void kvm_arch_hardware_disable(void)
1555 if (!is_protected_kvm_enabled())
1556 _kvm_arch_hardware_disable(NULL);
1559 #ifdef CONFIG_CPU_PM
1560 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1565 * kvm_arm_hardware_enabled is left with its old value over
1566 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1571 if (__this_cpu_read(kvm_arm_hardware_enabled))
1573 * don't update kvm_arm_hardware_enabled here
1574 * so that the hardware will be re-enabled
1575 * when we resume. See below.
1580 case CPU_PM_ENTER_FAILED:
1582 if (__this_cpu_read(kvm_arm_hardware_enabled))
1583 /* The hardware was enabled before suspend. */
1593 static struct notifier_block hyp_init_cpu_pm_nb = {
1594 .notifier_call = hyp_init_cpu_pm_notifier,
1597 static void hyp_cpu_pm_init(void)
1599 if (!is_protected_kvm_enabled())
1600 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1602 static void hyp_cpu_pm_exit(void)
1604 if (!is_protected_kvm_enabled())
1605 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1608 static inline void hyp_cpu_pm_init(void)
1611 static inline void hyp_cpu_pm_exit(void)
1616 static void init_cpu_logical_map(void)
1621 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1622 * Only copy the set of online CPUs whose features have been chacked
1623 * against the finalized system capabilities. The hypervisor will not
1624 * allow any other CPUs from the `possible` set to boot.
1626 for_each_online_cpu(cpu)
1627 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1630 #define init_psci_0_1_impl_state(config, what) \
1631 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1633 static bool init_psci_relay(void)
1636 * If PSCI has not been initialized, protected KVM cannot install
1637 * itself on newly booted CPUs.
1639 if (!psci_ops.get_version) {
1640 kvm_err("Cannot initialize protected mode without PSCI\n");
1644 kvm_host_psci_config.version = psci_ops.get_version();
1646 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1647 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1648 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1649 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1650 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1651 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1656 static int init_common_resources(void)
1658 return kvm_set_ipa_limit();
1661 static int init_subsystems(void)
1666 * Enable hardware so that subsystem initialisation can access EL2.
1668 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1671 * Register CPU lower-power notifier
1676 * Init HYP view of VGIC
1678 err = kvm_vgic_hyp_init();
1681 vgic_present = true;
1685 vgic_present = false;
1693 * Init HYP architected timer support
1695 err = kvm_timer_hyp_init(vgic_present);
1700 kvm_sys_reg_table_init();
1703 if (err || !is_protected_kvm_enabled())
1704 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1709 static void teardown_hyp_mode(void)
1714 for_each_possible_cpu(cpu) {
1715 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1716 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1720 static int do_pkvm_init(u32 hyp_va_bits)
1722 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1726 hyp_install_host_vector();
1727 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1728 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1735 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1737 void *addr = phys_to_virt(hyp_mem_base);
1740 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1741 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1743 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1747 ret = do_pkvm_init(hyp_va_bits);
1757 * Inits Hyp-mode on all online CPUs
1759 static int init_hyp_mode(void)
1766 * The protected Hyp-mode cannot be initialized if the memory pool
1767 * allocation has failed.
1769 if (is_protected_kvm_enabled() && !hyp_mem_base)
1773 * Allocate Hyp PGD and setup Hyp identity mapping
1775 err = kvm_mmu_init(&hyp_va_bits);
1780 * Allocate stack pages for Hypervisor-mode
1782 for_each_possible_cpu(cpu) {
1783 unsigned long stack_page;
1785 stack_page = __get_free_page(GFP_KERNEL);
1791 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1795 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1797 for_each_possible_cpu(cpu) {
1801 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1807 page_addr = page_address(page);
1808 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1809 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1813 * Map the Hyp-code called directly from the host
1815 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1816 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1818 kvm_err("Cannot map world-switch code\n");
1822 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1823 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1825 kvm_err("Cannot map .hyp.rodata section\n");
1829 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1830 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1832 kvm_err("Cannot map rodata section\n");
1837 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1838 * section thanks to an assertion in the linker script. Map it RW and
1839 * the rest of .bss RO.
