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/entry-kvm.h>
10 #include <linux/errno.h>
11 #include <linux/err.h>
12 #include <linux/kvm_host.h>
13 #include <linux/list.h>
14 #include <linux/module.h>
15 #include <linux/vmalloc.h>
17 #include <linux/mman.h>
18 #include <linux/sched.h>
19 #include <linux/kmemleak.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_irqfd.h>
22 #include <linux/irqbypass.h>
23 #include <linux/sched/stat.h>
24 #include <linux/psci.h>
25 #include <trace/events/kvm.h>
27 #define CREATE_TRACE_POINTS
28 #include "trace_arm.h"
30 #include <linux/uaccess.h>
31 #include <asm/ptrace.h>
33 #include <asm/tlbflush.h>
34 #include <asm/cacheflush.h>
35 #include <asm/cpufeature.h>
37 #include <asm/kvm_arm.h>
38 #include <asm/kvm_asm.h>
39 #include <asm/kvm_mmu.h>
40 #include <asm/kvm_emulate.h>
41 #include <asm/sections.h>
43 #include <kvm/arm_hypercalls.h>
44 #include <kvm/arm_pmu.h>
45 #include <kvm/arm_psci.h>
47 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
48 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized);
50 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
56 /* The VMID used in the VTTBR */
57 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
58 static u32 kvm_next_vmid;
59 static DEFINE_SPINLOCK(kvm_vmid_lock);
61 static bool vgic_present;
63 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
64 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
68 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
71 int kvm_arch_hardware_setup(void *opaque)
76 int kvm_arch_check_processor_compat(void *opaque)
81 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
82 struct kvm_enable_cap *cap)
90 case KVM_CAP_ARM_NISV_TO_USER:
92 kvm->arch.return_nisv_io_abort_to_user = true;
95 mutex_lock(&kvm->lock);
96 if (!system_supports_mte() || kvm->created_vcpus) {
100 kvm->arch.mte_enabled = true;
102 mutex_unlock(&kvm->lock);
112 static int kvm_arm_default_max_vcpus(void)
114 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
117 static void set_default_spectre(struct kvm *kvm)
120 * The default is to expose CSV2 == 1 if the HW isn't affected.
121 * Although this is a per-CPU feature, we make it global because
122 * asymmetric systems are just a nuisance.
124 * Userspace can override this as long as it doesn't promise
127 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
128 kvm->arch.pfr0_csv2 = 1;
129 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED)
130 kvm->arch.pfr0_csv3 = 1;
134 * kvm_arch_init_vm - initializes a VM data structure
135 * @kvm: pointer to the KVM struct
137 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
141 ret = kvm_arm_setup_stage2(kvm, type);
145 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
149 ret = kvm_share_hyp(kvm, kvm + 1);
151 goto out_free_stage2_pgd;
153 kvm_vgic_early_init(kvm);
155 /* The maximum number of VCPUs is limited by the host's GIC model */
156 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
158 set_default_spectre(kvm);
162 kvm_free_stage2_pgd(&kvm->arch.mmu);
166 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
168 return VM_FAULT_SIGBUS;
173 * kvm_arch_destroy_vm - destroy the VM data structure
174 * @kvm: pointer to the KVM struct
176 void kvm_arch_destroy_vm(struct kvm *kvm)
178 bitmap_free(kvm->arch.pmu_filter);
180 kvm_vgic_destroy(kvm);
182 kvm_destroy_vcpus(kvm);
184 kvm_unshare_hyp(kvm, kvm + 1);
187 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
191 case KVM_CAP_IRQCHIP:
194 case KVM_CAP_IOEVENTFD:
195 case KVM_CAP_DEVICE_CTRL:
196 case KVM_CAP_USER_MEMORY:
197 case KVM_CAP_SYNC_MMU:
198 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
199 case KVM_CAP_ONE_REG:
200 case KVM_CAP_ARM_PSCI:
201 case KVM_CAP_ARM_PSCI_0_2:
202 case KVM_CAP_READONLY_MEM:
203 case KVM_CAP_MP_STATE:
204 case KVM_CAP_IMMEDIATE_EXIT:
205 case KVM_CAP_VCPU_EVENTS:
206 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
207 case KVM_CAP_ARM_NISV_TO_USER:
208 case KVM_CAP_ARM_INJECT_EXT_DABT:
209 case KVM_CAP_SET_GUEST_DEBUG:
210 case KVM_CAP_VCPU_ATTRIBUTES:
211 case KVM_CAP_PTP_KVM:
214 case KVM_CAP_SET_GUEST_DEBUG2:
215 return KVM_GUESTDBG_VALID_MASK;
216 case KVM_CAP_ARM_SET_DEVICE_ADDR:
219 case KVM_CAP_NR_VCPUS:
221 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
222 * architectures, as it does not always bound it to
223 * KVM_CAP_MAX_VCPUS. It should not matter much because
224 * this is just an advisory value.
