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>
45 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
46 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
48 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
50 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
51 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
52 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
54 /* The VMID used in the VTTBR */
55 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
56 static u32 kvm_next_vmid;
57 static DEFINE_SPINLOCK(kvm_vmid_lock);
59 static bool vgic_present;
61 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
62 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
64 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
66 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
69 int kvm_arch_hardware_setup(void *opaque)
74 int kvm_arch_check_processor_compat(void *opaque)
79 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
80 struct kvm_enable_cap *cap)
88 case KVM_CAP_ARM_NISV_TO_USER:
90 kvm->arch.return_nisv_io_abort_to_user = true;
93 if (!system_supports_mte() || kvm->created_vcpus)
96 kvm->arch.mte_enabled = true;
106 static int kvm_arm_default_max_vcpus(void)
108 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
111 static void set_default_spectre(struct kvm *kvm)
114 * The default is to expose CSV2 == 1 if the HW isn't affected.
115 * Although this is a per-CPU feature, we make it global because
116 * asymmetric systems are just a nuisance.
118 * Userspace can override this as long as it doesn't promise
121 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
122 kvm->arch.pfr0_csv2 = 1;
123 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
124 kvm->arch.pfr0_csv3 = 1;
128 * kvm_arch_init_vm - initializes a VM data structure
129 * @kvm: pointer to the KVM struct
131 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
135 ret = kvm_arm_setup_stage2(kvm, type);
139 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
143 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
145 goto out_free_stage2_pgd;
147 kvm_vgic_early_init(kvm);
149 /* The maximum number of VCPUs is limited by the host's GIC model */
150 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
152 set_default_spectre(kvm);
156 kvm_free_stage2_pgd(&kvm->arch.mmu);
160 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
162 return VM_FAULT_SIGBUS;
167 * kvm_arch_destroy_vm - destroy the VM data structure
168 * @kvm: pointer to the KVM struct
170 void kvm_arch_destroy_vm(struct kvm *kvm)
174 bitmap_free(kvm->arch.pmu_filter);
176 kvm_vgic_destroy(kvm);
178 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
180 kvm_vcpu_destroy(kvm->vcpus[i]);
181 kvm->vcpus[i] = NULL;
184 atomic_set(&kvm->online_vcpus, 0);
187 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
191 case KVM_CAP_IRQCHIP:
194 case KVM_CAP_IOEVENTFD:
195 case KVM_CAP_DEVICE_CTRL:
196 case KVM_CAP_USER_MEMORY:
197 case KVM_CAP_SYNC_MMU:
198 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
199 case KVM_CAP_ONE_REG:
200 case KVM_CAP_ARM_PSCI:
201 case KVM_CAP_ARM_PSCI_0_2:
202 case KVM_CAP_READONLY_MEM:
203 case KVM_CAP_MP_STATE:
204 case KVM_CAP_IMMEDIATE_EXIT:
205 case KVM_CAP_VCPU_EVENTS:
206 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
207 case KVM_CAP_ARM_NISV_TO_USER:
208 case KVM_CAP_ARM_INJECT_EXT_DABT:
209 case KVM_CAP_SET_GUEST_DEBUG:
210 case KVM_CAP_VCPU_ATTRIBUTES:
211 case KVM_CAP_PTP_KVM:
214 case KVM_CAP_SET_GUEST_DEBUG2:
215 return KVM_GUESTDBG_VALID_MASK;
216 case KVM_CAP_ARM_SET_DEVICE_ADDR:
219 case KVM_CAP_NR_VCPUS:
220 r = num_online_cpus();
222 case KVM_CAP_MAX_VCPUS:
223 case KVM_CAP_MAX_VCPU_ID:
225 r = kvm->arch.max_vcpus;
227 r = kvm_arm_default_max_vcpus();
229 case KVM_CAP_MSI_DEVID:
233 r = kvm->arch.vgic.msis_require_devid;
235 case KVM_CAP_ARM_USER_IRQ:
237 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
238 * (bump this number if adding more devices)
242 case KVM_CAP_ARM_MTE:
243 r = system_supports_mte();
245 case KVM_CAP_STEAL_TIME:
246 r = kvm_arm_pvtime_supported();
248 case KVM_CAP_ARM_EL1_32BIT:
249 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
251 case KVM_CAP_GUEST_DEBUG_HW_BPS:
254 case KVM_CAP_GUEST_DEBUG_HW_WPS:
257 case KVM_CAP_ARM_PMU_V3:
258 r = kvm_arm_support_pmu_v3();
260 case KVM_CAP_ARM_INJECT_SERROR_ESR:
261 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
263 case KVM_CAP_ARM_VM_IPA_SIZE:
264 r = get_kvm_ipa_limit();
266 case KVM_CAP_ARM_SVE:
267 r = system_supports_sve();
269 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
270 case KVM_CAP_ARM_PTRAUTH_GENERIC:
271 r = system_has_full_ptr_auth();
280 long kvm_arch_dev_ioctl(struct file *filp,
281 unsigned int ioctl, unsigned long arg)
286 struct kvm *kvm_arch_alloc_vm(void)
289 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
291 return vzalloc(sizeof(struct kvm));
294 void kvm_arch_free_vm(struct kvm *kvm)
302 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
304 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
307 if (id >= kvm->arch.max_vcpus)
313 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
317 /* Force users to call KVM_ARM_VCPU_INIT */
318 vcpu->arch.target = -1;
319 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
321 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
323 /* Set up the timer */
324 kvm_timer_vcpu_init(vcpu);
326 kvm_pmu_vcpu_init(vcpu);
328 kvm_arm_reset_debug_ptr(vcpu);
330 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
332 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
334 err = kvm_vgic_vcpu_init(vcpu);
338 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
341 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
345 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
347 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
348 static_branch_dec(&userspace_irqchip_in_use);
350 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
351 kvm_timer_vcpu_terminate(vcpu);
352 kvm_pmu_vcpu_destroy(vcpu);
354 kvm_arm_vcpu_destroy(vcpu);
357 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
359 return kvm_timer_is_pending(vcpu);
362 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
365 * If we're about to block (most likely because we've just hit a
366 * WFI), we need to sync back the state of the GIC CPU interface
367 * so that we have the latest PMR and group enables. This ensures
368 * that kvm_arch_vcpu_runnable has up-to-date data to decide
369 * whether we have pending interrupts.
