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;
97 if (!system_supports_mte() || kvm->created_vcpus)
100 kvm->arch.mte_enabled = true;
110 static int kvm_arm_default_max_vcpus(void)
112 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
115 static void set_default_spectre(struct kvm *kvm)
118 * The default is to expose CSV2 == 1 if the HW isn't affected.
119 * Although this is a per-CPU feature, we make it global because
120 * asymmetric systems are just a nuisance.
122 * Userspace can override this as long as it doesn't promise
125 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
126 kvm->arch.pfr0_csv2 = 1;
127 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
128 kvm->arch.pfr0_csv3 = 1;
132 * kvm_arch_init_vm - initializes a VM data structure
133 * @kvm: pointer to the KVM struct
135 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
139 ret = kvm_arm_setup_stage2(kvm, type);
143 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
147 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
149 goto out_free_stage2_pgd;
151 kvm_vgic_early_init(kvm);
153 /* The maximum number of VCPUs is limited by the host's GIC model */
154 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
156 set_default_spectre(kvm);
160 kvm_free_stage2_pgd(&kvm->arch.mmu);
164 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
166 return VM_FAULT_SIGBUS;
171 * kvm_arch_destroy_vm - destroy the VM data structure
172 * @kvm: pointer to the KVM struct
174 void kvm_arch_destroy_vm(struct kvm *kvm)
178 bitmap_free(kvm->arch.pmu_filter);
180 kvm_vgic_destroy(kvm);
182 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
184 kvm_vcpu_destroy(kvm->vcpus[i]);
185 kvm->vcpus[i] = NULL;
188 atomic_set(&kvm->online_vcpus, 0);
191 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
195 case KVM_CAP_IRQCHIP:
198 case KVM_CAP_IOEVENTFD:
199 case KVM_CAP_DEVICE_CTRL:
200 case KVM_CAP_USER_MEMORY:
201 case KVM_CAP_SYNC_MMU:
202 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
203 case KVM_CAP_ONE_REG:
204 case KVM_CAP_ARM_PSCI:
205 case KVM_CAP_ARM_PSCI_0_2:
206 case KVM_CAP_READONLY_MEM:
207 case KVM_CAP_MP_STATE:
208 case KVM_CAP_IMMEDIATE_EXIT:
209 case KVM_CAP_VCPU_EVENTS:
210 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
211 case KVM_CAP_ARM_NISV_TO_USER:
212 case KVM_CAP_ARM_INJECT_EXT_DABT:
213 case KVM_CAP_SET_GUEST_DEBUG:
214 case KVM_CAP_VCPU_ATTRIBUTES:
215 case KVM_CAP_PTP_KVM:
218 case KVM_CAP_SET_GUEST_DEBUG2:
219 return KVM_GUESTDBG_VALID_MASK;
220 case KVM_CAP_ARM_SET_DEVICE_ADDR:
223 case KVM_CAP_NR_VCPUS:
224 r = num_online_cpus();
226 case KVM_CAP_MAX_VCPUS:
227 case KVM_CAP_MAX_VCPU_ID:
229 r = kvm->arch.max_vcpus;
231 r = kvm_arm_default_max_vcpus();
233 case KVM_CAP_MSI_DEVID:
237 r = kvm->arch.vgic.msis_require_devid;
239 case KVM_CAP_ARM_USER_IRQ:
241 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
242 * (bump this number if adding more devices)
246 case KVM_CAP_ARM_MTE:
247 r = system_supports_mte();
249 case KVM_CAP_STEAL_TIME:
250 r = kvm_arm_pvtime_supported();
252 case KVM_CAP_ARM_EL1_32BIT:
253 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
255 case KVM_CAP_GUEST_DEBUG_HW_BPS:
258 case KVM_CAP_GUEST_DEBUG_HW_WPS:
261 case KVM_CAP_ARM_PMU_V3:
262 r = kvm_arm_support_pmu_v3();
264 case KVM_CAP_ARM_INJECT_SERROR_ESR:
265 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
267 case KVM_CAP_ARM_VM_IPA_SIZE:
268 r = get_kvm_ipa_limit();
270 case KVM_CAP_ARM_SVE:
271 r = system_supports_sve();
273 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
274 case KVM_CAP_ARM_PTRAUTH_GENERIC:
275 r = system_has_full_ptr_auth();
284 long kvm_arch_dev_ioctl(struct file *filp,
285 unsigned int ioctl, unsigned long arg)
290 struct kvm *kvm_arch_alloc_vm(void)
293 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
295 return vzalloc(sizeof(struct kvm));
298 void kvm_arch_free_vm(struct kvm *kvm)
306 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
308 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
311 if (id >= kvm->arch.max_vcpus)
317 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
321 /* Force users to call KVM_ARM_VCPU_INIT */
322 vcpu->arch.target = -1;
323 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
325 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
327 /* Set up the timer */
328 kvm_timer_vcpu_init(vcpu);
330 kvm_pmu_vcpu_init(vcpu);
332 kvm_arm_reset_debug_ptr(vcpu);
334 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
336 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
338 err = kvm_vgic_vcpu_init(vcpu);
342 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
345 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
349 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
351 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
352 static_branch_dec(&userspace_irqchip_in_use);
354 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
355 kvm_timer_vcpu_terminate(vcpu);
356 kvm_pmu_vcpu_destroy(vcpu);
358 kvm_arm_vcpu_destroy(vcpu);
361 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
363 return kvm_timer_is_pending(vcpu);
366 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
369 * If we're about to block (most likely because we've just hit a
370 * WFI), we need to sync back the state of the GIC CPU interface
371 * so that we have the latest PMR and group enables. This ensures
372 * that kvm_arch_vcpu_runnable has up-to-date data to decide
373 * whether we have pending interrupts.
