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:
211 case KVM_CAP_ARM_SET_DEVICE_ADDR:
214 case KVM_CAP_NR_VCPUS:
215 r = num_online_cpus();
217 case KVM_CAP_MAX_VCPUS:
218 case KVM_CAP_MAX_VCPU_ID:
220 r = kvm->arch.max_vcpus;
222 r = kvm_arm_default_max_vcpus();
224 case KVM_CAP_MSI_DEVID:
228 r = kvm->arch.vgic.msis_require_devid;
230 case KVM_CAP_ARM_USER_IRQ:
232 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
233 * (bump this number if adding more devices)
237 case KVM_CAP_STEAL_TIME:
238 r = kvm_arm_pvtime_supported();
240 case KVM_CAP_ARM_EL1_32BIT:
241 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
243 case KVM_CAP_GUEST_DEBUG_HW_BPS:
246 case KVM_CAP_GUEST_DEBUG_HW_WPS:
249 case KVM_CAP_ARM_PMU_V3:
250 r = kvm_arm_support_pmu_v3();
252 case KVM_CAP_ARM_INJECT_SERROR_ESR:
253 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
255 case KVM_CAP_ARM_VM_IPA_SIZE:
256 r = get_kvm_ipa_limit();
258 case KVM_CAP_ARM_SVE:
259 r = system_supports_sve();
261 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
262 case KVM_CAP_ARM_PTRAUTH_GENERIC:
263 r = system_has_full_ptr_auth();
272 long kvm_arch_dev_ioctl(struct file *filp,
273 unsigned int ioctl, unsigned long arg)
278 struct kvm *kvm_arch_alloc_vm(void)
281 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
283 return vzalloc(sizeof(struct kvm));
286 void kvm_arch_free_vm(struct kvm *kvm)
294 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
296 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
299 if (id >= kvm->arch.max_vcpus)
305 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
309 /* Force users to call KVM_ARM_VCPU_INIT */
310 vcpu->arch.target = -1;
311 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
313 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
315 /* Set up the timer */
316 kvm_timer_vcpu_init(vcpu);
318 kvm_pmu_vcpu_init(vcpu);
320 kvm_arm_reset_debug_ptr(vcpu);
322 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
324 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
326 err = kvm_vgic_vcpu_init(vcpu);
330 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
333 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
337 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
339 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
340 static_branch_dec(&userspace_irqchip_in_use);
342 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
343 kvm_timer_vcpu_terminate(vcpu);
344 kvm_pmu_vcpu_destroy(vcpu);
346 kvm_arm_vcpu_destroy(vcpu);
349 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
351 return kvm_timer_is_pending(vcpu);
354 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
357 * If we're about to block (most likely because we've just hit a
358 * WFI), we need to sync back the state of the GIC CPU interface
359 * so that we have the latest PMR and group enables. This ensures
360 * that kvm_arch_vcpu_runnable has up-to-date data to decide
361 * whether we have pending interrupts.
363 * For the same reason, we want to tell GICv4 that we need
364 * doorbells to be signalled, should an interrupt become pending.
367 kvm_vgic_vmcr_sync(vcpu);
368 vgic_v4_put(vcpu, true);
372 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
379 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
381 struct kvm_s2_mmu *mmu;
384 mmu = vcpu->arch.hw_mmu;
385 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
388 * We might get preempted before the vCPU actually runs, but
389 * over-invalidation doesn't affect correctness.