226 r = min_t(unsigned int, num_online_cpus(),
227 kvm_arm_default_max_vcpus());
229 case KVM_CAP_MAX_VCPUS:
230 case KVM_CAP_MAX_VCPU_ID:
232 r = kvm->arch.max_vcpus;
234 r = kvm_arm_default_max_vcpus();
236 case KVM_CAP_MSI_DEVID:
240 r = kvm->arch.vgic.msis_require_devid;
242 case KVM_CAP_ARM_USER_IRQ:
244 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
245 * (bump this number if adding more devices)
249 case KVM_CAP_ARM_MTE:
250 r = system_supports_mte();
252 case KVM_CAP_STEAL_TIME:
253 r = kvm_arm_pvtime_supported();
255 case KVM_CAP_ARM_EL1_32BIT:
256 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
258 case KVM_CAP_GUEST_DEBUG_HW_BPS:
261 case KVM_CAP_GUEST_DEBUG_HW_WPS:
264 case KVM_CAP_ARM_PMU_V3:
265 r = kvm_arm_support_pmu_v3();
267 case KVM_CAP_ARM_INJECT_SERROR_ESR:
268 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
270 case KVM_CAP_ARM_VM_IPA_SIZE:
271 r = get_kvm_ipa_limit();
273 case KVM_CAP_ARM_SVE:
274 r = system_supports_sve();
276 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
277 case KVM_CAP_ARM_PTRAUTH_GENERIC:
278 r = system_has_full_ptr_auth();
287 long kvm_arch_dev_ioctl(struct file *filp,
288 unsigned int ioctl, unsigned long arg)
293 struct kvm *kvm_arch_alloc_vm(void)
295 size_t sz = sizeof(struct kvm);
298 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
300 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
303 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
305 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
308 if (id >= kvm->arch.max_vcpus)
314 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
318 /* Force users to call KVM_ARM_VCPU_INIT */
319 vcpu->arch.target = -1;
320 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
322 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
324 /* Set up the timer */
325 kvm_timer_vcpu_init(vcpu);
327 kvm_pmu_vcpu_init(vcpu);
329 kvm_arm_reset_debug_ptr(vcpu);
331 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
333 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
335 err = kvm_vgic_vcpu_init(vcpu);
339 return kvm_share_hyp(vcpu, vcpu + 1);
342 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
346 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
348 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm)))
349 static_branch_dec(&userspace_irqchip_in_use);
351 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
352 kvm_timer_vcpu_terminate(vcpu);
353 kvm_pmu_vcpu_destroy(vcpu);
355 kvm_arm_vcpu_destroy(vcpu);
358 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
360 return kvm_timer_is_pending(vcpu);
363 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
368 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
373 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
375 struct kvm_s2_mmu *mmu;
378 mmu = vcpu->arch.hw_mmu;
379 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
382 * We guarantee that both TLBs and I-cache are private to each
383 * vcpu. If detecting that a vcpu from the same VM has
384 * previously run on the same physical CPU, call into the
385 * hypervisor code to nuke the relevant contexts.
387 * We might get preempted before the vCPU actually runs, but
388 * over-invalidation doesn't affect correctness.