375 * For the same reason, we want to tell GICv4 that we need
376 * doorbells to be signalled, should an interrupt become pending.
379 kvm_vgic_vmcr_sync(vcpu);
380 vgic_v4_put(vcpu, true);
384 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
391 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
393 struct kvm_s2_mmu *mmu;
396 mmu = vcpu->arch.hw_mmu;
397 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
400 * We guarantee that both TLBs and I-cache are private to each
401 * vcpu. If detecting that a vcpu from the same VM has
402 * previously run on the same physical CPU, call into the
403 * hypervisor code to nuke the relevant contexts.
405 * We might get preempted before the vCPU actually runs, but
406 * over-invalidation doesn't affect correctness.
408 if (*last_ran != vcpu->vcpu_id) {
409 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
410 *last_ran = vcpu->vcpu_id;
416 kvm_timer_vcpu_load(vcpu);
418 kvm_vcpu_load_sysregs_vhe(vcpu);
419 kvm_arch_vcpu_load_fp(vcpu);
420 kvm_vcpu_pmu_restore_guest(vcpu);
421 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
422 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
424 if (single_task_running())
425 vcpu_clear_wfx_traps(vcpu);
427 vcpu_set_wfx_traps(vcpu);
429 if (vcpu_has_ptrauth(vcpu))
430 vcpu_ptrauth_disable(vcpu);
431 kvm_arch_vcpu_load_debug_state_flags(vcpu);
434 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
436 kvm_arch_vcpu_put_debug_state_flags(vcpu);
437 kvm_arch_vcpu_put_fp(vcpu);
439 kvm_vcpu_put_sysregs_vhe(vcpu);
440 kvm_timer_vcpu_put(vcpu);
442 kvm_vcpu_pmu_restore_host(vcpu);
447 static void vcpu_power_off(struct kvm_vcpu *vcpu)
449 vcpu->arch.power_off = true;
450 kvm_make_request(KVM_REQ_SLEEP, vcpu);
454 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
455 struct kvm_mp_state *mp_state)
457 if (vcpu->arch.power_off)
458 mp_state->mp_state = KVM_MP_STATE_STOPPED;
460 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
465 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
466 struct kvm_mp_state *mp_state)
470 switch (mp_state->mp_state) {
471 case KVM_MP_STATE_RUNNABLE:
472 vcpu->arch.power_off = false;
474 case KVM_MP_STATE_STOPPED:
475 vcpu_power_off(vcpu);
485 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
486 * @v: The VCPU pointer
488 * If the guest CPU is not waiting for interrupts or an interrupt line is
489 * asserted, the CPU is by definition runnable.
491 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
493 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
494 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
495 && !v->arch.power_off && !v->arch.pause);
498 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
500 return vcpu_mode_priv(vcpu);
503 /* Just ensure a guest exit from a particular CPU */
504 static void exit_vm_noop(void *info)
508 void force_vm_exit(const cpumask_t *mask)
511 smp_call_function_many(mask, exit_vm_noop, NULL, true);
516 * need_new_vmid_gen - check that the VMID is still valid
517 * @vmid: The VMID to check
519 * return true if there is a new generation of VMIDs being used
521 * The hardware supports a limited set of values with the value zero reserved
522 * for the host, so we check if an assigned value belongs to a previous
523 * generation, which requires us to assign a new value. If we're the first to
524 * use a VMID for the new generation, we must flush necessary caches and TLBs
527 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
529 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
530 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
531 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
535 * update_vmid - Update the vmid with a valid VMID for the current generation
536 * @vmid: The stage-2 VMID information struct
538 static void update_vmid(struct kvm_vmid *vmid)
540 if (!need_new_vmid_gen(vmid))
543 spin_lock(&kvm_vmid_lock);
546 * We need to re-check the vmid_gen here to ensure that if another vcpu
547 * already allocated a valid vmid for this vm, then this vcpu should
550 if (!need_new_vmid_gen(vmid)) {
551 spin_unlock(&kvm_vmid_lock);
555 /* First user of a new VMID generation? */
556 if (unlikely(kvm_next_vmid == 0)) {
557 atomic64_inc(&kvm_vmid_gen);
561 * On SMP we know no other CPUs can use this CPU's or each
562 * other's VMID after force_vm_exit returns since the
563 * kvm_vmid_lock blocks them from reentry to the guest.