391 if (*last_ran != vcpu->vcpu_id) {
392 kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu);
393 *last_ran = vcpu->vcpu_id;
399 kvm_timer_vcpu_load(vcpu);
401 kvm_vcpu_load_sysregs_vhe(vcpu);
402 kvm_arch_vcpu_load_fp(vcpu);
403 kvm_vcpu_pmu_restore_guest(vcpu);
404 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
405 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
407 if (single_task_running())
408 vcpu_clear_wfx_traps(vcpu);
410 vcpu_set_wfx_traps(vcpu);
412 if (vcpu_has_ptrauth(vcpu))
413 vcpu_ptrauth_disable(vcpu);
416 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
418 kvm_arch_vcpu_put_fp(vcpu);
420 kvm_vcpu_put_sysregs_vhe(vcpu);
421 kvm_timer_vcpu_put(vcpu);
423 kvm_vcpu_pmu_restore_host(vcpu);
428 static void vcpu_power_off(struct kvm_vcpu *vcpu)
430 vcpu->arch.power_off = true;
431 kvm_make_request(KVM_REQ_SLEEP, vcpu);
435 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
436 struct kvm_mp_state *mp_state)
438 if (vcpu->arch.power_off)
439 mp_state->mp_state = KVM_MP_STATE_STOPPED;
441 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
446 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
447 struct kvm_mp_state *mp_state)
451 switch (mp_state->mp_state) {
452 case KVM_MP_STATE_RUNNABLE:
453 vcpu->arch.power_off = false;
455 case KVM_MP_STATE_STOPPED:
456 vcpu_power_off(vcpu);
466 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
467 * @v: The VCPU pointer
469 * If the guest CPU is not waiting for interrupts or an interrupt line is
470 * asserted, the CPU is by definition runnable.
472 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
474 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
475 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
476 && !v->arch.power_off && !v->arch.pause);
479 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
481 return vcpu_mode_priv(vcpu);
484 /* Just ensure a guest exit from a particular CPU */
485 static void exit_vm_noop(void *info)
489 void force_vm_exit(const cpumask_t *mask)
492 smp_call_function_many(mask, exit_vm_noop, NULL, true);
497 * need_new_vmid_gen - check that the VMID is still valid
498 * @vmid: The VMID to check
500 * return true if there is a new generation of VMIDs being used
502 * The hardware supports a limited set of values with the value zero reserved
503 * for the host, so we check if an assigned value belongs to a previous
504 * generation, which requires us to assign a new value. If we're the first to
505 * use a VMID for the new generation, we must flush necessary caches and TLBs
508 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
510 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
511 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
512 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
516 * update_vmid - Update the vmid with a valid VMID for the current generation
517 * @vmid: The stage-2 VMID information struct
519 static void update_vmid(struct kvm_vmid *vmid)
521 if (!need_new_vmid_gen(vmid))
524 spin_lock(&kvm_vmid_lock);
527 * We need to re-check the vmid_gen here to ensure that if another vcpu
528 * already allocated a valid vmid for this vm, then this vcpu should
531 if (!need_new_vmid_gen(vmid)) {
532 spin_unlock(&kvm_vmid_lock);
536 /* First user of a new VMID generation? */
537 if (unlikely(kvm_next_vmid == 0)) {
538 atomic64_inc(&kvm_vmid_gen);
542 * On SMP we know no other CPUs can use this CPU's or each
543 * other's VMID after force_vm_exit returns since the
544 * kvm_vmid_lock blocks them from reentry to the guest.
546 force_vm_exit(cpu_all_mask);
548 * Now broadcast TLB + ICACHE invalidation over the inner
549 * shareable domain to make sure all data structures are
552 kvm_call_hyp(__kvm_flush_vm_context);
555 vmid->vmid = kvm_next_vmid;
557 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
560 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
562 spin_unlock(&kvm_vmid_lock);
565 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
567 struct kvm *kvm = vcpu->kvm;
570 if (likely(vcpu->arch.has_run_once))
573 if (!kvm_arm_vcpu_is_finalized(vcpu))
576 vcpu->arch.has_run_once = true;
578 if (likely(irqchip_in_kernel(kvm))) {
580 * Map the VGIC hardware resources before running a vcpu the
581 * first time on this VM.
583 ret = kvm_vgic_map_resources(kvm);
588 * Tell the rest of the code that there are userspace irqchip
591 static_branch_inc(&userspace_irqchip_in_use);
594 ret = kvm_timer_enable(vcpu);
598 ret = kvm_arm_pmu_v3_enable(vcpu);
603 bool kvm_arch_intc_initialized(struct kvm *kvm)
605 return vgic_initialized(kvm);
608 void kvm_arm_halt_guest(struct kvm *kvm)
611 struct kvm_vcpu *vcpu;
613 kvm_for_each_vcpu(i, vcpu, kvm)
614 vcpu->arch.pause = true;
615 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
618 void kvm_arm_resume_guest(struct kvm *kvm)
621 struct kvm_vcpu *vcpu;
623 kvm_for_each_vcpu(i, vcpu, kvm) {
624 vcpu->arch.pause = false;
625 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
629 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
631 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
633 rcuwait_wait_event(wait,
634 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
637 if (vcpu->arch.power_off || vcpu->arch.pause) {
638 /* Awaken to handle a signal, request we sleep again later. */
639 kvm_make_request(KVM_REQ_SLEEP, vcpu);
643 * Make sure we will observe a potential reset request if we've
644 * observed a change to the power state. Pairs with the smp_wmb() in
645 * kvm_psci_vcpu_on().