371 * For the same reason, we want to tell GICv4 that we need
372 * doorbells to be signalled, should an interrupt become pending.
375 kvm_vgic_vmcr_sync(vcpu);
376 vgic_v4_put(vcpu, true);
380 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
387 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
389 struct kvm_s2_mmu *mmu;
392 mmu = vcpu->arch.hw_mmu;
393 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
396 * We guarantee that both TLBs and I-cache are private to each
397 * vcpu. If detecting that a vcpu from the same VM has
398 * previously run on the same physical CPU, call into the
399 * hypervisor code to nuke the relevant contexts.
401 * We might get preempted before the vCPU actually runs, but
402 * over-invalidation doesn't affect correctness.
404 if (*last_ran != vcpu->vcpu_id) {
405 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
406 *last_ran = vcpu->vcpu_id;
412 kvm_timer_vcpu_load(vcpu);
414 kvm_vcpu_load_sysregs_vhe(vcpu);
415 kvm_arch_vcpu_load_fp(vcpu);
416 kvm_vcpu_pmu_restore_guest(vcpu);
417 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
418 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
420 if (single_task_running())
421 vcpu_clear_wfx_traps(vcpu);
423 vcpu_set_wfx_traps(vcpu);
425 if (vcpu_has_ptrauth(vcpu))
426 vcpu_ptrauth_disable(vcpu);
427 kvm_arch_vcpu_load_debug_state_flags(vcpu);
430 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
432 kvm_arch_vcpu_put_debug_state_flags(vcpu);
433 kvm_arch_vcpu_put_fp(vcpu);
435 kvm_vcpu_put_sysregs_vhe(vcpu);
436 kvm_timer_vcpu_put(vcpu);
438 kvm_vcpu_pmu_restore_host(vcpu);
443 static void vcpu_power_off(struct kvm_vcpu *vcpu)
445 vcpu->arch.power_off = true;
446 kvm_make_request(KVM_REQ_SLEEP, vcpu);
450 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
451 struct kvm_mp_state *mp_state)
453 if (vcpu->arch.power_off)
454 mp_state->mp_state = KVM_MP_STATE_STOPPED;
456 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
461 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
462 struct kvm_mp_state *mp_state)
466 switch (mp_state->mp_state) {
467 case KVM_MP_STATE_RUNNABLE:
468 vcpu->arch.power_off = false;
470 case KVM_MP_STATE_STOPPED:
471 vcpu_power_off(vcpu);
481 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
482 * @v: The VCPU pointer
484 * If the guest CPU is not waiting for interrupts or an interrupt line is
485 * asserted, the CPU is by definition runnable.
487 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
489 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
490 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
491 && !v->arch.power_off && !v->arch.pause);
494 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
496 return vcpu_mode_priv(vcpu);
499 /* Just ensure a guest exit from a particular CPU */
500 static void exit_vm_noop(void *info)
504 void force_vm_exit(const cpumask_t *mask)
507 smp_call_function_many(mask, exit_vm_noop, NULL, true);
512 * need_new_vmid_gen - check that the VMID is still valid
513 * @vmid: The VMID to check
515 * return true if there is a new generation of VMIDs being used
517 * The hardware supports a limited set of values with the value zero reserved
518 * for the host, so we check if an assigned value belongs to a previous
519 * generation, which requires us to assign a new value. If we're the first to
520 * use a VMID for the new generation, we must flush necessary caches and TLBs
523 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
525 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
526 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
527 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
531 * update_vmid - Update the vmid with a valid VMID for the current generation
532 * @vmid: The stage-2 VMID information struct
534 static void update_vmid(struct kvm_vmid *vmid)
536 if (!need_new_vmid_gen(vmid))
539 spin_lock(&kvm_vmid_lock);
542 * We need to re-check the vmid_gen here to ensure that if another vcpu
543 * already allocated a valid vmid for this vm, then this vcpu should
546 if (!need_new_vmid_gen(vmid)) {
547 spin_unlock(&kvm_vmid_lock);
551 /* First user of a new VMID generation? */
552 if (unlikely(kvm_next_vmid == 0)) {
553 atomic64_inc(&kvm_vmid_gen);
557 * On SMP we know no other CPUs can use this CPU's or each
558 * other's VMID after force_vm_exit returns since the
559 * kvm_vmid_lock blocks them from reentry to the guest.
