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 mutex_lock(&kvm->lock);
98 if (!system_supports_mte() || kvm->created_vcpus) {
102 kvm->arch.mte_enabled = true;
104 mutex_unlock(&kvm->lock);
114 static int kvm_arm_default_max_vcpus(void)
116 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
119 static void set_default_spectre(struct kvm *kvm)
122 * The default is to expose CSV2 == 1 if the HW isn't affected.
123 * Although this is a per-CPU feature, we make it global because
124 * asymmetric systems are just a nuisance.
126 * Userspace can override this as long as it doesn't promise
129 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
130 kvm->arch.pfr0_csv2 = 1;
131 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
132 kvm->arch.pfr0_csv3 = 1;
136 * kvm_arch_init_vm - initializes a VM data structure
137 * @kvm: pointer to the KVM struct
139 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
143 ret = kvm_arm_setup_stage2(kvm, type);
147 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
151 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
153 goto out_free_stage2_pgd;
155 kvm_vgic_early_init(kvm);
157 /* The maximum number of VCPUs is limited by the host's GIC model */
158 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
160 set_default_spectre(kvm);
164 kvm_free_stage2_pgd(&kvm->arch.mmu);
168 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
170 return VM_FAULT_SIGBUS;
175 * kvm_arch_destroy_vm - destroy the VM data structure
176 * @kvm: pointer to the KVM struct
178 void kvm_arch_destroy_vm(struct kvm *kvm)
182 bitmap_free(kvm->arch.pmu_filter);
184 kvm_vgic_destroy(kvm);
186 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
188 kvm_vcpu_destroy(kvm->vcpus[i]);
189 kvm->vcpus[i] = NULL;
192 atomic_set(&kvm->online_vcpus, 0);
195 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
199 case KVM_CAP_IRQCHIP:
202 case KVM_CAP_IOEVENTFD:
203 case KVM_CAP_DEVICE_CTRL:
204 case KVM_CAP_USER_MEMORY:
205 case KVM_CAP_SYNC_MMU:
206 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
207 case KVM_CAP_ONE_REG:
208 case KVM_CAP_ARM_PSCI:
209 case KVM_CAP_ARM_PSCI_0_2:
210 case KVM_CAP_READONLY_MEM:
211 case KVM_CAP_MP_STATE:
212 case KVM_CAP_IMMEDIATE_EXIT:
213 case KVM_CAP_VCPU_EVENTS:
214 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
215 case KVM_CAP_ARM_NISV_TO_USER:
216 case KVM_CAP_ARM_INJECT_EXT_DABT:
217 case KVM_CAP_SET_GUEST_DEBUG:
218 case KVM_CAP_VCPU_ATTRIBUTES:
219 case KVM_CAP_PTP_KVM:
222 case KVM_CAP_SET_GUEST_DEBUG2:
223 return KVM_GUESTDBG_VALID_MASK;
224 case KVM_CAP_ARM_SET_DEVICE_ADDR:
227 case KVM_CAP_NR_VCPUS:
228 r = num_online_cpus();
230 case KVM_CAP_MAX_VCPUS:
231 case KVM_CAP_MAX_VCPU_ID:
233 r = kvm->arch.max_vcpus;
235 r = kvm_arm_default_max_vcpus();
237 case KVM_CAP_MSI_DEVID:
241 r = kvm->arch.vgic.msis_require_devid;
243 case KVM_CAP_ARM_USER_IRQ:
245 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
246 * (bump this number if adding more devices)
250 case KVM_CAP_ARM_MTE:
251 r = system_supports_mte();
253 case KVM_CAP_STEAL_TIME:
254 r = kvm_arm_pvtime_supported();
256 case KVM_CAP_ARM_EL1_32BIT:
257 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
259 case KVM_CAP_GUEST_DEBUG_HW_BPS:
262 case KVM_CAP_GUEST_DEBUG_HW_WPS:
265 case KVM_CAP_ARM_PMU_V3:
266 r = kvm_arm_support_pmu_v3();
268 case KVM_CAP_ARM_INJECT_SERROR_ESR:
269 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
271 case KVM_CAP_ARM_VM_IPA_SIZE:
272 r = get_kvm_ipa_limit();
274 case KVM_CAP_ARM_SVE:
275 r = system_supports_sve();
277 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
278 case KVM_CAP_ARM_PTRAUTH_GENERIC:
279 r = system_has_full_ptr_auth();
288 long kvm_arch_dev_ioctl(struct file *filp,
289 unsigned int ioctl, unsigned long arg)
294 struct kvm *kvm_arch_alloc_vm(void)
297 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
299 return vzalloc(sizeof(struct kvm));
302 void kvm_arch_free_vm(struct kvm *kvm)
310 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
312 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
315 if (id >= kvm->arch.max_vcpus)
321 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
325 /* Force users to call KVM_ARM_VCPU_INIT */
326 vcpu->arch.target = -1;
327 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
329 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
331 /* Set up the timer */
332 kvm_timer_vcpu_init(vcpu);
334 kvm_pmu_vcpu_init(vcpu);
336 kvm_arm_reset_debug_ptr(vcpu);
338 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
340 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
342 err = kvm_vgic_vcpu_init(vcpu);
346 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
349 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
353 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
355 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
356 static_branch_dec(&userspace_irqchip_in_use);
358 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
359 kvm_timer_vcpu_terminate(vcpu);
360 kvm_pmu_vcpu_destroy(vcpu);
362 kvm_arm_vcpu_destroy(vcpu);
365 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
367 return kvm_timer_is_pending(vcpu);
370 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
373 * If we're about to block (most likely because we've just hit a
374 * WFI), we need to sync back the state of the GIC CPU interface
375 * so that we have the latest PMR and group enables. This ensures
376 * that kvm_arch_vcpu_runnable has up-to-date data to decide
377 * whether we have pending interrupts.