650 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
652 return vcpu->arch.target >= 0;
655 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
657 if (kvm_request_pending(vcpu)) {
658 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
659 vcpu_req_sleep(vcpu);
661 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
662 kvm_reset_vcpu(vcpu);
665 * Clear IRQ_PENDING requests that were made to guarantee
666 * that a VCPU sees new virtual interrupts.
668 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
670 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
671 kvm_update_stolen_time(vcpu);
673 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
674 /* The distributor enable bits were changed */
676 vgic_v4_put(vcpu, false);
684 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
685 * @vcpu: The VCPU pointer
687 * This function is called through the VCPU_RUN ioctl called from user space. It
688 * will execute VM code in a loop until the time slice for the process is used
689 * or some emulation is needed from user space in which case the function will
690 * return with return value 0 and with the kvm_run structure filled in with the
691 * required data for the requested emulation.
693 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
695 struct kvm_run *run = vcpu->run;
698 if (unlikely(!kvm_vcpu_initialized(vcpu)))
701 ret = kvm_vcpu_first_run_init(vcpu);
705 if (run->exit_reason == KVM_EXIT_MMIO) {
706 ret = kvm_handle_mmio_return(vcpu);
711 if (run->immediate_exit)
716 kvm_sigset_activate(vcpu);
719 run->exit_reason = KVM_EXIT_UNKNOWN;
722 * Check conditions before entering the guest
726 update_vmid(&vcpu->arch.hw_mmu->vmid);
728 check_vcpu_requests(vcpu);
731 * Preparing the interrupts to be injected also
732 * involves poking the GIC, which must be done in a
733 * non-preemptible context.
737 kvm_pmu_flush_hwstate(vcpu);
741 kvm_vgic_flush_hwstate(vcpu);
744 * Exit if we have a signal pending so that we can deliver the
745 * signal to user space.
747 if (signal_pending(current)) {
749 run->exit_reason = KVM_EXIT_INTR;
753 * If we're using a userspace irqchip, then check if we need
754 * to tell a userspace irqchip about timer or PMU level
755 * changes and if so, exit to userspace (the actual level
756 * state gets updated in kvm_timer_update_run and
757 * kvm_pmu_update_run below).
759 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
760 if (kvm_timer_should_notify_user(vcpu) ||
761 kvm_pmu_should_notify_user(vcpu)) {
763 run->exit_reason = KVM_EXIT_INTR;
768 * Ensure we set mode to IN_GUEST_MODE after we disable
769 * interrupts and before the final VCPU requests check.
770 * See the comment in kvm_vcpu_exiting_guest_mode() and
771 * Documentation/virt/kvm/vcpu-requests.rst
773 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
775 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
776 kvm_request_pending(vcpu)) {
777 vcpu->mode = OUTSIDE_GUEST_MODE;
778 isb(); /* Ensure work in x_flush_hwstate is committed */
779 kvm_pmu_sync_hwstate(vcpu);
780 if (static_branch_unlikely(&userspace_irqchip_in_use))
781 kvm_timer_sync_user(vcpu);
782 kvm_vgic_sync_hwstate(vcpu);
788 kvm_arm_setup_debug(vcpu);
790 /**************************************************************
793 trace_kvm_entry(*vcpu_pc(vcpu));
794 guest_enter_irqoff();
796 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
798 vcpu->mode = OUTSIDE_GUEST_MODE;
802 *************************************************************/
804 kvm_arm_clear_debug(vcpu);
807 * We must sync the PMU state before the vgic state so
808 * that the vgic can properly sample the updated state of the
811 kvm_pmu_sync_hwstate(vcpu);
814 * Sync the vgic state before syncing the timer state because
815 * the timer code needs to know if the virtual timer
816 * interrupts are active.
818 kvm_vgic_sync_hwstate(vcpu);
821 * Sync the timer hardware state before enabling interrupts as
822 * we don't want vtimer interrupts to race with syncing the
823 * timer virtual interrupt state.