561 force_vm_exit(cpu_all_mask);
563 * Now broadcast TLB + ICACHE invalidation over the inner
564 * shareable domain to make sure all data structures are
567 kvm_call_hyp(__kvm_flush_vm_context);
570 vmid->vmid = kvm_next_vmid;
572 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
575 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
577 spin_unlock(&kvm_vmid_lock);
580 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
582 struct kvm *kvm = vcpu->kvm;
585 if (likely(vcpu->arch.has_run_once))
588 if (!kvm_arm_vcpu_is_finalized(vcpu))
591 vcpu->arch.has_run_once = true;
593 kvm_arm_vcpu_init_debug(vcpu);
595 if (likely(irqchip_in_kernel(kvm))) {
597 * Map the VGIC hardware resources before running a vcpu the
598 * first time on this VM.
600 ret = kvm_vgic_map_resources(kvm);
605 * Tell the rest of the code that there are userspace irqchip
608 static_branch_inc(&userspace_irqchip_in_use);
611 ret = kvm_timer_enable(vcpu);
615 ret = kvm_arm_pmu_v3_enable(vcpu);
620 bool kvm_arch_intc_initialized(struct kvm *kvm)
622 return vgic_initialized(kvm);
625 void kvm_arm_halt_guest(struct kvm *kvm)
628 struct kvm_vcpu *vcpu;
630 kvm_for_each_vcpu(i, vcpu, kvm)
631 vcpu->arch.pause = true;
632 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
635 void kvm_arm_resume_guest(struct kvm *kvm)
638 struct kvm_vcpu *vcpu;
640 kvm_for_each_vcpu(i, vcpu, kvm) {
641 vcpu->arch.pause = false;
642 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
646 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
648 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
650 rcuwait_wait_event(wait,
651 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
654 if (vcpu->arch.power_off || vcpu->arch.pause) {
655 /* Awaken to handle a signal, request we sleep again later. */
656 kvm_make_request(KVM_REQ_SLEEP, vcpu);
660 * Make sure we will observe a potential reset request if we've
661 * observed a change to the power state. Pairs with the smp_wmb() in
662 * kvm_psci_vcpu_on().
667 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
669 return vcpu->arch.target >= 0;
672 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
674 if (kvm_request_pending(vcpu)) {
675 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
676 vcpu_req_sleep(vcpu);
678 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
679 kvm_reset_vcpu(vcpu);
682 * Clear IRQ_PENDING requests that were made to guarantee
683 * that a VCPU sees new virtual interrupts.
685 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
687 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
688 kvm_update_stolen_time(vcpu);
690 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
691 /* The distributor enable bits were changed */
693 vgic_v4_put(vcpu, false);
698 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
699 kvm_pmu_handle_pmcr(vcpu,
700 __vcpu_sys_reg(vcpu, PMCR_EL0));
704 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
706 if (likely(!vcpu_mode_is_32bit(vcpu)))
709 return !system_supports_32bit_el0() ||
710 static_branch_unlikely(&arm64_mismatched_32bit_el0);
714 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
715 * @vcpu: The VCPU pointer
717 * This function is called through the VCPU_RUN ioctl called from user space. It
718 * will execute VM code in a loop until the time slice for the process is used
719 * or some emulation is needed from user space in which case the function will
720 * return with return value 0 and with the kvm_run structure filled in with the
721 * required data for the requested emulation.
723 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
725 struct kvm_run *run = vcpu->run;
728 if (unlikely(!kvm_vcpu_initialized(vcpu)))
731 ret = kvm_vcpu_first_run_init(vcpu);
735 if (run->exit_reason == KVM_EXIT_MMIO) {
736 ret = kvm_handle_mmio_return(vcpu);
743 if (run->immediate_exit) {
748 kvm_sigset_activate(vcpu);
751 run->exit_reason = KVM_EXIT_UNKNOWN;
754 * Check conditions before entering the guest
758 update_vmid(&vcpu->arch.hw_mmu->vmid);
760 check_vcpu_requests(vcpu);
763 * Preparing the interrupts to be injected also
764 * involves poking the GIC, which must be done in a
765 * non-preemptible context.
769 kvm_pmu_flush_hwstate(vcpu);
773 kvm_vgic_flush_hwstate(vcpu);
776 * Exit if we have a signal pending so that we can deliver the
777 * signal to user space.
