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);
725 if (run->immediate_exit) {
730 kvm_sigset_activate(vcpu);
733 run->exit_reason = KVM_EXIT_UNKNOWN;
736 * Check conditions before entering the guest
740 update_vmid(&vcpu->arch.hw_mmu->vmid);
742 check_vcpu_requests(vcpu);
745 * Preparing the interrupts to be injected also
746 * involves poking the GIC, which must be done in a
747 * non-preemptible context.
751 kvm_pmu_flush_hwstate(vcpu);
755 kvm_vgic_flush_hwstate(vcpu);
758 * Exit if we have a signal pending so that we can deliver the
759 * signal to user space.
761 if (signal_pending(current)) {
763 run->exit_reason = KVM_EXIT_INTR;
767 * If we're using a userspace irqchip, then check if we need
768 * to tell a userspace irqchip about timer or PMU level
769 * changes and if so, exit to userspace (the actual level
770 * state gets updated in kvm_timer_update_run and
771 * kvm_pmu_update_run below).
773 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
774 if (kvm_timer_should_notify_user(vcpu) ||
775 kvm_pmu_should_notify_user(vcpu)) {
777 run->exit_reason = KVM_EXIT_INTR;
782 * Ensure we set mode to IN_GUEST_MODE after we disable
783 * interrupts and before the final VCPU requests check.
784 * See the comment in kvm_vcpu_exiting_guest_mode() and
785 * Documentation/virt/kvm/vcpu-requests.rst
787 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
789 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
790 kvm_request_pending(vcpu)) {
791 vcpu->mode = OUTSIDE_GUEST_MODE;
792 isb(); /* Ensure work in x_flush_hwstate is committed */
793 kvm_pmu_sync_hwstate(vcpu);
794 if (static_branch_unlikely(&userspace_irqchip_in_use))
795 kvm_timer_sync_user(vcpu);
796 kvm_vgic_sync_hwstate(vcpu);
802 kvm_arm_setup_debug(vcpu);
804 /**************************************************************
807 trace_kvm_entry(*vcpu_pc(vcpu));
808 guest_enter_irqoff();
810 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
812 vcpu->mode = OUTSIDE_GUEST_MODE;
816 *************************************************************/
818 kvm_arm_clear_debug(vcpu);
821 * We must sync the PMU state before the vgic state so
822 * that the vgic can properly sample the updated state of the
825 kvm_pmu_sync_hwstate(vcpu);
828 * Sync the vgic state before syncing the timer state because
829 * the timer code needs to know if the virtual timer
830 * interrupts are active.
832 kvm_vgic_sync_hwstate(vcpu);
835 * Sync the timer hardware state before enabling interrupts as
836 * we don't want vtimer interrupts to race with syncing the
837 * timer virtual interrupt state.
839 if (static_branch_unlikely(&userspace_irqchip_in_use))
840 kvm_timer_sync_user(vcpu);
842 kvm_arch_vcpu_ctxsync_fp(vcpu);
845 * We may have taken a host interrupt in HYP mode (ie
846 * while executing the guest). This interrupt is still
847 * pending, as we haven't serviced it yet!
849 * We're now back in SVC mode, with interrupts
850 * disabled. Enabling the interrupts now will have
851 * the effect of taking the interrupt again, in SVC
857 * We do local_irq_enable() before calling guest_exit() so
858 * that if a timer interrupt hits while running the guest we
859 * account that tick as being spent in the guest. We enable
860 * preemption after calling guest_exit() so that if we get
861 * preempted we make sure ticks after that is not counted as
865 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
867 /* Exit types that need handling before we can be preempted */
868 handle_exit_early(vcpu, ret);
873 * The ARMv8 architecture doesn't give the hypervisor
874 * a mechanism to prevent a guest from dropping to AArch32 EL0
875 * if implemented by the CPU. If we spot the guest in such
876 * state and that we decided it wasn't supposed to do so (like
877 * with the asymmetric AArch32 case), return to userspace with
880 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
882 * As we have caught the guest red-handed, decide that
883 * it isn't fit for purpose anymore by making the vcpu
884 * invalid. The VMM can try and fix it by issuing a
885 * KVM_ARM_VCPU_INIT if it really wants to.