825 if (static_branch_unlikely(&userspace_irqchip_in_use))
826 kvm_timer_sync_user(vcpu);
828 kvm_arch_vcpu_ctxsync_fp(vcpu);
831 * We may have taken a host interrupt in HYP mode (ie
832 * while executing the guest). This interrupt is still
833 * pending, as we haven't serviced it yet!
835 * We're now back in SVC mode, with interrupts
836 * disabled. Enabling the interrupts now will have
837 * the effect of taking the interrupt again, in SVC
843 * We do local_irq_enable() before calling guest_exit() so
844 * that if a timer interrupt hits while running the guest we
845 * account that tick as being spent in the guest. We enable
846 * preemption after calling guest_exit() so that if we get
847 * preempted we make sure ticks after that is not counted as
851 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
853 /* Exit types that need handling before we can be preempted */
854 handle_exit_early(vcpu, ret);
859 * The ARMv8 architecture doesn't give the hypervisor
860 * a mechanism to prevent a guest from dropping to AArch32 EL0
861 * if implemented by the CPU. If we spot the guest in such
862 * state and that we decided it wasn't supposed to do so (like
863 * with the asymmetric AArch32 case), return to userspace with
866 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
868 * As we have caught the guest red-handed, decide that
869 * it isn't fit for purpose anymore by making the vcpu
870 * invalid. The VMM can try and fix it by issuing a
871 * KVM_ARM_VCPU_INIT if it really wants to.
873 vcpu->arch.target = -1;
874 ret = ARM_EXCEPTION_IL;
877 ret = handle_exit(vcpu, ret);
880 /* Tell userspace about in-kernel device output levels */
881 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
882 kvm_timer_update_run(vcpu);
883 kvm_pmu_update_run(vcpu);
886 kvm_sigset_deactivate(vcpu);
892 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
898 if (number == KVM_ARM_IRQ_CPU_IRQ)
899 bit_index = __ffs(HCR_VI);
900 else /* KVM_ARM_IRQ_CPU_FIQ */
901 bit_index = __ffs(HCR_VF);
903 hcr = vcpu_hcr(vcpu);
905 set = test_and_set_bit(bit_index, hcr);
907 set = test_and_clear_bit(bit_index, hcr);
910 * If we didn't change anything, no need to wake up or kick other CPUs
916 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
917 * trigger a world-switch round on the running physical CPU to set the
918 * virtual IRQ/FIQ fields in the HCR appropriately.
920 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
926 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
929 u32 irq = irq_level->irq;
930 unsigned int irq_type, vcpu_idx, irq_num;
931 int nrcpus = atomic_read(&kvm->online_vcpus);
932 struct kvm_vcpu *vcpu = NULL;
933 bool level = irq_level->level;
935 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
936 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
937 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
938 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
940 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
943 case KVM_ARM_IRQ_TYPE_CPU:
944 if (irqchip_in_kernel(kvm))
947 if (vcpu_idx >= nrcpus)
950 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
954 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
957 return vcpu_interrupt_line(vcpu, irq_num, level);
958 case KVM_ARM_IRQ_TYPE_PPI:
959 if (!irqchip_in_kernel(kvm))
962 if (vcpu_idx >= nrcpus)
965 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
969 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
972 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
973 case KVM_ARM_IRQ_TYPE_SPI:
974 if (!irqchip_in_kernel(kvm))
977 if (irq_num < VGIC_NR_PRIVATE_IRQS)
980 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
986 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
987 const struct kvm_vcpu_init *init)
990 int phys_target = kvm_target_cpu();
992 if (init->target != phys_target)
996 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
997 * use the same target.
999 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1002 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1003 for (i = 0; i < sizeof(init->features) * 8; i++) {
1004 bool set = (init->features[i / 32] & (1 << (i % 32)));
1006 if (set && i >= KVM_VCPU_MAX_FEATURES)
1010 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1011 * use the same feature set.