779 if (signal_pending(current)) {
781 run->exit_reason = KVM_EXIT_INTR;
785 * If we're using a userspace irqchip, then check if we need
786 * to tell a userspace irqchip about timer or PMU level
787 * changes and if so, exit to userspace (the actual level
788 * state gets updated in kvm_timer_update_run and
789 * kvm_pmu_update_run below).
791 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
792 if (kvm_timer_should_notify_user(vcpu) ||
793 kvm_pmu_should_notify_user(vcpu)) {
795 run->exit_reason = KVM_EXIT_INTR;
800 * Ensure we set mode to IN_GUEST_MODE after we disable
801 * interrupts and before the final VCPU requests check.
802 * See the comment in kvm_vcpu_exiting_guest_mode() and
803 * Documentation/virt/kvm/vcpu-requests.rst
805 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
807 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
808 kvm_request_pending(vcpu)) {
809 vcpu->mode = OUTSIDE_GUEST_MODE;
810 isb(); /* Ensure work in x_flush_hwstate is committed */
811 kvm_pmu_sync_hwstate(vcpu);
812 if (static_branch_unlikely(&userspace_irqchip_in_use))
813 kvm_timer_sync_user(vcpu);
814 kvm_vgic_sync_hwstate(vcpu);
820 kvm_arm_setup_debug(vcpu);
822 /**************************************************************
825 trace_kvm_entry(*vcpu_pc(vcpu));
826 guest_enter_irqoff();
828 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
830 vcpu->mode = OUTSIDE_GUEST_MODE;
834 *************************************************************/
836 kvm_arm_clear_debug(vcpu);
839 * We must sync the PMU state before the vgic state so
840 * that the vgic can properly sample the updated state of the
843 kvm_pmu_sync_hwstate(vcpu);
846 * Sync the vgic state before syncing the timer state because
847 * the timer code needs to know if the virtual timer
848 * interrupts are active.
850 kvm_vgic_sync_hwstate(vcpu);
853 * Sync the timer hardware state before enabling interrupts as
854 * we don't want vtimer interrupts to race with syncing the
855 * timer virtual interrupt state.
857 if (static_branch_unlikely(&userspace_irqchip_in_use))
858 kvm_timer_sync_user(vcpu);
860 kvm_arch_vcpu_ctxsync_fp(vcpu);
863 * We may have taken a host interrupt in HYP mode (ie
864 * while executing the guest). This interrupt is still
865 * pending, as we haven't serviced it yet!
867 * We're now back in SVC mode, with interrupts
868 * disabled. Enabling the interrupts now will have
869 * the effect of taking the interrupt again, in SVC
875 * We do local_irq_enable() before calling guest_exit() so
876 * that if a timer interrupt hits while running the guest we
877 * account that tick as being spent in the guest. We enable
878 * preemption after calling guest_exit() so that if we get
879 * preempted we make sure ticks after that is not counted as
883 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
885 /* Exit types that need handling before we can be preempted */
886 handle_exit_early(vcpu, ret);
891 * The ARMv8 architecture doesn't give the hypervisor
892 * a mechanism to prevent a guest from dropping to AArch32 EL0
893 * if implemented by the CPU. If we spot the guest in such
894 * state and that we decided it wasn't supposed to do so (like
895 * with the asymmetric AArch32 case), return to userspace with
898 if (vcpu_mode_is_bad_32bit(vcpu)) {
900 * As we have caught the guest red-handed, decide that
901 * it isn't fit for purpose anymore by making the vcpu
902 * invalid. The VMM can try and fix it by issuing a
903 * KVM_ARM_VCPU_INIT if it really wants to.
905 vcpu->arch.target = -1;
906 ret = ARM_EXCEPTION_IL;
909 ret = handle_exit(vcpu, ret);
912 /* Tell userspace about in-kernel device output levels */
913 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
914 kvm_timer_update_run(vcpu);
915 kvm_pmu_update_run(vcpu);
918 kvm_sigset_deactivate(vcpu);
922 * In the unlikely event that we are returning to userspace
923 * with pending exceptions or PC adjustment, commit these
924 * adjustments in order to give userspace a consistent view of
925 * the vcpu state. Note that this relies on __kvm_adjust_pc()
926 * being preempt-safe on VHE.
928 if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
929 KVM_ARM64_INCREMENT_PC)))
930 kvm_call_hyp(__kvm_adjust_pc, vcpu);
936 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
942 if (number == KVM_ARM_IRQ_CPU_IRQ)
943 bit_index = __ffs(HCR_VI);
944 else /* KVM_ARM_IRQ_CPU_FIQ */
945 bit_index = __ffs(HCR_VF);
947 hcr = vcpu_hcr(vcpu);
949 set = test_and_set_bit(bit_index, hcr);
951 set = test_and_clear_bit(bit_index, hcr);
954 * If we didn't change anything, no need to wake up or kick other CPUs
960 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
961 * trigger a world-switch round on the running physical CPU to set the
962 * virtual IRQ/FIQ fields in the HCR appropriately.