887 vcpu->arch.target = -1;
888 ret = ARM_EXCEPTION_IL;
891 ret = handle_exit(vcpu, ret);
894 /* Tell userspace about in-kernel device output levels */
895 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
896 kvm_timer_update_run(vcpu);
897 kvm_pmu_update_run(vcpu);
900 kvm_sigset_deactivate(vcpu);
904 * In the unlikely event that we are returning to userspace
905 * with pending exceptions or PC adjustment, commit these
906 * adjustments in order to give userspace a consistent view of
907 * the vcpu state. Note that this relies on __kvm_adjust_pc()
908 * being preempt-safe on VHE.
910 if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
911 KVM_ARM64_INCREMENT_PC)))
912 kvm_call_hyp(__kvm_adjust_pc, vcpu);
918 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
924 if (number == KVM_ARM_IRQ_CPU_IRQ)
925 bit_index = __ffs(HCR_VI);
926 else /* KVM_ARM_IRQ_CPU_FIQ */
927 bit_index = __ffs(HCR_VF);
929 hcr = vcpu_hcr(vcpu);
931 set = test_and_set_bit(bit_index, hcr);
933 set = test_and_clear_bit(bit_index, hcr);
936 * If we didn't change anything, no need to wake up or kick other CPUs
942 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
943 * trigger a world-switch round on the running physical CPU to set the
944 * virtual IRQ/FIQ fields in the HCR appropriately.
946 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
952 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
955 u32 irq = irq_level->irq;
956 unsigned int irq_type, vcpu_idx, irq_num;
957 int nrcpus = atomic_read(&kvm->online_vcpus);
958 struct kvm_vcpu *vcpu = NULL;
959 bool level = irq_level->level;
961 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
962 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
963 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
964 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
966 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
969 case KVM_ARM_IRQ_TYPE_CPU:
970 if (irqchip_in_kernel(kvm))
973 if (vcpu_idx >= nrcpus)
976 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
980 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
983 return vcpu_interrupt_line(vcpu, irq_num, level);
984 case KVM_ARM_IRQ_TYPE_PPI:
985 if (!irqchip_in_kernel(kvm))
988 if (vcpu_idx >= nrcpus)
991 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
995 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
998 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
999 case KVM_ARM_IRQ_TYPE_SPI:
1000 if (!irqchip_in_kernel(kvm))
1003 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1006 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1012 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1013 const struct kvm_vcpu_init *init)
1015 unsigned int i, ret;
1016 int phys_target = kvm_target_cpu();
1018 if (init->target != phys_target)
1022 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1023 * use the same target.
1025 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1028 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1029 for (i = 0; i < sizeof(init->features) * 8; i++) {
1030 bool set = (init->features[i / 32] & (1 << (i % 32)));
1032 if (set && i >= KVM_VCPU_MAX_FEATURES)
1036 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1037 * use the same feature set.
1039 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1040 test_bit(i, vcpu->arch.features) != set)
1044 set_bit(i, vcpu->arch.features);
1047 vcpu->arch.target = phys_target;
1049 /* Now we know what it is, we can reset it. */
1050 ret = kvm_reset_vcpu(vcpu);
1052 vcpu->arch.target = -1;
1053 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1059 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1060 struct kvm_vcpu_init *init)
1064 ret = kvm_vcpu_set_target(vcpu, init);
1069 * Ensure a rebooted VM will fault in RAM pages and detect if the
1070 * guest MMU is turned off and flush the caches as needed.
1072 * S2FWB enforces all memory accesses to RAM being cacheable,
1073 * ensuring that the data side is always coherent. We still
1074 * need to invalidate the I-cache though, as FWB does *not*
1075 * imply CTR_EL0.DIC.
1077 if (vcpu->arch.has_run_once) {
1078 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1079 stage2_unmap_vm(vcpu->kvm);
1081 __flush_icache_all();
1084 vcpu_reset_hcr(vcpu);
1087 * Handle the "start in power-off" case.