390 if (*last_ran != vcpu->vcpu_id) {
391 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
392 *last_ran = vcpu->vcpu_id;
398 kvm_timer_vcpu_load(vcpu);
400 kvm_vcpu_load_sysregs_vhe(vcpu);
401 kvm_arch_vcpu_load_fp(vcpu);
402 kvm_vcpu_pmu_restore_guest(vcpu);
403 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
404 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
406 if (single_task_running())
407 vcpu_clear_wfx_traps(vcpu);
409 vcpu_set_wfx_traps(vcpu);
411 if (vcpu_has_ptrauth(vcpu))
412 vcpu_ptrauth_disable(vcpu);
413 kvm_arch_vcpu_load_debug_state_flags(vcpu);
416 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
418 kvm_arch_vcpu_put_debug_state_flags(vcpu);
419 kvm_arch_vcpu_put_fp(vcpu);
421 kvm_vcpu_put_sysregs_vhe(vcpu);
422 kvm_timer_vcpu_put(vcpu);
424 kvm_vcpu_pmu_restore_host(vcpu);
429 static void vcpu_power_off(struct kvm_vcpu *vcpu)
431 vcpu->arch.power_off = true;
432 kvm_make_request(KVM_REQ_SLEEP, vcpu);
436 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
437 struct kvm_mp_state *mp_state)
439 if (vcpu->arch.power_off)
440 mp_state->mp_state = KVM_MP_STATE_STOPPED;
442 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
447 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
448 struct kvm_mp_state *mp_state)
452 switch (mp_state->mp_state) {
453 case KVM_MP_STATE_RUNNABLE:
454 vcpu->arch.power_off = false;
456 case KVM_MP_STATE_STOPPED:
457 vcpu_power_off(vcpu);
467 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
468 * @v: The VCPU pointer
470 * If the guest CPU is not waiting for interrupts or an interrupt line is
471 * asserted, the CPU is by definition runnable.
473 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
475 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
476 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
477 && !v->arch.power_off && !v->arch.pause);
480 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
482 return vcpu_mode_priv(vcpu);
485 #ifdef CONFIG_GUEST_PERF_EVENTS
486 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
488 return *vcpu_pc(vcpu);
492 /* Just ensure a guest exit from a particular CPU */
493 static void exit_vm_noop(void *info)
497 void force_vm_exit(const cpumask_t *mask)
500 smp_call_function_many(mask, exit_vm_noop, NULL, true);
505 * need_new_vmid_gen - check that the VMID is still valid
506 * @vmid: The VMID to check
508 * return true if there is a new generation of VMIDs being used
510 * The hardware supports a limited set of values with the value zero reserved
511 * for the host, so we check if an assigned value belongs to a previous
512 * generation, which requires us to assign a new value. If we're the first to
513 * use a VMID for the new generation, we must flush necessary caches and TLBs
516 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
518 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
519 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
520 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
524 * update_vmid - Update the vmid with a valid VMID for the current generation
525 * @vmid: The stage-2 VMID information struct
527 static void update_vmid(struct kvm_vmid *vmid)
529 if (!need_new_vmid_gen(vmid))
532 spin_lock(&kvm_vmid_lock);
535 * We need to re-check the vmid_gen here to ensure that if another vcpu
536 * already allocated a valid vmid for this vm, then this vcpu should
539 if (!need_new_vmid_gen(vmid)) {
540 spin_unlock(&kvm_vmid_lock);
544 /* First user of a new VMID generation? */
545 if (unlikely(kvm_next_vmid == 0)) {
546 atomic64_inc(&kvm_vmid_gen);
550 * On SMP we know no other CPUs can use this CPU's or each
551 * other's VMID after force_vm_exit returns since the
552 * kvm_vmid_lock blocks them from reentry to the guest.
554 force_vm_exit(cpu_all_mask);
556 * Now broadcast TLB + ICACHE invalidation over the inner
557 * shareable domain to make sure all data structures are
560 kvm_call_hyp(__kvm_flush_vm_context);
563 WRITE_ONCE(vmid->vmid, kvm_next_vmid);
565 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
568 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
570 spin_unlock(&kvm_vmid_lock);
573 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
575 return vcpu->arch.target >= 0;
579 * Handle both the initialisation that is being done when the vcpu is
580 * run for the first time, as well as the updates that must be
581 * performed each time we get a new thread dealing with this vcpu.