964 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
970 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
973 u32 irq = irq_level->irq;
974 unsigned int irq_type, vcpu_idx, irq_num;
975 int nrcpus = atomic_read(&kvm->online_vcpus);
976 struct kvm_vcpu *vcpu = NULL;
977 bool level = irq_level->level;
979 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
980 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
981 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
982 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
984 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
987 case KVM_ARM_IRQ_TYPE_CPU:
988 if (irqchip_in_kernel(kvm))
991 if (vcpu_idx >= nrcpus)
994 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
998 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1001 return vcpu_interrupt_line(vcpu, irq_num, level);
1002 case KVM_ARM_IRQ_TYPE_PPI:
1003 if (!irqchip_in_kernel(kvm))
1006 if (vcpu_idx >= nrcpus)
1009 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1013 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1016 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1017 case KVM_ARM_IRQ_TYPE_SPI:
1018 if (!irqchip_in_kernel(kvm))
1021 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1024 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1030 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1031 const struct kvm_vcpu_init *init)
1033 unsigned int i, ret;
1034 u32 phys_target = kvm_target_cpu();
1036 if (init->target != phys_target)
1040 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1041 * use the same target.
1043 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1046 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1047 for (i = 0; i < sizeof(init->features) * 8; i++) {
1048 bool set = (init->features[i / 32] & (1 << (i % 32)));
1050 if (set && i >= KVM_VCPU_MAX_FEATURES)
1054 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1055 * use the same feature set.
1057 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1058 test_bit(i, vcpu->arch.features) != set)
1062 set_bit(i, vcpu->arch.features);
1065 vcpu->arch.target = phys_target;
1067 /* Now we know what it is, we can reset it. */
1068 ret = kvm_reset_vcpu(vcpu);
1070 vcpu->arch.target = -1;
1071 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1077 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1078 struct kvm_vcpu_init *init)
1082 ret = kvm_vcpu_set_target(vcpu, init);
1087 * Ensure a rebooted VM will fault in RAM pages and detect if the
1088 * guest MMU is turned off and flush the caches as needed.
1090 * S2FWB enforces all memory accesses to RAM being cacheable,
1091 * ensuring that the data side is always coherent. We still
1092 * need to invalidate the I-cache though, as FWB does *not*
1093 * imply CTR_EL0.DIC.
1095 if (vcpu->arch.has_run_once) {
1096 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1097 stage2_unmap_vm(vcpu->kvm);
1099 icache_inval_all_pou();
1102 vcpu_reset_hcr(vcpu);
1105 * Handle the "start in power-off" case.
1107 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1108 vcpu_power_off(vcpu);
1110 vcpu->arch.power_off = false;
1115 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1116 struct kvm_device_attr *attr)
1120 switch (attr->group) {
1122 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1129 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1130 struct kvm_device_attr *attr)
1134 switch (attr->group) {
1136 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1143 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1144 struct kvm_device_attr *attr)
1148 switch (attr->group) {
1150 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1157 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1158 struct kvm_vcpu_events *events)
1160 memset(events, 0, sizeof(*events));
1162 return __kvm_arm_vcpu_get_events(vcpu, events);
1165 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1166 struct kvm_vcpu_events *events)
1170 /* check whether the reserved field is zero */
1171 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1172 if (events->reserved[i])
1175 /* check whether the pad field is zero */
1176 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1177 if (events->exception.pad[i])
1180 return __kvm_arm_vcpu_set_events(vcpu, events);
1183 long kvm_arch_vcpu_ioctl(struct file *filp,
1184 unsigned int ioctl, unsigned long arg)
1186 struct kvm_vcpu *vcpu = filp->private_data;
1187 void __user *argp = (void __user *)arg;
1188 struct kvm_device_attr attr;
1192 case KVM_ARM_VCPU_INIT: {
1193 struct kvm_vcpu_init init;
1196 if (copy_from_user(&init, argp, sizeof(init)))
1199 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1202 case KVM_SET_ONE_REG:
1203 case KVM_GET_ONE_REG: {
1204 struct kvm_one_reg reg;
1207 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1211 if (copy_from_user(®, argp, sizeof(reg)))
1214 if (ioctl == KVM_SET_ONE_REG)
1215 r = kvm_arm_set_reg(vcpu, ®);
1217 r = kvm_arm_get_reg(vcpu, ®);
1220 case KVM_GET_REG_LIST: {
1221 struct kvm_reg_list __user *user_list = argp;
1222 struct kvm_reg_list reg_list;
1226 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1230 if (!kvm_arm_vcpu_is_finalized(vcpu))
1234 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1237 reg_list.n = kvm_arm_num_regs(vcpu);
1238 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1243 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1246 case KVM_SET_DEVICE_ATTR: {
1248 if (copy_from_user(&attr, argp, sizeof(attr)))
1250 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1253 case KVM_GET_DEVICE_ATTR: {
1255 if (copy_from_user(&attr, argp, sizeof(attr)))
1257 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1260 case KVM_HAS_DEVICE_ATTR: {
1262 if (copy_from_user(&attr, argp, sizeof(attr)))
1264 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1267 case KVM_GET_VCPU_EVENTS: {
1268 struct kvm_vcpu_events events;
1270 if (kvm_arm_vcpu_get_events(vcpu, &events))
1273 if (copy_to_user(argp, &events, sizeof(events)))
1278 case KVM_SET_VCPU_EVENTS: {