1089 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1090 vcpu_power_off(vcpu);
1092 vcpu->arch.power_off = false;
1097 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1098 struct kvm_device_attr *attr)
1102 switch (attr->group) {
1104 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1111 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1112 struct kvm_device_attr *attr)
1116 switch (attr->group) {
1118 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1125 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1126 struct kvm_device_attr *attr)
1130 switch (attr->group) {
1132 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1139 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1140 struct kvm_vcpu_events *events)
1142 memset(events, 0, sizeof(*events));
1144 return __kvm_arm_vcpu_get_events(vcpu, events);
1147 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1148 struct kvm_vcpu_events *events)
1152 /* check whether the reserved field is zero */
1153 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1154 if (events->reserved[i])
1157 /* check whether the pad field is zero */
1158 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1159 if (events->exception.pad[i])
1162 return __kvm_arm_vcpu_set_events(vcpu, events);
1165 long kvm_arch_vcpu_ioctl(struct file *filp,
1166 unsigned int ioctl, unsigned long arg)
1168 struct kvm_vcpu *vcpu = filp->private_data;
1169 void __user *argp = (void __user *)arg;
1170 struct kvm_device_attr attr;
1174 case KVM_ARM_VCPU_INIT: {
1175 struct kvm_vcpu_init init;
1178 if (copy_from_user(&init, argp, sizeof(init)))
1181 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1184 case KVM_SET_ONE_REG:
1185 case KVM_GET_ONE_REG: {
1186 struct kvm_one_reg reg;
1189 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1193 if (copy_from_user(®, argp, sizeof(reg)))
1196 if (ioctl == KVM_SET_ONE_REG)
1197 r = kvm_arm_set_reg(vcpu, ®);
1199 r = kvm_arm_get_reg(vcpu, ®);
1202 case KVM_GET_REG_LIST: {
1203 struct kvm_reg_list __user *user_list = argp;
1204 struct kvm_reg_list reg_list;
1208 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1212 if (!kvm_arm_vcpu_is_finalized(vcpu))
1216 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1219 reg_list.n = kvm_arm_num_regs(vcpu);
1220 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1225 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1228 case KVM_SET_DEVICE_ATTR: {
1230 if (copy_from_user(&attr, argp, sizeof(attr)))
1232 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1235 case KVM_GET_DEVICE_ATTR: {
1237 if (copy_from_user(&attr, argp, sizeof(attr)))
1239 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1242 case KVM_HAS_DEVICE_ATTR: {
1244 if (copy_from_user(&attr, argp, sizeof(attr)))
1246 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1249 case KVM_GET_VCPU_EVENTS: {
1250 struct kvm_vcpu_events events;
1252 if (kvm_arm_vcpu_get_events(vcpu, &events))
1255 if (copy_to_user(argp, &events, sizeof(events)))
1260 case KVM_SET_VCPU_EVENTS: {
1261 struct kvm_vcpu_events events;
1263 if (copy_from_user(&events, argp, sizeof(events)))
1266 return kvm_arm_vcpu_set_events(vcpu, &events);
1268 case KVM_ARM_VCPU_FINALIZE: {
1271 if (!kvm_vcpu_initialized(vcpu))
1274 if (get_user(what, (const int __user *)argp))
1277 return kvm_arm_vcpu_finalize(vcpu, what);
1286 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1291 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1292 const struct kvm_memory_slot *memslot)
1294 kvm_flush_remote_tlbs(kvm);
1297 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1298 struct kvm_arm_device_addr *dev_addr)
1300 unsigned long dev_id, type;
1302 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1303 KVM_ARM_DEVICE_ID_SHIFT;
1304 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1305 KVM_ARM_DEVICE_TYPE_SHIFT;
1308 case KVM_ARM_DEVICE_VGIC_V2:
1311 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1317 long kvm_arch_vm_ioctl(struct file *filp,
1318 unsigned int ioctl, unsigned long arg)
1320 struct kvm *kvm = filp->private_data;
1321 void __user *argp = (void __user *)arg;
1324 case KVM_CREATE_IRQCHIP: {
1328 mutex_lock(&kvm->lock);
1329 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1330 mutex_unlock(&kvm->lock);
1333 case KVM_ARM_SET_DEVICE_ADDR: {
1334 struct kvm_arm_device_addr dev_addr;
1336 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1338 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1340 case KVM_ARM_PREFERRED_TARGET: {
1342 struct kvm_vcpu_init init;
1344 err = kvm_vcpu_preferred_target(&init);
1348 if (copy_to_user(argp, &init, sizeof(init)))
1358 static unsigned long nvhe_percpu_size(void)
1360 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1361 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1364 static unsigned long nvhe_percpu_order(void)
1366 unsigned long size = nvhe_percpu_size();
1368 return size ? get_order(size) : 0;
1371 /* A lookup table holding the hypervisor VA for each vector slot */
1372 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1374 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1376 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1379 static int kvm_init_vector_slots(void)
1384 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1385 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1387 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1388 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1390 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1394 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1395 __BP_HARDEN_HYP_VECS_SZ, &base);
1400 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1401 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1405 static void cpu_prepare_hyp_mode(int cpu)
1407 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1411 * Calculate the raw per-cpu offset without a translation from the
1412 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1413 * so that we can use adr_l to access per-cpu variables in EL2.