379 * For the same reason, we want to tell GICv4 that we need
380 * doorbells to be signalled, should an interrupt become pending.
383 kvm_vgic_vmcr_sync(vcpu);
384 vgic_v4_put(vcpu, true);
388 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
395 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
397 struct kvm_s2_mmu *mmu;
400 mmu = vcpu->arch.hw_mmu;
401 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
404 * We guarantee that both TLBs and I-cache are private to each
405 * vcpu. If detecting that a vcpu from the same VM has
406 * previously run on the same physical CPU, call into the
407 * hypervisor code to nuke the relevant contexts.
409 * We might get preempted before the vCPU actually runs, but
410 * over-invalidation doesn't affect correctness.
412 if (*last_ran != vcpu->vcpu_id) {
413 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
414 *last_ran = vcpu->vcpu_id;
420 kvm_timer_vcpu_load(vcpu);
422 kvm_vcpu_load_sysregs_vhe(vcpu);
423 kvm_arch_vcpu_load_fp(vcpu);
424 kvm_vcpu_pmu_restore_guest(vcpu);
425 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
426 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
428 if (single_task_running())
429 vcpu_clear_wfx_traps(vcpu);
431 vcpu_set_wfx_traps(vcpu);
433 if (vcpu_has_ptrauth(vcpu))
434 vcpu_ptrauth_disable(vcpu);
435 kvm_arch_vcpu_load_debug_state_flags(vcpu);
438 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
440 kvm_arch_vcpu_put_debug_state_flags(vcpu);
441 kvm_arch_vcpu_put_fp(vcpu);
443 kvm_vcpu_put_sysregs_vhe(vcpu);
444 kvm_timer_vcpu_put(vcpu);
446 kvm_vcpu_pmu_restore_host(vcpu);
451 static void vcpu_power_off(struct kvm_vcpu *vcpu)
453 vcpu->arch.power_off = true;
454 kvm_make_request(KVM_REQ_SLEEP, vcpu);
458 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
459 struct kvm_mp_state *mp_state)
461 if (vcpu->arch.power_off)
462 mp_state->mp_state = KVM_MP_STATE_STOPPED;
464 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
469 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
470 struct kvm_mp_state *mp_state)
474 switch (mp_state->mp_state) {
475 case KVM_MP_STATE_RUNNABLE:
476 vcpu->arch.power_off = false;
478 case KVM_MP_STATE_STOPPED:
479 vcpu_power_off(vcpu);
489 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
490 * @v: The VCPU pointer
492 * If the guest CPU is not waiting for interrupts or an interrupt line is
493 * asserted, the CPU is by definition runnable.
495 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
497 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
498 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
499 && !v->arch.power_off && !v->arch.pause);
502 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
504 return vcpu_mode_priv(vcpu);
507 /* Just ensure a guest exit from a particular CPU */
508 static void exit_vm_noop(void *info)
512 void force_vm_exit(const cpumask_t *mask)
515 smp_call_function_many(mask, exit_vm_noop, NULL, true);
520 * need_new_vmid_gen - check that the VMID is still valid
521 * @vmid: The VMID to check
523 * return true if there is a new generation of VMIDs being used
525 * The hardware supports a limited set of values with the value zero reserved
526 * for the host, so we check if an assigned value belongs to a previous
527 * generation, which requires us to assign a new value. If we're the first to
528 * use a VMID for the new generation, we must flush necessary caches and TLBs
531 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
533 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
534 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
535 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
539 * update_vmid - Update the vmid with a valid VMID for the current generation
540 * @vmid: The stage-2 VMID information struct
542 static void update_vmid(struct kvm_vmid *vmid)
544 if (!need_new_vmid_gen(vmid))
547 spin_lock(&kvm_vmid_lock);
550 * We need to re-check the vmid_gen here to ensure that if another vcpu
551 * already allocated a valid vmid for this vm, then this vcpu should
554 if (!need_new_vmid_gen(vmid)) {
555 spin_unlock(&kvm_vmid_lock);
559 /* First user of a new VMID generation? */
560 if (unlikely(kvm_next_vmid == 0)) {
561 atomic64_inc(&kvm_vmid_gen);
565 * On SMP we know no other CPUs can use this CPU's or each
566 * other's VMID after force_vm_exit returns since the
567 * kvm_vmid_lock blocks them from reentry to the guest.