1841 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1842 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1844 kvm_err("Cannot map hyp bss section: %d\n", err);
1848 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1849 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1851 kvm_err("Cannot map bss section\n");
1856 * Map the Hyp stack pages
1858 for_each_possible_cpu(cpu) {
1859 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1860 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1864 kvm_err("Cannot map hyp stack\n");
1869 for_each_possible_cpu(cpu) {
1870 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1871 char *percpu_end = percpu_begin + nvhe_percpu_size();
1873 /* Map Hyp percpu pages */
1874 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1876 kvm_err("Cannot map hyp percpu region\n");
1880 /* Prepare the CPU initialization parameters */
1881 cpu_prepare_hyp_mode(cpu);
1884 if (is_protected_kvm_enabled()) {
1885 init_cpu_logical_map();
1887 if (!init_psci_relay()) {
1893 if (is_protected_kvm_enabled()) {
1894 err = kvm_hyp_init_protection(hyp_va_bits);
1896 kvm_err("Failed to init hyp memory protection\n");
1904 teardown_hyp_mode();
1905 kvm_err("error initializing Hyp mode: %d\n", err);
1909 static void _kvm_host_prot_finalize(void *discard)
1911 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1914 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1916 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1919 #define pkvm_mark_hyp_section(__section) \
1920 pkvm_mark_hyp(__pa_symbol(__section##_start), \
1921 __pa_symbol(__section##_end))
1923 static int finalize_hyp_mode(void)
1927 if (!is_protected_kvm_enabled())
1930 ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1934 ret = pkvm_mark_hyp_section(__hyp_text);
1938 ret = pkvm_mark_hyp_section(__hyp_rodata);
1942 ret = pkvm_mark_hyp_section(__hyp_bss);
1946 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1950 for_each_possible_cpu(cpu) {
1951 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1952 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
1954 ret = pkvm_mark_hyp(start, end);
1958 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
1959 end = start + PAGE_SIZE;
1960 ret = pkvm_mark_hyp(start, end);
1966 * Flip the static key upfront as that may no longer be possible
1967 * once the host stage 2 is installed.
1969 static_branch_enable(&kvm_protected_mode_initialized);
1970 on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
1975 static void check_kvm_target_cpu(void *ret)
1977 *(int *)ret = kvm_target_cpu();
1980 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1982 struct kvm_vcpu *vcpu;
1985 mpidr &= MPIDR_HWID_BITMASK;
1986 kvm_for_each_vcpu(i, vcpu, kvm) {
1987 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1993 bool kvm_arch_has_irq_bypass(void)
1998 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1999 struct irq_bypass_producer *prod)
2001 struct kvm_kernel_irqfd *irqfd =
2002 container_of(cons, struct kvm_kernel_irqfd, consumer);
2004 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2007 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2008 struct irq_bypass_producer *prod)
2010 struct kvm_kernel_irqfd *irqfd =
2011 container_of(cons, struct kvm_kernel_irqfd, consumer);
2013 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2017 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2019 struct kvm_kernel_irqfd *irqfd =
2020 container_of(cons, struct kvm_kernel_irqfd, consumer);
2022 kvm_arm_halt_guest(irqfd->kvm);
2025 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2027 struct kvm_kernel_irqfd *irqfd =
2028 container_of(cons, struct kvm_kernel_irqfd, consumer);
2030 kvm_arm_resume_guest(irqfd->kvm);
2034 * Initialize Hyp-mode and memory mappings on all CPUs.
2036 int kvm_arch_init(void *opaque)
2042 if (!is_hyp_mode_available()) {
2043 kvm_info("HYP mode not available\n");
2047 in_hyp_mode = is_kernel_in_hyp_mode();
2049 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2050 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2051 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2052 "Only trusted guests should be used on this system.\n");
2054 for_each_online_cpu(cpu) {
2055 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
2057 kvm_err("Error, CPU %d not supported!\n", cpu);
2062 err = init_common_resources();
2066 err = kvm_arm_init_sve();
2071 err = init_hyp_mode();
2076 err = kvm_init_vector_slots();
2078 kvm_err("Cannot initialise vector slots\n");
2082 err = init_subsystems();
2087 err = finalize_hyp_mode();
2089 kvm_err("Failed to finalize Hyp protection\n");
2094 if (is_protected_kvm_enabled()) {
2095 kvm_info("Protected nVHE mode initialized successfully\n");
2096 } else if (in_hyp_mode) {
2097 kvm_info("VHE mode initialized successfully\n");
2099 kvm_info("Hyp mode initialized successfully\n");
2107 teardown_hyp_mode();
2112 /* NOP: Compiling as a module not supported */
2113 void kvm_arch_exit(void)
2115 kvm_perf_teardown();
2118 static int __init early_kvm_mode_cfg(char *arg)
2123 if (strcmp(arg, "protected") == 0) {
2124 kvm_mode = KVM_MODE_PROTECTED;
2128 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2133 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2135 enum kvm_mode kvm_get_mode(void)
2140 static int arm_init(void)
2142 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2146 module_init(arm_init);