1013 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1014 test_bit(i, vcpu->arch.features) != set)
1018 set_bit(i, vcpu->arch.features);
1021 vcpu->arch.target = phys_target;
1023 /* Now we know what it is, we can reset it. */
1024 ret = kvm_reset_vcpu(vcpu);
1026 vcpu->arch.target = -1;
1027 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1033 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1034 struct kvm_vcpu_init *init)
1038 ret = kvm_vcpu_set_target(vcpu, init);
1043 * Ensure a rebooted VM will fault in RAM pages and detect if the
1044 * guest MMU is turned off and flush the caches as needed.
1046 * S2FWB enforces all memory accesses to RAM being cacheable,
1047 * ensuring that the data side is always coherent. We still
1048 * need to invalidate the I-cache though, as FWB does *not*
1049 * imply CTR_EL0.DIC.
1051 if (vcpu->arch.has_run_once) {
1052 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1053 stage2_unmap_vm(vcpu->kvm);
1055 __flush_icache_all();
1058 vcpu_reset_hcr(vcpu);
1061 * Handle the "start in power-off" case.
1063 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1064 vcpu_power_off(vcpu);
1066 vcpu->arch.power_off = false;
1071 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1072 struct kvm_device_attr *attr)
1076 switch (attr->group) {
1078 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1085 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1086 struct kvm_device_attr *attr)
1090 switch (attr->group) {
1092 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1099 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1100 struct kvm_device_attr *attr)
1104 switch (attr->group) {
1106 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1113 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1114 struct kvm_vcpu_events *events)
1116 memset(events, 0, sizeof(*events));
1118 return __kvm_arm_vcpu_get_events(vcpu, events);
1121 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1122 struct kvm_vcpu_events *events)
1126 /* check whether the reserved field is zero */
1127 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1128 if (events->reserved[i])
1131 /* check whether the pad field is zero */
1132 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1133 if (events->exception.pad[i])
1136 return __kvm_arm_vcpu_set_events(vcpu, events);
1139 long kvm_arch_vcpu_ioctl(struct file *filp,
1140 unsigned int ioctl, unsigned long arg)
1142 struct kvm_vcpu *vcpu = filp->private_data;
1143 void __user *argp = (void __user *)arg;
1144 struct kvm_device_attr attr;
1148 case KVM_ARM_VCPU_INIT: {
1149 struct kvm_vcpu_init init;
1152 if (copy_from_user(&init, argp, sizeof(init)))
1155 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1158 case KVM_SET_ONE_REG:
1159 case KVM_GET_ONE_REG: {
1160 struct kvm_one_reg reg;
1163 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1167 if (copy_from_user(®, argp, sizeof(reg)))
1170 if (ioctl == KVM_SET_ONE_REG)
1171 r = kvm_arm_set_reg(vcpu, ®);
1173 r = kvm_arm_get_reg(vcpu, ®);
1176 case KVM_GET_REG_LIST: {
1177 struct kvm_reg_list __user *user_list = argp;
1178 struct kvm_reg_list reg_list;
1182 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1186 if (!kvm_arm_vcpu_is_finalized(vcpu))
1190 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1193 reg_list.n = kvm_arm_num_regs(vcpu);
1194 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1199 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1202 case KVM_SET_DEVICE_ATTR: {
1204 if (copy_from_user(&attr, argp, sizeof(attr)))
1206 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1209 case KVM_GET_DEVICE_ATTR: {
1211 if (copy_from_user(&attr, argp, sizeof(attr)))
1213 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1216 case KVM_HAS_DEVICE_ATTR: {
1218 if (copy_from_user(&attr, argp, sizeof(attr)))
1220 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1223 case KVM_GET_VCPU_EVENTS: {
1224 struct kvm_vcpu_events events;
1226 if (kvm_arm_vcpu_get_events(vcpu, &events))
1229 if (copy_to_user(argp, &events, sizeof(events)))
1234 case KVM_SET_VCPU_EVENTS: {
1235 struct kvm_vcpu_events events;
1237 if (copy_from_user(&events, argp, sizeof(events)))
1240 return kvm_arm_vcpu_set_events(vcpu, &events);
1242 case KVM_ARM_VCPU_FINALIZE: {
1245 if (!kvm_vcpu_initialized(vcpu))
1248 if (get_user(what, (const int __user *)argp))
1251 return kvm_arm_vcpu_finalize(vcpu, what);
1260 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1265 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1266 struct kvm_memory_slot *memslot)
1268 kvm_flush_remote_tlbs(kvm);
1271 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1272 struct kvm_arm_device_addr *dev_addr)
1274 unsigned long dev_id, type;
1276 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1277 KVM_ARM_DEVICE_ID_SHIFT;
1278 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1279 KVM_ARM_DEVICE_TYPE_SHIFT;