569 force_vm_exit(cpu_all_mask);
571 * Now broadcast TLB + ICACHE invalidation over the inner
572 * shareable domain to make sure all data structures are
575 kvm_call_hyp(__kvm_flush_vm_context);
578 vmid->vmid = kvm_next_vmid;
580 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
583 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
585 spin_unlock(&kvm_vmid_lock);
588 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
590 struct kvm *kvm = vcpu->kvm;
593 if (likely(vcpu->arch.has_run_once))
596 if (!kvm_arm_vcpu_is_finalized(vcpu))
599 vcpu->arch.has_run_once = true;
601 kvm_arm_vcpu_init_debug(vcpu);
603 if (likely(irqchip_in_kernel(kvm))) {
605 * Map the VGIC hardware resources before running a vcpu the
606 * first time on this VM.
608 ret = kvm_vgic_map_resources(kvm);
613 * Tell the rest of the code that there are userspace irqchip
616 static_branch_inc(&userspace_irqchip_in_use);
619 ret = kvm_timer_enable(vcpu);
623 ret = kvm_arm_pmu_v3_enable(vcpu);
628 bool kvm_arch_intc_initialized(struct kvm *kvm)
630 return vgic_initialized(kvm);
633 void kvm_arm_halt_guest(struct kvm *kvm)
636 struct kvm_vcpu *vcpu;
638 kvm_for_each_vcpu(i, vcpu, kvm)
639 vcpu->arch.pause = true;
640 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
643 void kvm_arm_resume_guest(struct kvm *kvm)
646 struct kvm_vcpu *vcpu;
648 kvm_for_each_vcpu(i, vcpu, kvm) {
649 vcpu->arch.pause = false;
650 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
654 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
656 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
658 rcuwait_wait_event(wait,
659 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
662 if (vcpu->arch.power_off || vcpu->arch.pause) {
663 /* Awaken to handle a signal, request we sleep again later. */
664 kvm_make_request(KVM_REQ_SLEEP, vcpu);
668 * Make sure we will observe a potential reset request if we've
669 * observed a change to the power state. Pairs with the smp_wmb() in
670 * kvm_psci_vcpu_on().
675 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
677 return vcpu->arch.target >= 0;
680 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
682 if (kvm_request_pending(vcpu)) {
683 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
684 vcpu_req_sleep(vcpu);
686 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
687 kvm_reset_vcpu(vcpu);
690 * Clear IRQ_PENDING requests that were made to guarantee
691 * that a VCPU sees new virtual interrupts.
693 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
695 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
696 kvm_update_stolen_time(vcpu);
698 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
699 /* The distributor enable bits were changed */
701 vgic_v4_put(vcpu, false);
706 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
707 kvm_pmu_handle_pmcr(vcpu,
708 __vcpu_sys_reg(vcpu, PMCR_EL0));
712 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
714 if (likely(!vcpu_mode_is_32bit(vcpu)))
717 return !system_supports_32bit_el0() ||
718 static_branch_unlikely(&arm64_mismatched_32bit_el0);
722 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
723 * @vcpu: The VCPU pointer
725 * This function is called through the VCPU_RUN ioctl called from user space. It
726 * will execute VM code in a loop until the time slice for the process is used
727 * or some emulation is needed from user space in which case the function will
728 * return with return value 0 and with the kvm_run structure filled in with the
729 * required data for the requested emulation.
731 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
733 struct kvm_run *run = vcpu->run;
736 if (unlikely(!kvm_vcpu_initialized(vcpu)))
739 ret = kvm_vcpu_first_run_init(vcpu);
743 if (run->exit_reason == KVM_EXIT_MMIO) {
744 ret = kvm_handle_mmio_return(vcpu);
751 if (run->immediate_exit) {
756 kvm_sigset_activate(vcpu);
759 run->exit_reason = KVM_EXIT_UNKNOWN;
762 * Check conditions before entering the guest
766 update_vmid(&vcpu->arch.hw_mmu->vmid);
768 check_vcpu_requests(vcpu);
771 * Preparing the interrupts to be injected also
772 * involves poking the GIC, which must be done in a
773 * non-preemptible context.
777 kvm_pmu_flush_hwstate(vcpu);
781 kvm_vgic_flush_hwstate(vcpu);
784 * Exit if we have a signal pending so that we can deliver the
785 * signal to user space.
787 if (signal_pending(current)) {
789 run->exit_reason = KVM_EXIT_INTR;
793 * If we're using a userspace irqchip, then check if we need
794 * to tell a userspace irqchip about timer or PMU level
795 * changes and if so, exit to userspace (the actual level
796 * state gets updated in kvm_timer_update_run and
797 * kvm_pmu_update_run below).
799 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
800 if (kvm_timer_should_notify_user(vcpu) ||
801 kvm_pmu_should_notify_user(vcpu)) {
803 run->exit_reason = KVM_EXIT_INTR;
808 * Ensure we set mode to IN_GUEST_MODE after we disable
809 * interrupts and before the final VCPU requests check.