1279 struct kvm_vcpu_events events;
1281 if (copy_from_user(&events, argp, sizeof(events)))
1284 return kvm_arm_vcpu_set_events(vcpu, &events);
1286 case KVM_ARM_VCPU_FINALIZE: {
1289 if (!kvm_vcpu_initialized(vcpu))
1292 if (get_user(what, (const int __user *)argp))
1295 return kvm_arm_vcpu_finalize(vcpu, what);
1304 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1309 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1310 const struct kvm_memory_slot *memslot)
1312 kvm_flush_remote_tlbs(kvm);
1315 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1316 struct kvm_arm_device_addr *dev_addr)
1318 unsigned long dev_id, type;
1320 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1321 KVM_ARM_DEVICE_ID_SHIFT;
1322 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1323 KVM_ARM_DEVICE_TYPE_SHIFT;
1326 case KVM_ARM_DEVICE_VGIC_V2:
1329 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1335 long kvm_arch_vm_ioctl(struct file *filp,
1336 unsigned int ioctl, unsigned long arg)
1338 struct kvm *kvm = filp->private_data;
1339 void __user *argp = (void __user *)arg;
1342 case KVM_CREATE_IRQCHIP: {
1346 mutex_lock(&kvm->lock);
1347 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1348 mutex_unlock(&kvm->lock);
1351 case KVM_ARM_SET_DEVICE_ADDR: {
1352 struct kvm_arm_device_addr dev_addr;
1354 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1356 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1358 case KVM_ARM_PREFERRED_TARGET: {
1360 struct kvm_vcpu_init init;
1362 err = kvm_vcpu_preferred_target(&init);
1366 if (copy_to_user(argp, &init, sizeof(init)))
1371 case KVM_ARM_MTE_COPY_TAGS: {
1372 struct kvm_arm_copy_mte_tags copy_tags;
1374 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1376 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1383 static unsigned long nvhe_percpu_size(void)
1385 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1386 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1389 static unsigned long nvhe_percpu_order(void)
1391 unsigned long size = nvhe_percpu_size();
1393 return size ? get_order(size) : 0;
1396 /* A lookup table holding the hypervisor VA for each vector slot */
1397 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1399 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1401 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1404 static int kvm_init_vector_slots(void)
1409 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1410 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1412 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1413 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1415 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1419 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1420 __BP_HARDEN_HYP_VECS_SZ, &base);
1425 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1426 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1430 static void cpu_prepare_hyp_mode(int cpu)
1432 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1436 * Calculate the raw per-cpu offset without a translation from the
1437 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1438 * so that we can use adr_l to access per-cpu variables in EL2.
1439 * Also drop the KASAN tag which gets in the way...
1441 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1442 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1444 params->mair_el2 = read_sysreg(mair_el1);
1447 * The ID map may be configured to use an extended virtual address
1448 * range. This is only the case if system RAM is out of range for the
1449 * currently configured page size and VA_BITS, in which case we will
1450 * also need the extended virtual range for the HYP ID map, or we won't
1451 * be able to enable the EL2 MMU.
1453 * However, at EL2, there is only one TTBR register, and we can't switch
1454 * between translation tables *and* update TCR_EL2.T0SZ at the same
1455 * time. Bottom line: we need to use the extended range with *both* our
1456 * translation tables.
1458 * So use the same T0SZ value we use for the ID map.
1460 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1461 tcr &= ~TCR_T0SZ_MASK;
1462 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1463 params->tcr_el2 = tcr;
1465 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1466 params->pgd_pa = kvm_mmu_get_httbr();
1467 if (is_protected_kvm_enabled())
1468 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1470 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1471 params->vttbr = params->vtcr = 0;
1474 * Flush the init params from the data cache because the struct will
1475 * be read while the MMU is off.
1477 kvm_flush_dcache_to_poc(params, sizeof(*params));
1480 static void hyp_install_host_vector(void)
1482 struct kvm_nvhe_init_params *params;
1483 struct arm_smccc_res res;
1485 /* Switch from the HYP stub to our own HYP init vector */
1486 __hyp_set_vectors(kvm_get_idmap_vector());
1489 * Call initialization code, and switch to the full blown HYP code.
1490 * If the cpucaps haven't been finalized yet, something has gone very
1491 * wrong, and hyp will crash and burn when it uses any
1492 * cpus_have_const_cap() wrapper.
1494 BUG_ON(!system_capabilities_finalized());
1495 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1496 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1497 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1500 static void cpu_init_hyp_mode(void)
1502 hyp_install_host_vector();
1505 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1508 if (this_cpu_has_cap(ARM64_SSBS) &&
1509 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1510 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1514 static void cpu_hyp_reset(void)
1516 if (!is_kernel_in_hyp_mode())
1517 __hyp_reset_vectors();
1521 * EL2 vectors can be mapped and rerouted in a number of ways,
1522 * depending on the kernel configuration and CPU present:
1524 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1525 * placed in one of the vector slots, which is executed before jumping
1526 * to the real vectors.