1282 case KVM_ARM_DEVICE_VGIC_V2:
1285 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1291 long kvm_arch_vm_ioctl(struct file *filp,
1292 unsigned int ioctl, unsigned long arg)
1294 struct kvm *kvm = filp->private_data;
1295 void __user *argp = (void __user *)arg;
1298 case KVM_CREATE_IRQCHIP: {
1302 mutex_lock(&kvm->lock);
1303 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1304 mutex_unlock(&kvm->lock);
1307 case KVM_ARM_SET_DEVICE_ADDR: {
1308 struct kvm_arm_device_addr dev_addr;
1310 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1312 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1314 case KVM_ARM_PREFERRED_TARGET: {
1316 struct kvm_vcpu_init init;
1318 err = kvm_vcpu_preferred_target(&init);
1322 if (copy_to_user(argp, &init, sizeof(init)))
1332 static unsigned long nvhe_percpu_size(void)
1334 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1335 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1338 static unsigned long nvhe_percpu_order(void)
1340 unsigned long size = nvhe_percpu_size();
1342 return size ? get_order(size) : 0;
1345 /* A lookup table holding the hypervisor VA for each vector slot */
1346 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1348 static int __kvm_vector_slot2idx(enum arm64_hyp_spectre_vector slot)
1350 return slot - (slot != HYP_VECTOR_DIRECT);
1353 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1355 int idx = __kvm_vector_slot2idx(slot);
1357 hyp_spectre_vector_selector[slot] = base + (idx * SZ_2K);
1360 static int kvm_init_vector_slots(void)
1365 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1366 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1368 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1369 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1371 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1375 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1376 __BP_HARDEN_HYP_VECS_SZ, &base);
1381 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1382 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1386 static void cpu_init_hyp_mode(void)
1388 struct kvm_nvhe_init_params *params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1389 struct arm_smccc_res res;
1392 /* Switch from the HYP stub to our own HYP init vector */
1393 __hyp_set_vectors(kvm_get_idmap_vector());
1396 * Calculate the raw per-cpu offset without a translation from the
1397 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1398 * so that we can use adr_l to access per-cpu variables in EL2.
1399 * Also drop the KASAN tag which gets in the way...
1401 params->tpidr_el2 = (unsigned long)kasan_reset_tag(this_cpu_ptr_nvhe_sym(__per_cpu_start)) -
1402 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1404 params->mair_el2 = read_sysreg(mair_el1);
1407 * The ID map may be configured to use an extended virtual address
1408 * range. This is only the case if system RAM is out of range for the
1409 * currently configured page size and VA_BITS, in which case we will
1410 * also need the extended virtual range for the HYP ID map, or we won't
1411 * be able to enable the EL2 MMU.
1413 * However, at EL2, there is only one TTBR register, and we can't switch
1414 * between translation tables *and* update TCR_EL2.T0SZ at the same
1415 * time. Bottom line: we need to use the extended range with *both* our
1416 * translation tables.
1418 * So use the same T0SZ value we use for the ID map.
1420 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1421 tcr &= ~TCR_T0SZ_MASK;
1422 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1423 params->tcr_el2 = tcr;
1425 params->stack_hyp_va = kern_hyp_va(__this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE);
1426 params->pgd_pa = kvm_mmu_get_httbr();
1429 * Flush the init params from the data cache because the struct will
1430 * be read while the MMU is off.
1432 kvm_flush_dcache_to_poc(params, sizeof(*params));
1435 * Call initialization code, and switch to the full blown HYP code.
1436 * If the cpucaps haven't been finalized yet, something has gone very
1437 * wrong, and hyp will crash and burn when it uses any
1438 * cpus_have_const_cap() wrapper.
1440 BUG_ON(!system_capabilities_finalized());
1441 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1442 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1445 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1448 if (this_cpu_has_cap(ARM64_SSBS) &&
1449 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1450 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1454 static void cpu_hyp_reset(void)
1456 if (!is_kernel_in_hyp_mode())
1457 __hyp_reset_vectors();
1461 * EL2 vectors can be mapped and rerouted in a number of ways,
1462 * depending on the kernel configuration and CPU present:
1464 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1465 * placed in one of the vector slots, which is executed before jumping
1466 * to the real vectors.