583 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
585 struct kvm *kvm = vcpu->kvm;
588 if (!kvm_vcpu_initialized(vcpu))
591 if (!kvm_arm_vcpu_is_finalized(vcpu))
594 ret = kvm_arch_vcpu_run_map_fp(vcpu);
598 if (likely(vcpu_has_run_once(vcpu)))
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 ret = kvm_timer_enable(vcpu);
617 ret = kvm_arm_pmu_v3_enable(vcpu);
621 if (!irqchip_in_kernel(kvm)) {
623 * Tell the rest of the code that there are userspace irqchip
626 static_branch_inc(&userspace_irqchip_in_use);
630 * Initialize traps for protected VMs.
631 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
632 * the code is in place for first run initialization at EL2.
634 if (kvm_vm_is_protected(kvm))
635 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
640 bool kvm_arch_intc_initialized(struct kvm *kvm)
642 return vgic_initialized(kvm);
645 void kvm_arm_halt_guest(struct kvm *kvm)
648 struct kvm_vcpu *vcpu;
650 kvm_for_each_vcpu(i, vcpu, kvm)
651 vcpu->arch.pause = true;
652 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
655 void kvm_arm_resume_guest(struct kvm *kvm)
658 struct kvm_vcpu *vcpu;
660 kvm_for_each_vcpu(i, vcpu, kvm) {
661 vcpu->arch.pause = false;
662 __kvm_vcpu_wake_up(vcpu);
666 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
668 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
670 rcuwait_wait_event(wait,
671 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
674 if (vcpu->arch.power_off || vcpu->arch.pause) {
675 /* Awaken to handle a signal, request we sleep again later. */
676 kvm_make_request(KVM_REQ_SLEEP, vcpu);
680 * Make sure we will observe a potential reset request if we've
681 * observed a change to the power state. Pairs with the smp_wmb() in
682 * kvm_psci_vcpu_on().
688 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
689 * @vcpu: The VCPU pointer
691 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
692 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending
693 * on when a wake event arrives, e.g. there may already be a pending wake event.
695 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
698 * Sync back the state of the GIC CPU interface so that we have
699 * the latest PMR and group enables. This ensures that
700 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
701 * we have pending interrupts, e.g. when determining if the
704 * For the same reason, we want to tell GICv4 that we need
705 * doorbells to be signalled, should an interrupt become pending.
708 kvm_vgic_vmcr_sync(vcpu);
709 vgic_v4_put(vcpu, true);
713 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
720 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
722 if (kvm_request_pending(vcpu)) {
723 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
724 vcpu_req_sleep(vcpu);
726 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
727 kvm_reset_vcpu(vcpu);
730 * Clear IRQ_PENDING requests that were made to guarantee
731 * that a VCPU sees new virtual interrupts.
733 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
735 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
736 kvm_update_stolen_time(vcpu);
738 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
739 /* The distributor enable bits were changed */
741 vgic_v4_put(vcpu, false);
746 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
747 kvm_pmu_handle_pmcr(vcpu,
748 __vcpu_sys_reg(vcpu, PMCR_EL0));
752 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
754 if (likely(!vcpu_mode_is_32bit(vcpu)))
757 return !system_supports_32bit_el0() ||
758 static_branch_unlikely(&arm64_mismatched_32bit_el0);
762 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
763 * @vcpu: The VCPU pointer
764 * @ret: Pointer to write optional return code
766 * Returns: true if the VCPU needs to return to a preemptible + interruptible
767 * and skip guest entry.
769 * This function disambiguates between two different types of exits: exits to a
770 * preemptible + interruptible kernel context and exits to userspace. For an
771 * exit to userspace, this function will write the return code to ret and return
772 * true. For an exit to preemptible + interruptible kernel context (i.e. check
773 * for pending work and re-enter), return true without writing to ret.
775 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
777 struct kvm_run *run = vcpu->run;
780 * If we're using a userspace irqchip, then check if we need
781 * to tell a userspace irqchip about timer or PMU level
782 * changes and if so, exit to userspace (the actual level
783 * state gets updated in kvm_timer_update_run and
784 * kvm_pmu_update_run below).
786 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
787 if (kvm_timer_should_notify_user(vcpu) ||
788 kvm_pmu_should_notify_user(vcpu)) {
790 run->exit_reason = KVM_EXIT_INTR;
795 return kvm_request_pending(vcpu) ||
796 need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
797 xfer_to_guest_mode_work_pending();
801 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
802 * the vCPU is running.