565 force_vm_exit(cpu_all_mask);
567 * Now broadcast TLB + ICACHE invalidation over the inner
568 * shareable domain to make sure all data structures are
571 kvm_call_hyp(__kvm_flush_vm_context);
574 vmid->vmid = kvm_next_vmid;
576 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
579 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
581 spin_unlock(&kvm_vmid_lock);
584 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
586 struct kvm *kvm = vcpu->kvm;
589 if (likely(vcpu->arch.has_run_once))
592 if (!kvm_arm_vcpu_is_finalized(vcpu))
595 vcpu->arch.has_run_once = true;
597 kvm_arm_vcpu_init_debug(vcpu);
599 if (likely(irqchip_in_kernel(kvm))) {
601 * Map the VGIC hardware resources before running a vcpu the
602 * first time on this VM.
604 ret = kvm_vgic_map_resources(kvm);
609 * Tell the rest of the code that there are userspace irqchip
612 static_branch_inc(&userspace_irqchip_in_use);
615 ret = kvm_timer_enable(vcpu);
619 ret = kvm_arm_pmu_v3_enable(vcpu);
624 bool kvm_arch_intc_initialized(struct kvm *kvm)
626 return vgic_initialized(kvm);
629 void kvm_arm_halt_guest(struct kvm *kvm)
632 struct kvm_vcpu *vcpu;
634 kvm_for_each_vcpu(i, vcpu, kvm)
635 vcpu->arch.pause = true;
636 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
639 void kvm_arm_resume_guest(struct kvm *kvm)
642 struct kvm_vcpu *vcpu;
644 kvm_for_each_vcpu(i, vcpu, kvm) {
645 vcpu->arch.pause = false;
646 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
650 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
652 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
654 rcuwait_wait_event(wait,
655 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
658 if (vcpu->arch.power_off || vcpu->arch.pause) {
659 /* Awaken to handle a signal, request we sleep again later. */
660 kvm_make_request(KVM_REQ_SLEEP, vcpu);
664 * Make sure we will observe a potential reset request if we've
665 * observed a change to the power state. Pairs with the smp_wmb() in
666 * kvm_psci_vcpu_on().
671 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
673 return vcpu->arch.target >= 0;
676 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
678 if (kvm_request_pending(vcpu)) {
679 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
680 vcpu_req_sleep(vcpu);
682 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
683 kvm_reset_vcpu(vcpu);
686 * Clear IRQ_PENDING requests that were made to guarantee
687 * that a VCPU sees new virtual interrupts.
689 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
691 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
692 kvm_update_stolen_time(vcpu);
694 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
695 /* The distributor enable bits were changed */
697 vgic_v4_put(vcpu, false);
705 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
706 * @vcpu: The VCPU pointer
708 * This function is called through the VCPU_RUN ioctl called from user space. It
709 * will execute VM code in a loop until the time slice for the process is used
710 * or some emulation is needed from user space in which case the function will
711 * return with return value 0 and with the kvm_run structure filled in with the
712 * required data for the requested emulation.
714 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
716 struct kvm_run *run = vcpu->run;
719 if (unlikely(!kvm_vcpu_initialized(vcpu)))
722 ret = kvm_vcpu_first_run_init(vcpu);
726 if (run->exit_reason == KVM_EXIT_MMIO) {
727 ret = kvm_handle_mmio_return(vcpu);
734 if (run->immediate_exit) {
739 kvm_sigset_activate(vcpu);
742 run->exit_reason = KVM_EXIT_UNKNOWN;
745 * Check conditions before entering the guest
749 update_vmid(&vcpu->arch.hw_mmu->vmid);
751 check_vcpu_requests(vcpu);
754 * Preparing the interrupts to be injected also
755 * involves poking the GIC, which must be done in a
756 * non-preemptible context.
760 kvm_pmu_flush_hwstate(vcpu);
764 kvm_vgic_flush_hwstate(vcpu);
767 * Exit if we have a signal pending so that we can deliver the
768 * signal to user space.
770 if (signal_pending(current)) {
772 run->exit_reason = KVM_EXIT_INTR;
776 * If we're using a userspace irqchip, then check if we need
777 * to tell a userspace irqchip about timer or PMU level
778 * changes and if so, exit to userspace (the actual level
779 * state gets updated in kvm_timer_update_run and
780 * kvm_pmu_update_run below).
782 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
783 if (kvm_timer_should_notify_user(vcpu) ||
784 kvm_pmu_should_notify_user(vcpu)) {
786 run->exit_reason = KVM_EXIT_INTR;
791 * Ensure we set mode to IN_GUEST_MODE after we disable
792 * interrupts and before the final VCPU requests check.