810 * See the comment in kvm_vcpu_exiting_guest_mode() and
811 * Documentation/virt/kvm/vcpu-requests.rst
813 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
815 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
816 kvm_request_pending(vcpu)) {
817 vcpu->mode = OUTSIDE_GUEST_MODE;
818 isb(); /* Ensure work in x_flush_hwstate is committed */
819 kvm_pmu_sync_hwstate(vcpu);
820 if (static_branch_unlikely(&userspace_irqchip_in_use))
821 kvm_timer_sync_user(vcpu);
822 kvm_vgic_sync_hwstate(vcpu);
828 kvm_arm_setup_debug(vcpu);
830 /**************************************************************
833 trace_kvm_entry(*vcpu_pc(vcpu));
834 guest_enter_irqoff();
836 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
838 vcpu->mode = OUTSIDE_GUEST_MODE;
842 *************************************************************/
844 kvm_arm_clear_debug(vcpu);
847 * We must sync the PMU state before the vgic state so
848 * that the vgic can properly sample the updated state of the
851 kvm_pmu_sync_hwstate(vcpu);
854 * Sync the vgic state before syncing the timer state because
855 * the timer code needs to know if the virtual timer
856 * interrupts are active.
858 kvm_vgic_sync_hwstate(vcpu);
861 * Sync the timer hardware state before enabling interrupts as
862 * we don't want vtimer interrupts to race with syncing the
863 * timer virtual interrupt state.
865 if (static_branch_unlikely(&userspace_irqchip_in_use))
866 kvm_timer_sync_user(vcpu);
868 kvm_arch_vcpu_ctxsync_fp(vcpu);
871 * We may have taken a host interrupt in HYP mode (ie
872 * while executing the guest). This interrupt is still
873 * pending, as we haven't serviced it yet!
875 * We're now back in SVC mode, with interrupts
876 * disabled. Enabling the interrupts now will have
877 * the effect of taking the interrupt again, in SVC
883 * We do local_irq_enable() before calling guest_exit() so
884 * that if a timer interrupt hits while running the guest we
885 * account that tick as being spent in the guest. We enable
886 * preemption after calling guest_exit() so that if we get
887 * preempted we make sure ticks after that is not counted as
891 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
893 /* Exit types that need handling before we can be preempted */
894 handle_exit_early(vcpu, ret);
899 * The ARMv8 architecture doesn't give the hypervisor
900 * a mechanism to prevent a guest from dropping to AArch32 EL0
901 * if implemented by the CPU. If we spot the guest in such
902 * state and that we decided it wasn't supposed to do so (like
903 * with the asymmetric AArch32 case), return to userspace with
906 if (vcpu_mode_is_bad_32bit(vcpu)) {
908 * As we have caught the guest red-handed, decide that
909 * it isn't fit for purpose anymore by making the vcpu
910 * invalid. The VMM can try and fix it by issuing a
911 * KVM_ARM_VCPU_INIT if it really wants to.
913 vcpu->arch.target = -1;
914 ret = ARM_EXCEPTION_IL;
917 ret = handle_exit(vcpu, ret);
920 /* Tell userspace about in-kernel device output levels */
921 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
922 kvm_timer_update_run(vcpu);
923 kvm_pmu_update_run(vcpu);
926 kvm_sigset_deactivate(vcpu);
930 * In the unlikely event that we are returning to userspace
931 * with pending exceptions or PC adjustment, commit these
932 * adjustments in order to give userspace a consistent view of
933 * the vcpu state. Note that this relies on __kvm_adjust_pc()
934 * being preempt-safe on VHE.
936 if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
937 KVM_ARM64_INCREMENT_PC)))
938 kvm_call_hyp(__kvm_adjust_pc, vcpu);
944 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
950 if (number == KVM_ARM_IRQ_CPU_IRQ)
951 bit_index = __ffs(HCR_VI);
952 else /* KVM_ARM_IRQ_CPU_FIQ */
953 bit_index = __ffs(HCR_VF);
955 hcr = vcpu_hcr(vcpu);
957 set = test_and_set_bit(bit_index, hcr);
959 set = test_and_clear_bit(bit_index, hcr);
962 * If we didn't change anything, no need to wake up or kick other CPUs
968 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
969 * trigger a world-switch round on the running physical CPU to set the
970 * virtual IRQ/FIQ fields in the HCR appropriately.
972 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
978 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
981 u32 irq = irq_level->irq;
982 unsigned int irq_type, vcpu_idx, irq_num;
983 int nrcpus = atomic_read(&kvm->online_vcpus);
984 struct kvm_vcpu *vcpu = NULL;
985 bool level = irq_level->level;
987 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
988 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
989 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
990 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
992 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
995 case KVM_ARM_IRQ_TYPE_CPU:
996 if (irqchip_in_kernel(kvm))
999 if (vcpu_idx >= nrcpus)
1002 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1006 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1009 return vcpu_interrupt_line(vcpu, irq_num, level);
1010 case KVM_ARM_IRQ_TYPE_PPI:
1011 if (!irqchip_in_kernel(kvm))
1014 if (vcpu_idx >= nrcpus)
1017 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1021 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1024 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1025 case KVM_ARM_IRQ_TYPE_SPI:
1026 if (!irqchip_in_kernel(kvm))
1029 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1032 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1038 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1039 const struct kvm_vcpu_init *init)
1041 unsigned int i, ret;
1042 int phys_target = kvm_target_cpu();
1044 if (init->target != phys_target)
1048 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1049 * use the same target.