1468 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1469 * containing the hardening sequence is mapped next to the idmap page,
1470 * and executed before jumping to the real vectors.
1472 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1473 * empty slot is selected, mapped next to the idmap page, and
1474 * executed before jumping to the real vectors.
1476 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1477 * VHE, as we don't have hypervisor-specific mappings. If the system
1478 * is VHE and yet selects this capability, it will be ignored.
1480 static void cpu_set_hyp_vector(void)
1482 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1483 void *vector = hyp_spectre_vector_selector[data->slot];
1485 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1488 static void cpu_hyp_reinit(void)
1490 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1493 cpu_set_hyp_vector();
1495 if (is_kernel_in_hyp_mode())
1496 kvm_timer_init_vhe();
1498 cpu_init_hyp_mode();
1500 kvm_arm_init_debug();
1503 kvm_vgic_init_cpu_hardware();
1506 static void _kvm_arch_hardware_enable(void *discard)
1508 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1510 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1514 int kvm_arch_hardware_enable(void)
1516 _kvm_arch_hardware_enable(NULL);
1520 static void _kvm_arch_hardware_disable(void *discard)
1522 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1524 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1528 void kvm_arch_hardware_disable(void)
1530 if (!is_protected_kvm_enabled())
1531 _kvm_arch_hardware_disable(NULL);
1534 #ifdef CONFIG_CPU_PM
1535 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1540 * kvm_arm_hardware_enabled is left with its old value over
1541 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1546 if (__this_cpu_read(kvm_arm_hardware_enabled))
1548 * don't update kvm_arm_hardware_enabled here
1549 * so that the hardware will be re-enabled
1550 * when we resume. See below.
1555 case CPU_PM_ENTER_FAILED:
1557 if (__this_cpu_read(kvm_arm_hardware_enabled))
1558 /* The hardware was enabled before suspend. */
1568 static struct notifier_block hyp_init_cpu_pm_nb = {
1569 .notifier_call = hyp_init_cpu_pm_notifier,
1572 static void hyp_cpu_pm_init(void)
1574 if (!is_protected_kvm_enabled())
1575 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1577 static void hyp_cpu_pm_exit(void)
1579 if (!is_protected_kvm_enabled())
1580 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1583 static inline void hyp_cpu_pm_init(void)
1586 static inline void hyp_cpu_pm_exit(void)
1591 static void init_cpu_logical_map(void)
1596 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1597 * Only copy the set of online CPUs whose features have been chacked
1598 * against the finalized system capabilities. The hypervisor will not
1599 * allow any other CPUs from the `possible` set to boot.
1601 for_each_online_cpu(cpu)
1602 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1605 #define init_psci_0_1_impl_state(config, what) \
1606 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1608 static bool init_psci_relay(void)
1611 * If PSCI has not been initialized, protected KVM cannot install
1612 * itself on newly booted CPUs.
1614 if (!psci_ops.get_version) {
1615 kvm_err("Cannot initialize protected mode without PSCI\n");
1619 kvm_host_psci_config.version = psci_ops.get_version();
1621 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1622 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1623 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1624 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1625 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1626 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1631 static int init_common_resources(void)
1633 return kvm_set_ipa_limit();
1636 static int init_subsystems(void)
1641 * Enable hardware so that subsystem initialisation can access EL2.