793 * See the comment in kvm_vcpu_exiting_guest_mode() and
794 * Documentation/virt/kvm/vcpu-requests.rst
796 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
798 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
799 kvm_request_pending(vcpu)) {
800 vcpu->mode = OUTSIDE_GUEST_MODE;
801 isb(); /* Ensure work in x_flush_hwstate is committed */
802 kvm_pmu_sync_hwstate(vcpu);
803 if (static_branch_unlikely(&userspace_irqchip_in_use))
804 kvm_timer_sync_user(vcpu);
805 kvm_vgic_sync_hwstate(vcpu);
811 kvm_arm_setup_debug(vcpu);
813 /**************************************************************
816 trace_kvm_entry(*vcpu_pc(vcpu));
817 guest_enter_irqoff();
819 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
821 vcpu->mode = OUTSIDE_GUEST_MODE;
825 *************************************************************/
827 kvm_arm_clear_debug(vcpu);
830 * We must sync the PMU state before the vgic state so
831 * that the vgic can properly sample the updated state of the
834 kvm_pmu_sync_hwstate(vcpu);
837 * Sync the vgic state before syncing the timer state because
838 * the timer code needs to know if the virtual timer
839 * interrupts are active.
841 kvm_vgic_sync_hwstate(vcpu);
844 * Sync the timer hardware state before enabling interrupts as
845 * we don't want vtimer interrupts to race with syncing the
846 * timer virtual interrupt state.
848 if (static_branch_unlikely(&userspace_irqchip_in_use))
849 kvm_timer_sync_user(vcpu);
851 kvm_arch_vcpu_ctxsync_fp(vcpu);
854 * We may have taken a host interrupt in HYP mode (ie
855 * while executing the guest). This interrupt is still
856 * pending, as we haven't serviced it yet!
858 * We're now back in SVC mode, with interrupts
859 * disabled. Enabling the interrupts now will have
860 * the effect of taking the interrupt again, in SVC
866 * We do local_irq_enable() before calling guest_exit() so
867 * that if a timer interrupt hits while running the guest we
868 * account that tick as being spent in the guest. We enable
869 * preemption after calling guest_exit() so that if we get
870 * preempted we make sure ticks after that is not counted as
874 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
876 /* Exit types that need handling before we can be preempted */
877 handle_exit_early(vcpu, ret);
882 * The ARMv8 architecture doesn't give the hypervisor
883 * a mechanism to prevent a guest from dropping to AArch32 EL0
884 * if implemented by the CPU. If we spot the guest in such
885 * state and that we decided it wasn't supposed to do so (like
886 * with the asymmetric AArch32 case), return to userspace with
889 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
891 * As we have caught the guest red-handed, decide that
892 * it isn't fit for purpose anymore by making the vcpu
893 * invalid. The VMM can try and fix it by issuing a
894 * KVM_ARM_VCPU_INIT if it really wants to.
896 vcpu->arch.target = -1;
897 ret = ARM_EXCEPTION_IL;
900 ret = handle_exit(vcpu, ret);
903 /* Tell userspace about in-kernel device output levels */
904 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
905 kvm_timer_update_run(vcpu);
906 kvm_pmu_update_run(vcpu);
909 kvm_sigset_deactivate(vcpu);
913 * In the unlikely event that we are returning to userspace
914 * with pending exceptions or PC adjustment, commit these
915 * adjustments in order to give userspace a consistent view of
916 * the vcpu state. Note that this relies on __kvm_adjust_pc()
917 * being preempt-safe on VHE.
919 if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
920 KVM_ARM64_INCREMENT_PC)))
921 kvm_call_hyp(__kvm_adjust_pc, vcpu);
927 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
933 if (number == KVM_ARM_IRQ_CPU_IRQ)
934 bit_index = __ffs(HCR_VI);
935 else /* KVM_ARM_IRQ_CPU_FIQ */
936 bit_index = __ffs(HCR_VF);
938 hcr = vcpu_hcr(vcpu);
940 set = test_and_set_bit(bit_index, hcr);
942 set = test_and_clear_bit(bit_index, hcr);
945 * If we didn't change anything, no need to wake up or kick other CPUs
951 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
952 * trigger a world-switch round on the running physical CPU to set the
953 * virtual IRQ/FIQ fields in the HCR appropriately.
955 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
961 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
964 u32 irq = irq_level->irq;
965 unsigned int irq_type, vcpu_idx, irq_num;
966 int nrcpus = atomic_read(&kvm->online_vcpus);
967 struct kvm_vcpu *vcpu = NULL;
968 bool level = irq_level->level;
970 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
971 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
972 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
973 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
975 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
978 case KVM_ARM_IRQ_TYPE_CPU:
979 if (irqchip_in_kernel(kvm))
982 if (vcpu_idx >= nrcpus)
985 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
989 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
992 return vcpu_interrupt_line(vcpu, irq_num, level);
993 case KVM_ARM_IRQ_TYPE_PPI:
994 if (!irqchip_in_kernel(kvm))
997 if (vcpu_idx >= nrcpus)
1000 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1004 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1007 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1008 case KVM_ARM_IRQ_TYPE_SPI:
1009 if (!irqchip_in_kernel(kvm))
1012 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1015 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1021 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1022 const struct kvm_vcpu_init *init)
1024 unsigned int i, ret;
1025 int phys_target = kvm_target_cpu();
1027 if (init->target != phys_target)
1031 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1032 * use the same target.