1051 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1054 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1055 for (i = 0; i < sizeof(init->features) * 8; i++) {
1056 bool set = (init->features[i / 32] & (1 << (i % 32)));
1058 if (set && i >= KVM_VCPU_MAX_FEATURES)
1062 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1063 * use the same feature set.
1065 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1066 test_bit(i, vcpu->arch.features) != set)
1070 set_bit(i, vcpu->arch.features);
1073 vcpu->arch.target = phys_target;
1075 /* Now we know what it is, we can reset it. */
1076 ret = kvm_reset_vcpu(vcpu);
1078 vcpu->arch.target = -1;
1079 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1085 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1086 struct kvm_vcpu_init *init)
1090 ret = kvm_vcpu_set_target(vcpu, init);
1095 * Ensure a rebooted VM will fault in RAM pages and detect if the
1096 * guest MMU is turned off and flush the caches as needed.
1098 * S2FWB enforces all memory accesses to RAM being cacheable,
1099 * ensuring that the data side is always coherent. We still
1100 * need to invalidate the I-cache though, as FWB does *not*
1101 * imply CTR_EL0.DIC.
1103 if (vcpu->arch.has_run_once) {
1104 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1105 stage2_unmap_vm(vcpu->kvm);
1107 icache_inval_all_pou();
1110 vcpu_reset_hcr(vcpu);
1113 * Handle the "start in power-off" case.
1115 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1116 vcpu_power_off(vcpu);
1118 vcpu->arch.power_off = false;
1123 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1124 struct kvm_device_attr *attr)
1128 switch (attr->group) {
1130 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1137 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1138 struct kvm_device_attr *attr)
1142 switch (attr->group) {
1144 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1151 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1152 struct kvm_device_attr *attr)
1156 switch (attr->group) {
1158 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1165 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1166 struct kvm_vcpu_events *events)
1168 memset(events, 0, sizeof(*events));
1170 return __kvm_arm_vcpu_get_events(vcpu, events);
1173 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1174 struct kvm_vcpu_events *events)
1178 /* check whether the reserved field is zero */
1179 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1180 if (events->reserved[i])
1183 /* check whether the pad field is zero */
1184 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1185 if (events->exception.pad[i])
1188 return __kvm_arm_vcpu_set_events(vcpu, events);
1191 long kvm_arch_vcpu_ioctl(struct file *filp,
1192 unsigned int ioctl, unsigned long arg)
1194 struct kvm_vcpu *vcpu = filp->private_data;
1195 void __user *argp = (void __user *)arg;
1196 struct kvm_device_attr attr;
1200 case KVM_ARM_VCPU_INIT: {
1201 struct kvm_vcpu_init init;
1204 if (copy_from_user(&init, argp, sizeof(init)))
1207 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1210 case KVM_SET_ONE_REG:
1211 case KVM_GET_ONE_REG: {
1212 struct kvm_one_reg reg;
1215 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1219 if (copy_from_user(®, argp, sizeof(reg)))
1222 if (ioctl == KVM_SET_ONE_REG)
1223 r = kvm_arm_set_reg(vcpu, ®);
1225 r = kvm_arm_get_reg(vcpu, ®);
1228 case KVM_GET_REG_LIST: {
1229 struct kvm_reg_list __user *user_list = argp;
1230 struct kvm_reg_list reg_list;
1234 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1238 if (!kvm_arm_vcpu_is_finalized(vcpu))
1242 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1245 reg_list.n = kvm_arm_num_regs(vcpu);
1246 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1251 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1254 case KVM_SET_DEVICE_ATTR: {
1256 if (copy_from_user(&attr, argp, sizeof(attr)))
1258 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1261 case KVM_GET_DEVICE_ATTR: {
1263 if (copy_from_user(&attr, argp, sizeof(attr)))
1265 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1268 case KVM_HAS_DEVICE_ATTR: {
1270 if (copy_from_user(&attr, argp, sizeof(attr)))
1272 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1275 case KVM_GET_VCPU_EVENTS: {
1276 struct kvm_vcpu_events events;
1278 if (kvm_arm_vcpu_get_events(vcpu, &events))
1281 if (copy_to_user(argp, &events, sizeof(events)))
1286 case KVM_SET_VCPU_EVENTS: {
1287 struct kvm_vcpu_events events;
1289 if (copy_from_user(&events, argp, sizeof(events)))
1292 return kvm_arm_vcpu_set_events(vcpu, &events);
1294 case KVM_ARM_VCPU_FINALIZE: {
1297 if (!kvm_vcpu_initialized(vcpu))
1300 if (get_user(what, (const int __user *)argp))
1303 return kvm_arm_vcpu_finalize(vcpu, what);
1312 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1317 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1318 const struct kvm_memory_slot *memslot)
1320 kvm_flush_remote_tlbs(kvm);
1323 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1324 struct kvm_arm_device_addr *dev_addr)
1326 unsigned long dev_id, type;
1328 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1329 KVM_ARM_DEVICE_ID_SHIFT;
1330 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1331 KVM_ARM_DEVICE_TYPE_SHIFT;
1334 case KVM_ARM_DEVICE_VGIC_V2:
1337 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1343 long kvm_arch_vm_ioctl(struct file *filp,
1344 unsigned int ioctl, unsigned long arg)
1346 struct kvm *kvm = filp->private_data;
1347 void __user *argp = (void __user *)arg;