1414 * Also drop the KASAN tag which gets in the way...
1416 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1417 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1419 params->mair_el2 = read_sysreg(mair_el1);
1422 * The ID map may be configured to use an extended virtual address
1423 * range. This is only the case if system RAM is out of range for the
1424 * currently configured page size and VA_BITS, in which case we will
1425 * also need the extended virtual range for the HYP ID map, or we won't
1426 * be able to enable the EL2 MMU.
1428 * However, at EL2, there is only one TTBR register, and we can't switch
1429 * between translation tables *and* update TCR_EL2.T0SZ at the same
1430 * time. Bottom line: we need to use the extended range with *both* our
1431 * translation tables.
1433 * So use the same T0SZ value we use for the ID map.
1435 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1436 tcr &= ~TCR_T0SZ_MASK;
1437 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1438 params->tcr_el2 = tcr;
1440 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1441 params->pgd_pa = kvm_mmu_get_httbr();
1442 if (is_protected_kvm_enabled())
1443 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1445 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1446 params->vttbr = params->vtcr = 0;
1449 * Flush the init params from the data cache because the struct will
1450 * be read while the MMU is off.
1452 kvm_flush_dcache_to_poc(params, sizeof(*params));
1455 static void hyp_install_host_vector(void)
1457 struct kvm_nvhe_init_params *params;
1458 struct arm_smccc_res res;
1460 /* Switch from the HYP stub to our own HYP init vector */
1461 __hyp_set_vectors(kvm_get_idmap_vector());
1464 * Call initialization code, and switch to the full blown HYP code.
1465 * If the cpucaps haven't been finalized yet, something has gone very
1466 * wrong, and hyp will crash and burn when it uses any
1467 * cpus_have_const_cap() wrapper.
1469 BUG_ON(!system_capabilities_finalized());
1470 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1471 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1472 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1475 static void cpu_init_hyp_mode(void)
1477 hyp_install_host_vector();
1480 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1483 if (this_cpu_has_cap(ARM64_SSBS) &&
1484 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1485 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1489 static void cpu_hyp_reset(void)
1491 if (!is_kernel_in_hyp_mode())
1492 __hyp_reset_vectors();
1496 * EL2 vectors can be mapped and rerouted in a number of ways,
1497 * depending on the kernel configuration and CPU present:
1499 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1500 * placed in one of the vector slots, which is executed before jumping
1501 * to the real vectors.
1503 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1504 * containing the hardening sequence is mapped next to the idmap page,
1505 * and executed before jumping to the real vectors.
1507 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1508 * empty slot is selected, mapped next to the idmap page, and
1509 * executed before jumping to the real vectors.
1511 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1512 * VHE, as we don't have hypervisor-specific mappings. If the system
1513 * is VHE and yet selects this capability, it will be ignored.
1515 static void cpu_set_hyp_vector(void)
1517 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1518 void *vector = hyp_spectre_vector_selector[data->slot];
1520 if (!is_protected_kvm_enabled())
1521 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1523 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1526 static void cpu_hyp_reinit(void)
1528 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1532 if (is_kernel_in_hyp_mode())
1533 kvm_timer_init_vhe();
1535 cpu_init_hyp_mode();
1537 cpu_set_hyp_vector();
1539 kvm_arm_init_debug();
1542 kvm_vgic_init_cpu_hardware();
1545 static void _kvm_arch_hardware_enable(void *discard)
1547 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1549 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1553 int kvm_arch_hardware_enable(void)
1555 _kvm_arch_hardware_enable(NULL);
1559 static void _kvm_arch_hardware_disable(void *discard)
1561 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1563 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1567 void kvm_arch_hardware_disable(void)
1569 if (!is_protected_kvm_enabled())
1570 _kvm_arch_hardware_disable(NULL);
1573 #ifdef CONFIG_CPU_PM
1574 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1579 * kvm_arm_hardware_enabled is left with its old value over
1580 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1585 if (__this_cpu_read(kvm_arm_hardware_enabled))
1587 * don't update kvm_arm_hardware_enabled here
1588 * so that the hardware will be re-enabled
1589 * when we resume. See below.