1034 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1037 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1038 for (i = 0; i < sizeof(init->features) * 8; i++) {
1039 bool set = (init->features[i / 32] & (1 << (i % 32)));
1041 if (set && i >= KVM_VCPU_MAX_FEATURES)
1045 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1046 * use the same feature set.
1048 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1049 test_bit(i, vcpu->arch.features) != set)
1053 set_bit(i, vcpu->arch.features);
1056 vcpu->arch.target = phys_target;
1058 /* Now we know what it is, we can reset it. */
1059 ret = kvm_reset_vcpu(vcpu);
1061 vcpu->arch.target = -1;
1062 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1068 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1069 struct kvm_vcpu_init *init)
1073 ret = kvm_vcpu_set_target(vcpu, init);
1078 * Ensure a rebooted VM will fault in RAM pages and detect if the
1079 * guest MMU is turned off and flush the caches as needed.
1081 * S2FWB enforces all memory accesses to RAM being cacheable,
1082 * ensuring that the data side is always coherent. We still
1083 * need to invalidate the I-cache though, as FWB does *not*
1084 * imply CTR_EL0.DIC.
1086 if (vcpu->arch.has_run_once) {
1087 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1088 stage2_unmap_vm(vcpu->kvm);
1090 __flush_icache_all();
1093 vcpu_reset_hcr(vcpu);
1096 * Handle the "start in power-off" case.
1098 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1099 vcpu_power_off(vcpu);
1101 vcpu->arch.power_off = false;
1106 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1107 struct kvm_device_attr *attr)
1111 switch (attr->group) {
1113 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1120 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1121 struct kvm_device_attr *attr)
1125 switch (attr->group) {
1127 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1134 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1135 struct kvm_device_attr *attr)
1139 switch (attr->group) {
1141 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1148 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1149 struct kvm_vcpu_events *events)
1151 memset(events, 0, sizeof(*events));
1153 return __kvm_arm_vcpu_get_events(vcpu, events);
1156 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1157 struct kvm_vcpu_events *events)
1161 /* check whether the reserved field is zero */
1162 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1163 if (events->reserved[i])
1166 /* check whether the pad field is zero */
1167 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1168 if (events->exception.pad[i])
1171 return __kvm_arm_vcpu_set_events(vcpu, events);
1174 long kvm_arch_vcpu_ioctl(struct file *filp,
1175 unsigned int ioctl, unsigned long arg)
1177 struct kvm_vcpu *vcpu = filp->private_data;
1178 void __user *argp = (void __user *)arg;
1179 struct kvm_device_attr attr;
1183 case KVM_ARM_VCPU_INIT: {
1184 struct kvm_vcpu_init init;
1187 if (copy_from_user(&init, argp, sizeof(init)))
1190 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1193 case KVM_SET_ONE_REG:
1194 case KVM_GET_ONE_REG: {
1195 struct kvm_one_reg reg;
1198 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1202 if (copy_from_user(®, argp, sizeof(reg)))
1205 if (ioctl == KVM_SET_ONE_REG)
1206 r = kvm_arm_set_reg(vcpu, ®);
1208 r = kvm_arm_get_reg(vcpu, ®);
1211 case KVM_GET_REG_LIST: {
1212 struct kvm_reg_list __user *user_list = argp;
1213 struct kvm_reg_list reg_list;
1217 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1221 if (!kvm_arm_vcpu_is_finalized(vcpu))
1225 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1228 reg_list.n = kvm_arm_num_regs(vcpu);
1229 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1234 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1237 case KVM_SET_DEVICE_ATTR: {
1239 if (copy_from_user(&attr, argp, sizeof(attr)))
1241 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1244 case KVM_GET_DEVICE_ATTR: {
1246 if (copy_from_user(&attr, argp, sizeof(attr)))
1248 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1251 case KVM_HAS_DEVICE_ATTR: {
1253 if (copy_from_user(&attr, argp, sizeof(attr)))
1255 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1258 case KVM_GET_VCPU_EVENTS: {
1259 struct kvm_vcpu_events events;
1261 if (kvm_arm_vcpu_get_events(vcpu, &events))
1264 if (copy_to_user(argp, &events, sizeof(events)))
1269 case KVM_SET_VCPU_EVENTS: {
1270 struct kvm_vcpu_events events;
1272 if (copy_from_user(&events, argp, sizeof(events)))
1275 return kvm_arm_vcpu_set_events(vcpu, &events);
1277 case KVM_ARM_VCPU_FINALIZE: {
1280 if (!kvm_vcpu_initialized(vcpu))
1283 if (get_user(what, (const int __user *)argp))
1286 return kvm_arm_vcpu_finalize(vcpu, what);
1295 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1300 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1301 const struct kvm_memory_slot *memslot)
1303 kvm_flush_remote_tlbs(kvm);
1306 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1307 struct kvm_arm_device_addr *dev_addr)
1309 unsigned long dev_id, type;
1311 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1312 KVM_ARM_DEVICE_ID_SHIFT;
1313 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1314 KVM_ARM_DEVICE_TYPE_SHIFT;
1317 case KVM_ARM_DEVICE_VGIC_V2:
1320 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1326 long kvm_arch_vm_ioctl(struct file *filp,