1350 case KVM_CREATE_IRQCHIP: {
1354 mutex_lock(&kvm->lock);
1355 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1356 mutex_unlock(&kvm->lock);
1359 case KVM_ARM_SET_DEVICE_ADDR: {
1360 struct kvm_arm_device_addr dev_addr;
1362 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1364 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1366 case KVM_ARM_PREFERRED_TARGET: {
1368 struct kvm_vcpu_init init;
1370 err = kvm_vcpu_preferred_target(&init);
1374 if (copy_to_user(argp, &init, sizeof(init)))
1379 case KVM_ARM_MTE_COPY_TAGS: {
1380 struct kvm_arm_copy_mte_tags copy_tags;
1382 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1384 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1391 static unsigned long nvhe_percpu_size(void)
1393 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1394 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1397 static unsigned long nvhe_percpu_order(void)
1399 unsigned long size = nvhe_percpu_size();
1401 return size ? get_order(size) : 0;
1404 /* A lookup table holding the hypervisor VA for each vector slot */
1405 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1407 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1409 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1412 static int kvm_init_vector_slots(void)
1417 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1418 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1420 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1421 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1423 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1427 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1428 __BP_HARDEN_HYP_VECS_SZ, &base);
1433 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1434 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1438 static void cpu_prepare_hyp_mode(int cpu)
1440 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1444 * Calculate the raw per-cpu offset without a translation from the
1445 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1446 * so that we can use adr_l to access per-cpu variables in EL2.
1447 * Also drop the KASAN tag which gets in the way...
1449 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1450 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1452 params->mair_el2 = read_sysreg(mair_el1);
1455 * The ID map may be configured to use an extended virtual address
1456 * range. This is only the case if system RAM is out of range for the
1457 * currently configured page size and VA_BITS, in which case we will
1458 * also need the extended virtual range for the HYP ID map, or we won't
1459 * be able to enable the EL2 MMU.
1461 * However, at EL2, there is only one TTBR register, and we can't switch
1462 * between translation tables *and* update TCR_EL2.T0SZ at the same
1463 * time. Bottom line: we need to use the extended range with *both* our
1464 * translation tables.
1466 * So use the same T0SZ value we use for the ID map.
1468 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1469 tcr &= ~TCR_T0SZ_MASK;
1470 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1471 params->tcr_el2 = tcr;
1473 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1474 params->pgd_pa = kvm_mmu_get_httbr();
1475 if (is_protected_kvm_enabled())
1476 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1478 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1479 params->vttbr = params->vtcr = 0;
1482 * Flush the init params from the data cache because the struct will
1483 * be read while the MMU is off.
1485 kvm_flush_dcache_to_poc(params, sizeof(*params));
1488 static void hyp_install_host_vector(void)
1490 struct kvm_nvhe_init_params *params;
1491 struct arm_smccc_res res;
1493 /* Switch from the HYP stub to our own HYP init vector */
1494 __hyp_set_vectors(kvm_get_idmap_vector());
1497 * Call initialization code, and switch to the full blown HYP code.
1498 * If the cpucaps haven't been finalized yet, something has gone very
1499 * wrong, and hyp will crash and burn when it uses any
1500 * cpus_have_const_cap() wrapper.
1502 BUG_ON(!system_capabilities_finalized());
1503 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1504 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1505 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1508 static void cpu_init_hyp_mode(void)
1510 hyp_install_host_vector();
1513 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1516 if (this_cpu_has_cap(ARM64_SSBS) &&
1517 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1518 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1522 static void cpu_hyp_reset(void)
1524 if (!is_kernel_in_hyp_mode())
1525 __hyp_reset_vectors();
1529 * EL2 vectors can be mapped and rerouted in a number of ways,
1530 * depending on the kernel configuration and CPU present:
1532 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1533 * placed in one of the vector slots, which is executed before jumping
1534 * to the real vectors.
1536 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1537 * containing the hardening sequence is mapped next to the idmap page,
1538 * and executed before jumping to the real vectors.
1540 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1541 * empty slot is selected, mapped next to the idmap page, and
1542 * executed before jumping to the real vectors.
1544 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1545 * VHE, as we don't have hypervisor-specific mappings. If the system
1546 * is VHE and yet selects this capability, it will be ignored.