1528 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1529 * containing the hardening sequence is mapped next to the idmap page,
1530 * and executed before jumping to the real vectors.
1532 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1533 * empty slot is selected, mapped next to the idmap page, and
1534 * executed before jumping to the real vectors.
1536 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1537 * VHE, as we don't have hypervisor-specific mappings. If the system
1538 * is VHE and yet selects this capability, it will be ignored.
1540 static void cpu_set_hyp_vector(void)
1542 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1543 void *vector = hyp_spectre_vector_selector[data->slot];
1545 if (!is_protected_kvm_enabled())
1546 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1548 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1551 static void cpu_hyp_reinit(void)
1553 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1557 if (is_kernel_in_hyp_mode())
1558 kvm_timer_init_vhe();
1560 cpu_init_hyp_mode();
1562 cpu_set_hyp_vector();
1564 kvm_arm_init_debug();
1567 kvm_vgic_init_cpu_hardware();
1570 static void _kvm_arch_hardware_enable(void *discard)
1572 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1574 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1578 int kvm_arch_hardware_enable(void)
1580 _kvm_arch_hardware_enable(NULL);
1584 static void _kvm_arch_hardware_disable(void *discard)
1586 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1588 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1592 void kvm_arch_hardware_disable(void)
1594 if (!is_protected_kvm_enabled())
1595 _kvm_arch_hardware_disable(NULL);
1598 #ifdef CONFIG_CPU_PM
1599 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1604 * kvm_arm_hardware_enabled is left with its old value over
1605 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1610 if (__this_cpu_read(kvm_arm_hardware_enabled))
1612 * don't update kvm_arm_hardware_enabled here
1613 * so that the hardware will be re-enabled
1614 * when we resume. See below.
1619 case CPU_PM_ENTER_FAILED:
1621 if (__this_cpu_read(kvm_arm_hardware_enabled))
1622 /* The hardware was enabled before suspend. */
1632 static struct notifier_block hyp_init_cpu_pm_nb = {
1633 .notifier_call = hyp_init_cpu_pm_notifier,
1636 static void hyp_cpu_pm_init(void)
1638 if (!is_protected_kvm_enabled())
1639 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1641 static void hyp_cpu_pm_exit(void)
1643 if (!is_protected_kvm_enabled())
1644 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1647 static inline void hyp_cpu_pm_init(void)
1650 static inline void hyp_cpu_pm_exit(void)
1655 static void init_cpu_logical_map(void)
1660 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1661 * Only copy the set of online CPUs whose features have been chacked
1662 * against the finalized system capabilities. The hypervisor will not
1663 * allow any other CPUs from the `possible` set to boot.
1665 for_each_online_cpu(cpu)
1666 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1669 #define init_psci_0_1_impl_state(config, what) \
1670 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1672 static bool init_psci_relay(void)
1675 * If PSCI has not been initialized, protected KVM cannot install
1676 * itself on newly booted CPUs.
1678 if (!psci_ops.get_version) {
1679 kvm_err("Cannot initialize protected mode without PSCI\n");
1683 kvm_host_psci_config.version = psci_ops.get_version();
1685 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1686 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1687 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1688 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1689 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1690 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1695 static int init_subsystems(void)
1700 * Enable hardware so that subsystem initialisation can access EL2.
1702 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1705 * Register CPU lower-power notifier
1710 * Init HYP view of VGIC
1712 err = kvm_vgic_hyp_init();
1715 vgic_present = true;
1719 vgic_present = false;
1727 * Init HYP architected timer support
1729 err = kvm_timer_hyp_init(vgic_present);
1734 kvm_sys_reg_table_init();
1737 if (err || !is_protected_kvm_enabled())
1738 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1743 static void teardown_hyp_mode(void)
1748 for_each_possible_cpu(cpu) {
1749 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1750 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1754 static int do_pkvm_init(u32 hyp_va_bits)
1756 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1760 hyp_install_host_vector();
1761 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1762 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1769 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1771 void *addr = phys_to_virt(hyp_mem_base);
1774 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1775 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1777 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1781 ret = do_pkvm_init(hyp_va_bits);
1791 * Inits Hyp-mode on all online CPUs
1793 static int init_hyp_mode(void)
1800 * The protected Hyp-mode cannot be initialized if the memory pool
1801 * allocation has failed.
1803 if (is_protected_kvm_enabled() && !hyp_mem_base)
1807 * Allocate Hyp PGD and setup Hyp identity mapping
1809 err = kvm_mmu_init(&hyp_va_bits);
1814 * Allocate stack pages for Hypervisor-mode
1816 for_each_possible_cpu(cpu) {
1817 unsigned long stack_page;
1819 stack_page = __get_free_page(GFP_KERNEL);
1825 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1829 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1831 for_each_possible_cpu(cpu) {
1835 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1841 page_addr = page_address(page);
1842 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1843 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1847 * Map the Hyp-code called directly from the host
1849 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1850 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1852 kvm_err("Cannot map world-switch code\n");
1856 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1857 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1859 kvm_err("Cannot map .hyp.rodata section\n");
1863 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1864 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1866 kvm_err("Cannot map rodata section\n");
1871 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1872 * section thanks to an assertion in the linker script. Map it RW and
1873 * the rest of .bss RO.