1327 unsigned int ioctl, unsigned long arg)
1329 struct kvm *kvm = filp->private_data;
1330 void __user *argp = (void __user *)arg;
1333 case KVM_CREATE_IRQCHIP: {
1337 mutex_lock(&kvm->lock);
1338 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1339 mutex_unlock(&kvm->lock);
1342 case KVM_ARM_SET_DEVICE_ADDR: {
1343 struct kvm_arm_device_addr dev_addr;
1345 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1347 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1349 case KVM_ARM_PREFERRED_TARGET: {
1351 struct kvm_vcpu_init init;
1353 err = kvm_vcpu_preferred_target(&init);
1357 if (copy_to_user(argp, &init, sizeof(init)))
1367 static unsigned long nvhe_percpu_size(void)
1369 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1370 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1373 static unsigned long nvhe_percpu_order(void)
1375 unsigned long size = nvhe_percpu_size();
1377 return size ? get_order(size) : 0;
1380 /* A lookup table holding the hypervisor VA for each vector slot */
1381 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1383 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1385 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1388 static int kvm_init_vector_slots(void)
1393 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1394 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1396 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1397 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1399 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1403 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1404 __BP_HARDEN_HYP_VECS_SZ, &base);
1409 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1410 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1414 static void cpu_prepare_hyp_mode(int cpu)
1416 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1420 * Calculate the raw per-cpu offset without a translation from the
1421 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1422 * so that we can use adr_l to access per-cpu variables in EL2.
1423 * Also drop the KASAN tag which gets in the way...
1425 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1426 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1428 params->mair_el2 = read_sysreg(mair_el1);
1431 * The ID map may be configured to use an extended virtual address
1432 * range. This is only the case if system RAM is out of range for the
1433 * currently configured page size and VA_BITS, in which case we will
1434 * also need the extended virtual range for the HYP ID map, or we won't
1435 * be able to enable the EL2 MMU.
1437 * However, at EL2, there is only one TTBR register, and we can't switch
1438 * between translation tables *and* update TCR_EL2.T0SZ at the same
1439 * time. Bottom line: we need to use the extended range with *both* our
1440 * translation tables.
1442 * So use the same T0SZ value we use for the ID map.
1444 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1445 tcr &= ~TCR_T0SZ_MASK;
1446 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1447 params->tcr_el2 = tcr;
1449 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1450 params->pgd_pa = kvm_mmu_get_httbr();
1451 if (is_protected_kvm_enabled())
1452 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1454 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1455 params->vttbr = params->vtcr = 0;
1458 * Flush the init params from the data cache because the struct will
1459 * be read while the MMU is off.
1461 kvm_flush_dcache_to_poc(params, sizeof(*params));
1464 static void hyp_install_host_vector(void)
1466 struct kvm_nvhe_init_params *params;
1467 struct arm_smccc_res res;
1469 /* Switch from the HYP stub to our own HYP init vector */
1470 __hyp_set_vectors(kvm_get_idmap_vector());
1473 * Call initialization code, and switch to the full blown HYP code.
1474 * If the cpucaps haven't been finalized yet, something has gone very
1475 * wrong, and hyp will crash and burn when it uses any
1476 * cpus_have_const_cap() wrapper.
1478 BUG_ON(!system_capabilities_finalized());
1479 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1480 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1481 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1484 static void cpu_init_hyp_mode(void)
1486 hyp_install_host_vector();
1489 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1492 if (this_cpu_has_cap(ARM64_SSBS) &&
1493 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1494 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1498 static void cpu_hyp_reset(void)
1500 if (!is_kernel_in_hyp_mode())
1501 __hyp_reset_vectors();
1505 * EL2 vectors can be mapped and rerouted in a number of ways,
1506 * depending on the kernel configuration and CPU present:
1508 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1509 * placed in one of the vector slots, which is executed before jumping
1510 * to the real vectors.
1512 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1513 * containing the hardening sequence is mapped next to the idmap page,
1514 * and executed before jumping to the real vectors.
1516 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1517 * empty slot is selected, mapped next to the idmap page, and
1518 * executed before jumping to the real vectors.
1520 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1521 * VHE, as we don't have hypervisor-specific mappings. If the system
1522 * is VHE and yet selects this capability, it will be ignored.