1548 static void cpu_set_hyp_vector(void)
1550 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1551 void *vector = hyp_spectre_vector_selector[data->slot];
1553 if (!is_protected_kvm_enabled())
1554 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1556 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1559 static void cpu_hyp_reinit(void)
1561 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1565 if (is_kernel_in_hyp_mode())
1566 kvm_timer_init_vhe();
1568 cpu_init_hyp_mode();
1570 cpu_set_hyp_vector();
1572 kvm_arm_init_debug();
1575 kvm_vgic_init_cpu_hardware();
1578 static void _kvm_arch_hardware_enable(void *discard)
1580 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1582 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1586 int kvm_arch_hardware_enable(void)
1588 _kvm_arch_hardware_enable(NULL);
1592 static void _kvm_arch_hardware_disable(void *discard)
1594 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1596 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1600 void kvm_arch_hardware_disable(void)
1602 if (!is_protected_kvm_enabled())
1603 _kvm_arch_hardware_disable(NULL);
1606 #ifdef CONFIG_CPU_PM
1607 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1612 * kvm_arm_hardware_enabled is left with its old value over
1613 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1618 if (__this_cpu_read(kvm_arm_hardware_enabled))
1620 * don't update kvm_arm_hardware_enabled here
1621 * so that the hardware will be re-enabled
1622 * when we resume. See below.
1627 case CPU_PM_ENTER_FAILED:
1629 if (__this_cpu_read(kvm_arm_hardware_enabled))
1630 /* The hardware was enabled before suspend. */
1640 static struct notifier_block hyp_init_cpu_pm_nb = {
1641 .notifier_call = hyp_init_cpu_pm_notifier,
1644 static void hyp_cpu_pm_init(void)
1646 if (!is_protected_kvm_enabled())
1647 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1649 static void hyp_cpu_pm_exit(void)
1651 if (!is_protected_kvm_enabled())
1652 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1655 static inline void hyp_cpu_pm_init(void)
1658 static inline void hyp_cpu_pm_exit(void)
1663 static void init_cpu_logical_map(void)
1668 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1669 * Only copy the set of online CPUs whose features have been chacked
1670 * against the finalized system capabilities. The hypervisor will not
1671 * allow any other CPUs from the `possible` set to boot.
1673 for_each_online_cpu(cpu)
1674 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1677 #define init_psci_0_1_impl_state(config, what) \
1678 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1680 static bool init_psci_relay(void)
1683 * If PSCI has not been initialized, protected KVM cannot install
1684 * itself on newly booted CPUs.
1686 if (!psci_ops.get_version) {
1687 kvm_err("Cannot initialize protected mode without PSCI\n");
1691 kvm_host_psci_config.version = psci_ops.get_version();
1693 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1694 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1695 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1696 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1697 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1698 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1703 static int init_common_resources(void)
1705 return kvm_set_ipa_limit();
1708 static int init_subsystems(void)
1713 * Enable hardware so that subsystem initialisation can access EL2.
1715 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1718 * Register CPU lower-power notifier
1723 * Init HYP view of VGIC
1725 err = kvm_vgic_hyp_init();
1728 vgic_present = true;
1732 vgic_present = false;
1740 * Init HYP architected timer support
1742 err = kvm_timer_hyp_init(vgic_present);
1747 kvm_sys_reg_table_init();
1750 if (err || !is_protected_kvm_enabled())
1751 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1756 static void teardown_hyp_mode(void)
1761 for_each_possible_cpu(cpu) {
1762 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1763 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1767 static int do_pkvm_init(u32 hyp_va_bits)
1769 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1773 hyp_install_host_vector();
1774 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1775 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1782 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1784 void *addr = phys_to_virt(hyp_mem_base);
1787 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1788 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1790 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1794 ret = do_pkvm_init(hyp_va_bits);
1804 * Inits Hyp-mode on all online CPUs
1806 static int init_hyp_mode(void)
1813 * The protected Hyp-mode cannot be initialized if the memory pool
1814 * allocation has failed.
1816 if (is_protected_kvm_enabled() && !hyp_mem_base)
1820 * Allocate Hyp PGD and setup Hyp identity mapping
1822 err = kvm_mmu_init(&hyp_va_bits);
1827 * Allocate stack pages for Hypervisor-mode
1829 for_each_possible_cpu(cpu) {
1830 unsigned long stack_page;
1832 stack_page = __get_free_page(GFP_KERNEL);
1838 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1842 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1844 for_each_possible_cpu(cpu) {
1848 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1854 page_addr = page_address(page);
1855 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1856 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1860 * Map the Hyp-code called directly from the host
1862 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1863 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1865 kvm_err("Cannot map world-switch code\n");
1869 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1870 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1872 kvm_err("Cannot map .hyp.rodata section\n");
1876 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1877 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1879 kvm_err("Cannot map rodata section\n");
1884 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1885 * section thanks to an assertion in the linker script. Map it RW and
1886 * the rest of .bss RO.