1594 case CPU_PM_ENTER_FAILED:
1596 if (__this_cpu_read(kvm_arm_hardware_enabled))
1597 /* The hardware was enabled before suspend. */
1607 static struct notifier_block hyp_init_cpu_pm_nb = {
1608 .notifier_call = hyp_init_cpu_pm_notifier,
1611 static void hyp_cpu_pm_init(void)
1613 if (!is_protected_kvm_enabled())
1614 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1616 static void hyp_cpu_pm_exit(void)
1618 if (!is_protected_kvm_enabled())
1619 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1622 static inline void hyp_cpu_pm_init(void)
1625 static inline void hyp_cpu_pm_exit(void)
1630 static void init_cpu_logical_map(void)
1635 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1636 * Only copy the set of online CPUs whose features have been chacked
1637 * against the finalized system capabilities. The hypervisor will not
1638 * allow any other CPUs from the `possible` set to boot.
1640 for_each_online_cpu(cpu)
1641 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1644 #define init_psci_0_1_impl_state(config, what) \
1645 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1647 static bool init_psci_relay(void)
1650 * If PSCI has not been initialized, protected KVM cannot install
1651 * itself on newly booted CPUs.
1653 if (!psci_ops.get_version) {
1654 kvm_err("Cannot initialize protected mode without PSCI\n");
1658 kvm_host_psci_config.version = psci_ops.get_version();
1660 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1661 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1662 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1663 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1664 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1665 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1670 static int init_common_resources(void)
1672 return kvm_set_ipa_limit();
1675 static int init_subsystems(void)
1680 * Enable hardware so that subsystem initialisation can access EL2.
1682 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1685 * Register CPU lower-power notifier
1690 * Init HYP view of VGIC
1692 err = kvm_vgic_hyp_init();
1695 vgic_present = true;
1699 vgic_present = false;
1707 * Init HYP architected timer support
1709 err = kvm_timer_hyp_init(vgic_present);
1714 kvm_sys_reg_table_init();
1717 if (err || !is_protected_kvm_enabled())
1718 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1723 static void teardown_hyp_mode(void)
1728 for_each_possible_cpu(cpu) {
1729 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1730 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1734 static int do_pkvm_init(u32 hyp_va_bits)
1736 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1740 hyp_install_host_vector();
1741 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1742 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1749 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1751 void *addr = phys_to_virt(hyp_mem_base);
1754 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1755 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1757 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1761 ret = do_pkvm_init(hyp_va_bits);
1771 * Inits Hyp-mode on all online CPUs
1773 static int init_hyp_mode(void)
1780 * The protected Hyp-mode cannot be initialized if the memory pool
1781 * allocation has failed.
1783 if (is_protected_kvm_enabled() && !hyp_mem_base)
1787 * Allocate Hyp PGD and setup Hyp identity mapping
1789 err = kvm_mmu_init(&hyp_va_bits);
1794 * Allocate stack pages for Hypervisor-mode
1796 for_each_possible_cpu(cpu) {
1797 unsigned long stack_page;
1799 stack_page = __get_free_page(GFP_KERNEL);
1805 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1809 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1811 for_each_possible_cpu(cpu) {
1815 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1821 page_addr = page_address(page);
1822 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1823 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1827 * Map the Hyp-code called directly from the host
1829 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1830 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1832 kvm_err("Cannot map world-switch code\n");
1836 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1837 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1839 kvm_err("Cannot map .hyp.rodata section\n");
1843 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1844 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1846 kvm_err("Cannot map rodata section\n");
1851 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1852 * section thanks to an assertion in the linker script. Map it RW and
1853 * the rest of .bss RO.