1524 static void cpu_set_hyp_vector(void)
1526 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1527 void *vector = hyp_spectre_vector_selector[data->slot];
1529 if (!is_protected_kvm_enabled())
1530 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1532 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1535 static void cpu_hyp_reinit(void)
1537 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1541 if (is_kernel_in_hyp_mode())
1542 kvm_timer_init_vhe();
1544 cpu_init_hyp_mode();
1546 cpu_set_hyp_vector();
1548 kvm_arm_init_debug();
1551 kvm_vgic_init_cpu_hardware();
1554 static void _kvm_arch_hardware_enable(void *discard)
1556 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1558 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1562 int kvm_arch_hardware_enable(void)
1564 _kvm_arch_hardware_enable(NULL);
1568 static void _kvm_arch_hardware_disable(void *discard)
1570 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1572 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1576 void kvm_arch_hardware_disable(void)
1578 if (!is_protected_kvm_enabled())
1579 _kvm_arch_hardware_disable(NULL);
1582 #ifdef CONFIG_CPU_PM
1583 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1588 * kvm_arm_hardware_enabled is left with its old value over
1589 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1594 if (__this_cpu_read(kvm_arm_hardware_enabled))
1596 * don't update kvm_arm_hardware_enabled here
1597 * so that the hardware will be re-enabled
1598 * when we resume. See below.
1603 case CPU_PM_ENTER_FAILED:
1605 if (__this_cpu_read(kvm_arm_hardware_enabled))
1606 /* The hardware was enabled before suspend. */
1616 static struct notifier_block hyp_init_cpu_pm_nb = {
1617 .notifier_call = hyp_init_cpu_pm_notifier,
1620 static void hyp_cpu_pm_init(void)
1622 if (!is_protected_kvm_enabled())
1623 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1625 static void hyp_cpu_pm_exit(void)
1627 if (!is_protected_kvm_enabled())
1628 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1631 static inline void hyp_cpu_pm_init(void)
1634 static inline void hyp_cpu_pm_exit(void)
1639 static void init_cpu_logical_map(void)
1644 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1645 * Only copy the set of online CPUs whose features have been chacked
1646 * against the finalized system capabilities. The hypervisor will not
1647 * allow any other CPUs from the `possible` set to boot.
1649 for_each_online_cpu(cpu)
1650 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1653 #define init_psci_0_1_impl_state(config, what) \
1654 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1656 static bool init_psci_relay(void)
1659 * If PSCI has not been initialized, protected KVM cannot install
1660 * itself on newly booted CPUs.
1662 if (!psci_ops.get_version) {
1663 kvm_err("Cannot initialize protected mode without PSCI\n");
1667 kvm_host_psci_config.version = psci_ops.get_version();
1669 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1670 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1671 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1672 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1673 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1674 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1679 static int init_common_resources(void)
1681 return kvm_set_ipa_limit();
1684 static int init_subsystems(void)
1689 * Enable hardware so that subsystem initialisation can access EL2.
1691 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1694 * Register CPU lower-power notifier
1699 * Init HYP view of VGIC
1701 err = kvm_vgic_hyp_init();
1704 vgic_present = true;
1708 vgic_present = false;
1716 * Init HYP architected timer support
1718 err = kvm_timer_hyp_init(vgic_present);
1723 kvm_sys_reg_table_init();
1726 if (err || !is_protected_kvm_enabled())
1727 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1732 static void teardown_hyp_mode(void)
1737 for_each_possible_cpu(cpu) {
1738 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1739 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1743 static int do_pkvm_init(u32 hyp_va_bits)
1745 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1749 hyp_install_host_vector();
1750 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1751 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1758 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1760 void *addr = phys_to_virt(hyp_mem_base);
1763 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1764 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1766 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1770 ret = do_pkvm_init(hyp_va_bits);
1780 * Inits Hyp-mode on all online CPUs
1782 static int init_hyp_mode(void)
1789 * The protected Hyp-mode cannot be initialized if the memory pool
1790 * allocation has failed.
1792 if (is_protected_kvm_enabled() && !hyp_mem_base)
1796 * Allocate Hyp PGD and setup Hyp identity mapping
1798 err = kvm_mmu_init(&hyp_va_bits);
1803 * Allocate stack pages for Hypervisor-mode
1805 for_each_possible_cpu(cpu) {
1806 unsigned long stack_page;
1808 stack_page = __get_free_page(GFP_KERNEL);
1814 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1818 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1820 for_each_possible_cpu(cpu) {
1824 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1830 page_addr = page_address(page);
1831 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1832 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1836 * Map the Hyp-code called directly from the host
1838 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1839 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1841 kvm_err("Cannot map world-switch code\n");
1845 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1846 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1848 kvm_err("Cannot map .hyp.rodata section\n");
1852 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1853 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1855 kvm_err("Cannot map rodata section\n");
1860 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1861 * section thanks to an assertion in the linker script. Map it RW and
1862 * the rest of .bss RO.