1875 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1876 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1878 kvm_err("Cannot map hyp bss section: %d\n", err);
1882 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1883 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1885 kvm_err("Cannot map bss section\n");
1890 * Map the Hyp stack pages
1892 for_each_possible_cpu(cpu) {
1893 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1894 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1898 kvm_err("Cannot map hyp stack\n");
1903 for_each_possible_cpu(cpu) {
1904 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1905 char *percpu_end = percpu_begin + nvhe_percpu_size();
1907 /* Map Hyp percpu pages */
1908 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1910 kvm_err("Cannot map hyp percpu region\n");
1914 /* Prepare the CPU initialization parameters */
1915 cpu_prepare_hyp_mode(cpu);
1918 if (is_protected_kvm_enabled()) {
1919 init_cpu_logical_map();
1921 if (!init_psci_relay()) {
1927 if (is_protected_kvm_enabled()) {
1928 err = kvm_hyp_init_protection(hyp_va_bits);
1930 kvm_err("Failed to init hyp memory protection\n");
1938 teardown_hyp_mode();
1939 kvm_err("error initializing Hyp mode: %d\n", err);
1943 static void _kvm_host_prot_finalize(void *discard)
1945 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1948 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1950 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1953 #define pkvm_mark_hyp_section(__section) \
1954 pkvm_mark_hyp(__pa_symbol(__section##_start), \
1955 __pa_symbol(__section##_end))
1957 static int finalize_hyp_mode(void)
1961 if (!is_protected_kvm_enabled())
1964 ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1968 ret = pkvm_mark_hyp_section(__hyp_text);
1972 ret = pkvm_mark_hyp_section(__hyp_rodata);
1976 ret = pkvm_mark_hyp_section(__hyp_bss);
1980 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1984 for_each_possible_cpu(cpu) {
1985 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1986 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
1988 ret = pkvm_mark_hyp(start, end);
1992 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
1993 end = start + PAGE_SIZE;
1994 ret = pkvm_mark_hyp(start, end);
2000 * Flip the static key upfront as that may no longer be possible
2001 * once the host stage 2 is installed.
2003 static_branch_enable(&kvm_protected_mode_initialized);
2004 on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
2009 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2011 struct kvm_vcpu *vcpu;
2014 mpidr &= MPIDR_HWID_BITMASK;
2015 kvm_for_each_vcpu(i, vcpu, kvm) {
2016 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2022 bool kvm_arch_has_irq_bypass(void)
2027 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2028 struct irq_bypass_producer *prod)
2030 struct kvm_kernel_irqfd *irqfd =
2031 container_of(cons, struct kvm_kernel_irqfd, consumer);
2033 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2036 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2037 struct irq_bypass_producer *prod)
2039 struct kvm_kernel_irqfd *irqfd =
2040 container_of(cons, struct kvm_kernel_irqfd, consumer);
2042 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2046 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2048 struct kvm_kernel_irqfd *irqfd =
2049 container_of(cons, struct kvm_kernel_irqfd, consumer);
2051 kvm_arm_halt_guest(irqfd->kvm);
2054 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2056 struct kvm_kernel_irqfd *irqfd =
2057 container_of(cons, struct kvm_kernel_irqfd, consumer);
2059 kvm_arm_resume_guest(irqfd->kvm);
2063 * Initialize Hyp-mode and memory mappings on all CPUs.
2065 int kvm_arch_init(void *opaque)
2070 if (!is_hyp_mode_available()) {
2071 kvm_info("HYP mode not available\n");
2075 in_hyp_mode = is_kernel_in_hyp_mode();
2077 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2078 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2079 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2080 "Only trusted guests should be used on this system.\n");
2082 err = kvm_set_ipa_limit();
2086 err = kvm_arm_init_sve();
2091 err = init_hyp_mode();
2096 err = kvm_init_vector_slots();
2098 kvm_err("Cannot initialise vector slots\n");
2102 err = init_subsystems();
2107 err = finalize_hyp_mode();
2109 kvm_err("Failed to finalize Hyp protection\n");
2114 if (is_protected_kvm_enabled()) {
2115 kvm_info("Protected nVHE mode initialized successfully\n");
2116 } else if (in_hyp_mode) {
2117 kvm_info("VHE mode initialized successfully\n");
2119 kvm_info("Hyp mode initialized successfully\n");
2127 teardown_hyp_mode();
2132 /* NOP: Compiling as a module not supported */
2133 void kvm_arch_exit(void)
2135 kvm_perf_teardown();
2138 static int __init early_kvm_mode_cfg(char *arg)
2143 if (strcmp(arg, "protected") == 0) {
2144 kvm_mode = KVM_MODE_PROTECTED;
2148 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2153 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2155 enum kvm_mode kvm_get_mode(void)
2160 static int arm_init(void)
2162 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2166 module_init(arm_init);