1855 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1856 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1858 kvm_err("Cannot map hyp bss section: %d\n", err);
1862 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1863 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1865 kvm_err("Cannot map bss section\n");
1870 * Map the Hyp stack pages
1872 for_each_possible_cpu(cpu) {
1873 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1874 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1878 kvm_err("Cannot map hyp stack\n");
1883 for_each_possible_cpu(cpu) {
1884 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1885 char *percpu_end = percpu_begin + nvhe_percpu_size();
1887 /* Map Hyp percpu pages */
1888 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1890 kvm_err("Cannot map hyp percpu region\n");
1894 /* Prepare the CPU initialization parameters */
1895 cpu_prepare_hyp_mode(cpu);
1898 if (is_protected_kvm_enabled()) {
1899 init_cpu_logical_map();
1901 if (!init_psci_relay()) {
1907 if (is_protected_kvm_enabled()) {
1908 err = kvm_hyp_init_protection(hyp_va_bits);
1910 kvm_err("Failed to init hyp memory protection\n");
1918 teardown_hyp_mode();
1919 kvm_err("error initializing Hyp mode: %d\n", err);
1923 static void _kvm_host_prot_finalize(void *discard)
1925 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1928 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1930 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1933 #define pkvm_mark_hyp_section(__section) \
1934 pkvm_mark_hyp(__pa_symbol(__section##_start), \
1935 __pa_symbol(__section##_end))
1937 static int finalize_hyp_mode(void)
1941 if (!is_protected_kvm_enabled())
1944 ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1948 ret = pkvm_mark_hyp_section(__hyp_text);
1952 ret = pkvm_mark_hyp_section(__hyp_rodata);
1956 ret = pkvm_mark_hyp_section(__hyp_bss);
1960 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1964 for_each_possible_cpu(cpu) {
1965 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1966 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
1968 ret = pkvm_mark_hyp(start, end);
1972 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
1973 end = start + PAGE_SIZE;
1974 ret = pkvm_mark_hyp(start, end);
1980 * Flip the static key upfront as that may no longer be possible
1981 * once the host stage 2 is installed.
1983 static_branch_enable(&kvm_protected_mode_initialized);
1984 on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
1989 static void check_kvm_target_cpu(void *ret)
1991 *(int *)ret = kvm_target_cpu();
1994 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1996 struct kvm_vcpu *vcpu;
1999 mpidr &= MPIDR_HWID_BITMASK;
2000 kvm_for_each_vcpu(i, vcpu, kvm) {
2001 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2007 bool kvm_arch_has_irq_bypass(void)
2012 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2013 struct irq_bypass_producer *prod)
2015 struct kvm_kernel_irqfd *irqfd =
2016 container_of(cons, struct kvm_kernel_irqfd, consumer);
2018 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2021 void kvm_arch_irq_bypass_del_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 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2031 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2033 struct kvm_kernel_irqfd *irqfd =
2034 container_of(cons, struct kvm_kernel_irqfd, consumer);
2036 kvm_arm_halt_guest(irqfd->kvm);
2039 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2041 struct kvm_kernel_irqfd *irqfd =
2042 container_of(cons, struct kvm_kernel_irqfd, consumer);
2044 kvm_arm_resume_guest(irqfd->kvm);
2048 * Initialize Hyp-mode and memory mappings on all CPUs.
2050 int kvm_arch_init(void *opaque)
2056 if (!is_hyp_mode_available()) {
2057 kvm_info("HYP mode not available\n");
2061 in_hyp_mode = is_kernel_in_hyp_mode();
2063 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2064 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2065 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2066 "Only trusted guests should be used on this system.\n");
2068 for_each_online_cpu(cpu) {
2069 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
2071 kvm_err("Error, CPU %d not supported!\n", cpu);
2076 err = init_common_resources();
2080 err = kvm_arm_init_sve();
2085 err = init_hyp_mode();
2090 err = kvm_init_vector_slots();
2092 kvm_err("Cannot initialise vector slots\n");
2096 err = init_subsystems();
2101 err = finalize_hyp_mode();
2103 kvm_err("Failed to finalize Hyp protection\n");
2108 if (is_protected_kvm_enabled()) {
2109 kvm_info("Protected nVHE mode initialized successfully\n");
2110 } else if (in_hyp_mode) {
2111 kvm_info("VHE mode initialized successfully\n");
2113 kvm_info("Hyp mode initialized successfully\n");
2121 teardown_hyp_mode();
2126 /* NOP: Compiling as a module not supported */
2127 void kvm_arch_exit(void)
2129 kvm_perf_teardown();
2132 static int __init early_kvm_mode_cfg(char *arg)
2137 if (strcmp(arg, "protected") == 0) {
2138 kvm_mode = KVM_MODE_PROTECTED;
2142 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2147 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2149 enum kvm_mode kvm_get_mode(void)
2154 static int arm_init(void)
2156 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2160 module_init(arm_init);