1864 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1865 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1867 kvm_err("Cannot map hyp bss section: %d\n", err);
1871 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1872 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1874 kvm_err("Cannot map bss section\n");
1879 * Map the Hyp stack pages
1881 for_each_possible_cpu(cpu) {
1882 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1883 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1887 kvm_err("Cannot map hyp stack\n");
1892 for_each_possible_cpu(cpu) {
1893 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1894 char *percpu_end = percpu_begin + nvhe_percpu_size();
1896 /* Map Hyp percpu pages */
1897 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1899 kvm_err("Cannot map hyp percpu region\n");
1903 /* Prepare the CPU initialization parameters */
1904 cpu_prepare_hyp_mode(cpu);
1907 if (is_protected_kvm_enabled()) {
1908 init_cpu_logical_map();
1910 if (!init_psci_relay()) {
1916 if (is_protected_kvm_enabled()) {
1917 err = kvm_hyp_init_protection(hyp_va_bits);
1919 kvm_err("Failed to init hyp memory protection\n");
1927 teardown_hyp_mode();
1928 kvm_err("error initializing Hyp mode: %d\n", err);
1932 static void _kvm_host_prot_finalize(void *discard)
1934 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1937 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1939 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1942 #define pkvm_mark_hyp_section(__section) \
1943 pkvm_mark_hyp(__pa_symbol(__section##_start), \
1944 __pa_symbol(__section##_end))
1946 static int finalize_hyp_mode(void)
1950 if (!is_protected_kvm_enabled())
1953 ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1957 ret = pkvm_mark_hyp_section(__hyp_text);
1961 ret = pkvm_mark_hyp_section(__hyp_rodata);
1965 ret = pkvm_mark_hyp_section(__hyp_bss);
1969 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1973 for_each_possible_cpu(cpu) {
1974 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1975 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
1977 ret = pkvm_mark_hyp(start, end);
1981 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
1982 end = start + PAGE_SIZE;
1983 ret = pkvm_mark_hyp(start, end);
1989 * Flip the static key upfront as that may no longer be possible
1990 * once the host stage 2 is installed.
1992 static_branch_enable(&kvm_protected_mode_initialized);
1993 on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
1998 static void check_kvm_target_cpu(void *ret)
2000 *(int *)ret = kvm_target_cpu();
2003 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2005 struct kvm_vcpu *vcpu;
2008 mpidr &= MPIDR_HWID_BITMASK;
2009 kvm_for_each_vcpu(i, vcpu, kvm) {
2010 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2016 bool kvm_arch_has_irq_bypass(void)
2021 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2022 struct irq_bypass_producer *prod)
2024 struct kvm_kernel_irqfd *irqfd =
2025 container_of(cons, struct kvm_kernel_irqfd, consumer);
2027 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2030 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2031 struct irq_bypass_producer *prod)
2033 struct kvm_kernel_irqfd *irqfd =
2034 container_of(cons, struct kvm_kernel_irqfd, consumer);
2036 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2040 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2042 struct kvm_kernel_irqfd *irqfd =
2043 container_of(cons, struct kvm_kernel_irqfd, consumer);
2045 kvm_arm_halt_guest(irqfd->kvm);
2048 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2050 struct kvm_kernel_irqfd *irqfd =
2051 container_of(cons, struct kvm_kernel_irqfd, consumer);
2053 kvm_arm_resume_guest(irqfd->kvm);
2057 * Initialize Hyp-mode and memory mappings on all CPUs.
2059 int kvm_arch_init(void *opaque)
2065 if (!is_hyp_mode_available()) {
2066 kvm_info("HYP mode not available\n");
2070 in_hyp_mode = is_kernel_in_hyp_mode();
2072 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2073 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2074 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2075 "Only trusted guests should be used on this system.\n");
2077 for_each_online_cpu(cpu) {
2078 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
2080 kvm_err("Error, CPU %d not supported!\n", cpu);
2085 err = init_common_resources();
2089 err = kvm_arm_init_sve();
2094 err = init_hyp_mode();
2099 err = kvm_init_vector_slots();
2101 kvm_err("Cannot initialise vector slots\n");
2105 err = init_subsystems();
2110 err = finalize_hyp_mode();
2112 kvm_err("Failed to finalize Hyp protection\n");
2117 if (is_protected_kvm_enabled()) {
2118 kvm_info("Protected nVHE mode initialized successfully\n");
2119 } else if (in_hyp_mode) {
2120 kvm_info("VHE mode initialized successfully\n");
2122 kvm_info("Hyp mode initialized successfully\n");
2130 teardown_hyp_mode();
2135 /* NOP: Compiling as a module not supported */
2136 void kvm_arch_exit(void)
2138 kvm_perf_teardown();
2141 static int __init early_kvm_mode_cfg(char *arg)
2146 if (strcmp(arg, "protected") == 0) {
2147 kvm_mode = KVM_MODE_PROTECTED;
2151 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2156 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2158 enum kvm_mode kvm_get_mode(void)
2163 static int arm_init(void)
2165 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2169 module_init(arm_init);