1888 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1889 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1891 kvm_err("Cannot map hyp bss section: %d\n", err);
1895 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1896 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1898 kvm_err("Cannot map bss section\n");
1903 * Map the Hyp stack pages
1905 for_each_possible_cpu(cpu) {
1906 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1907 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1911 kvm_err("Cannot map hyp stack\n");
1916 for_each_possible_cpu(cpu) {
1917 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1918 char *percpu_end = percpu_begin + nvhe_percpu_size();
1920 /* Map Hyp percpu pages */
1921 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1923 kvm_err("Cannot map hyp percpu region\n");
1927 /* Prepare the CPU initialization parameters */
1928 cpu_prepare_hyp_mode(cpu);
1931 if (is_protected_kvm_enabled()) {
1932 init_cpu_logical_map();
1934 if (!init_psci_relay()) {
1940 if (is_protected_kvm_enabled()) {
1941 err = kvm_hyp_init_protection(hyp_va_bits);
1943 kvm_err("Failed to init hyp memory protection\n");
1951 teardown_hyp_mode();
1952 kvm_err("error initializing Hyp mode: %d\n", err);
1956 static void _kvm_host_prot_finalize(void *discard)
1958 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize));
1961 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end)
1963 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end);
1966 #define pkvm_mark_hyp_section(__section) \
1967 pkvm_mark_hyp(__pa_symbol(__section##_start), \
1968 __pa_symbol(__section##_end))
1970 static int finalize_hyp_mode(void)
1974 if (!is_protected_kvm_enabled())
1977 ret = pkvm_mark_hyp_section(__hyp_idmap_text);
1981 ret = pkvm_mark_hyp_section(__hyp_text);
1985 ret = pkvm_mark_hyp_section(__hyp_rodata);
1989 ret = pkvm_mark_hyp_section(__hyp_bss);
1993 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size);
1997 for_each_possible_cpu(cpu) {
1998 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]);
1999 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order());
2001 ret = pkvm_mark_hyp(start, end);
2005 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu));
2006 end = start + PAGE_SIZE;
2007 ret = pkvm_mark_hyp(start, end);
2013 * Flip the static key upfront as that may no longer be possible
2014 * once the host stage 2 is installed.
2016 static_branch_enable(&kvm_protected_mode_initialized);
2017 on_each_cpu(_kvm_host_prot_finalize, NULL, 1);
2022 static void check_kvm_target_cpu(void *ret)
2024 *(int *)ret = kvm_target_cpu();
2027 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2029 struct kvm_vcpu *vcpu;
2032 mpidr &= MPIDR_HWID_BITMASK;
2033 kvm_for_each_vcpu(i, vcpu, kvm) {
2034 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2040 bool kvm_arch_has_irq_bypass(void)
2045 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2046 struct irq_bypass_producer *prod)
2048 struct kvm_kernel_irqfd *irqfd =
2049 container_of(cons, struct kvm_kernel_irqfd, consumer);
2051 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2054 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2055 struct irq_bypass_producer *prod)
2057 struct kvm_kernel_irqfd *irqfd =
2058 container_of(cons, struct kvm_kernel_irqfd, consumer);
2060 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2064 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2066 struct kvm_kernel_irqfd *irqfd =
2067 container_of(cons, struct kvm_kernel_irqfd, consumer);
2069 kvm_arm_halt_guest(irqfd->kvm);
2072 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2074 struct kvm_kernel_irqfd *irqfd =
2075 container_of(cons, struct kvm_kernel_irqfd, consumer);
2077 kvm_arm_resume_guest(irqfd->kvm);
2081 * Initialize Hyp-mode and memory mappings on all CPUs.
2083 int kvm_arch_init(void *opaque)
2089 if (!is_hyp_mode_available()) {
2090 kvm_info("HYP mode not available\n");
2094 in_hyp_mode = is_kernel_in_hyp_mode();
2096 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2097 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2098 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2099 "Only trusted guests should be used on this system.\n");
2101 for_each_online_cpu(cpu) {
2102 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
2104 kvm_err("Error, CPU %d not supported!\n", cpu);
2109 err = init_common_resources();
2113 err = kvm_arm_init_sve();
2118 err = init_hyp_mode();
2123 err = kvm_init_vector_slots();
2125 kvm_err("Cannot initialise vector slots\n");
2129 err = init_subsystems();
2134 err = finalize_hyp_mode();
2136 kvm_err("Failed to finalize Hyp protection\n");
2141 if (is_protected_kvm_enabled()) {
2142 kvm_info("Protected nVHE mode initialized successfully\n");
2143 } else if (in_hyp_mode) {
2144 kvm_info("VHE mode initialized successfully\n");
2146 kvm_info("Hyp mode initialized successfully\n");
2154 teardown_hyp_mode();
2159 /* NOP: Compiling as a module not supported */
2160 void kvm_arch_exit(void)
2162 kvm_perf_teardown();
2165 static int __init early_kvm_mode_cfg(char *arg)
2170 if (strcmp(arg, "protected") == 0) {
2171 kvm_mode = KVM_MODE_PROTECTED;
2175 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
2180 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2182 enum kvm_mode kvm_get_mode(void)
2187 static int arm_init(void)
2189 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2193 module_init(arm_init);