1643 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1646 * Register CPU lower-power notifier
1651 * Init HYP view of VGIC
1653 err = kvm_vgic_hyp_init();
1656 vgic_present = true;
1660 vgic_present = false;
1668 * Init HYP architected timer support
1670 err = kvm_timer_hyp_init(vgic_present);
1675 kvm_sys_reg_table_init();
1678 if (err || !is_protected_kvm_enabled())
1679 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1684 static void teardown_hyp_mode(void)
1689 for_each_possible_cpu(cpu) {
1690 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1691 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1696 * Inits Hyp-mode on all online CPUs
1698 static int init_hyp_mode(void)
1704 * Allocate Hyp PGD and setup Hyp identity mapping
1706 err = kvm_mmu_init();
1711 * Allocate stack pages for Hypervisor-mode
1713 for_each_possible_cpu(cpu) {
1714 unsigned long stack_page;
1716 stack_page = __get_free_page(GFP_KERNEL);
1722 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1726 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1728 for_each_possible_cpu(cpu) {
1732 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1738 page_addr = page_address(page);
1739 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1740 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1744 * Map the Hyp-code called directly from the host
1746 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1747 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1749 kvm_err("Cannot map world-switch code\n");
1753 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1754 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1756 kvm_err("Cannot map .hyp.rodata section\n");
1760 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1761 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1763 kvm_err("Cannot map rodata section\n");
1767 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1768 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1770 kvm_err("Cannot map bss section\n");
1775 * Map the Hyp stack pages
1777 for_each_possible_cpu(cpu) {
1778 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1779 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1783 kvm_err("Cannot map hyp stack\n");
1789 * Map Hyp percpu pages
1791 for_each_possible_cpu(cpu) {
1792 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1793 char *percpu_end = percpu_begin + nvhe_percpu_size();
1795 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1798 kvm_err("Cannot map hyp percpu region\n");
1803 if (is_protected_kvm_enabled()) {
1804 init_cpu_logical_map();
1806 if (!init_psci_relay())
1813 teardown_hyp_mode();
1814 kvm_err("error initializing Hyp mode: %d\n", err);
1818 static void check_kvm_target_cpu(void *ret)
1820 *(int *)ret = kvm_target_cpu();
1823 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1825 struct kvm_vcpu *vcpu;
1828 mpidr &= MPIDR_HWID_BITMASK;
1829 kvm_for_each_vcpu(i, vcpu, kvm) {
1830 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1836 bool kvm_arch_has_irq_bypass(void)
1841 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1842 struct irq_bypass_producer *prod)
1844 struct kvm_kernel_irqfd *irqfd =
1845 container_of(cons, struct kvm_kernel_irqfd, consumer);
1847 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1850 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1851 struct irq_bypass_producer *prod)
1853 struct kvm_kernel_irqfd *irqfd =
1854 container_of(cons, struct kvm_kernel_irqfd, consumer);
1856 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1860 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1862 struct kvm_kernel_irqfd *irqfd =
1863 container_of(cons, struct kvm_kernel_irqfd, consumer);
1865 kvm_arm_halt_guest(irqfd->kvm);
1868 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1870 struct kvm_kernel_irqfd *irqfd =
1871 container_of(cons, struct kvm_kernel_irqfd, consumer);
1873 kvm_arm_resume_guest(irqfd->kvm);
1877 * Initialize Hyp-mode and memory mappings on all CPUs.
1879 int kvm_arch_init(void *opaque)
1885 if (!is_hyp_mode_available()) {
1886 kvm_info("HYP mode not available\n");
1890 in_hyp_mode = is_kernel_in_hyp_mode();
1892 if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1893 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1897 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
1898 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
1899 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
1900 "Only trusted guests should be used on this system.\n");
1902 for_each_online_cpu(cpu) {
1903 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1905 kvm_err("Error, CPU %d not supported!\n", cpu);
1910 err = init_common_resources();
1914 err = kvm_arm_init_sve();
1919 err = init_hyp_mode();
1924 err = kvm_init_vector_slots();
1926 kvm_err("Cannot initialise vector slots\n");
1930 err = init_subsystems();
1934 if (is_protected_kvm_enabled()) {
1935 static_branch_enable(&kvm_protected_mode_initialized);
1936 kvm_info("Protected nVHE mode initialized successfully\n");
1937 } else if (in_hyp_mode) {
1938 kvm_info("VHE mode initialized successfully\n");
1940 kvm_info("Hyp mode initialized successfully\n");
1948 teardown_hyp_mode();
1953 /* NOP: Compiling as a module not supported */
1954 void kvm_arch_exit(void)
1956 kvm_perf_teardown();
1959 static int __init early_kvm_mode_cfg(char *arg)
1964 if (strcmp(arg, "protected") == 0) {
1965 kvm_mode = KVM_MODE_PROTECTED;
1969 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode()))
1974 early_param("kvm-arm.mode", early_kvm_mode_cfg);
1976 enum kvm_mode kvm_get_mode(void)
1981 static int arm_init(void)
1983 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1987 module_init(arm_init);