804 * This must be noinstr as instrumentation may make use of RCU, and this is not
805 * safe during the EQS.
807 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
811 guest_state_enter_irqoff();
812 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
813 guest_state_exit_irqoff();
819 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
820 * @vcpu: The VCPU pointer
822 * This function is called through the VCPU_RUN ioctl called from user space. It
823 * will execute VM code in a loop until the time slice for the process is used
824 * or some emulation is needed from user space in which case the function will
825 * return with return value 0 and with the kvm_run structure filled in with the
826 * required data for the requested emulation.
828 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
830 struct kvm_run *run = vcpu->run;
833 if (run->exit_reason == KVM_EXIT_MMIO) {
834 ret = kvm_handle_mmio_return(vcpu);
841 if (run->immediate_exit) {
846 kvm_sigset_activate(vcpu);
849 run->exit_reason = KVM_EXIT_UNKNOWN;
852 * Check conditions before entering the guest
854 ret = xfer_to_guest_mode_handle_work(vcpu);
858 update_vmid(&vcpu->arch.hw_mmu->vmid);
860 check_vcpu_requests(vcpu);
863 * Preparing the interrupts to be injected also
864 * involves poking the GIC, which must be done in a
865 * non-preemptible context.
869 kvm_pmu_flush_hwstate(vcpu);
873 kvm_vgic_flush_hwstate(vcpu);
876 * Ensure we set mode to IN_GUEST_MODE after we disable
877 * interrupts and before the final VCPU requests check.
878 * See the comment in kvm_vcpu_exiting_guest_mode() and
879 * Documentation/virt/kvm/vcpu-requests.rst
881 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
883 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
884 vcpu->mode = OUTSIDE_GUEST_MODE;
885 isb(); /* Ensure work in x_flush_hwstate is committed */
886 kvm_pmu_sync_hwstate(vcpu);
887 if (static_branch_unlikely(&userspace_irqchip_in_use))
888 kvm_timer_sync_user(vcpu);
889 kvm_vgic_sync_hwstate(vcpu);
895 kvm_arm_setup_debug(vcpu);
896 kvm_arch_vcpu_ctxflush_fp(vcpu);
898 /**************************************************************
901 trace_kvm_entry(*vcpu_pc(vcpu));
902 guest_timing_enter_irqoff();
904 ret = kvm_arm_vcpu_enter_exit(vcpu);
906 vcpu->mode = OUTSIDE_GUEST_MODE;
910 *************************************************************/
912 kvm_arm_clear_debug(vcpu);
915 * We must sync the PMU state before the vgic state so
916 * that the vgic can properly sample the updated state of the
919 kvm_pmu_sync_hwstate(vcpu);
922 * Sync the vgic state before syncing the timer state because
923 * the timer code needs to know if the virtual timer
924 * interrupts are active.
926 kvm_vgic_sync_hwstate(vcpu);
929 * Sync the timer hardware state before enabling interrupts as
930 * we don't want vtimer interrupts to race with syncing the
931 * timer virtual interrupt state.
933 if (static_branch_unlikely(&userspace_irqchip_in_use))
934 kvm_timer_sync_user(vcpu);
936 kvm_arch_vcpu_ctxsync_fp(vcpu);
939 * We must ensure that any pending interrupts are taken before
940 * we exit guest timing so that timer ticks are accounted as
941 * guest time. Transiently unmask interrupts so that any
942 * pending interrupts are taken.
944 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
945 * context synchronization event) is necessary to ensure that
946 * pending interrupts are taken.
952 guest_timing_exit_irqoff();
956 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
958 /* Exit types that need handling before we can be preempted */
959 handle_exit_early(vcpu, ret);
964 * The ARMv8 architecture doesn't give the hypervisor
965 * a mechanism to prevent a guest from dropping to AArch32 EL0
966 * if implemented by the CPU. If we spot the guest in such
967 * state and that we decided it wasn't supposed to do so (like
968 * with the asymmetric AArch32 case), return to userspace with
971 if (vcpu_mode_is_bad_32bit(vcpu)) {
973 * As we have caught the guest red-handed, decide that
974 * it isn't fit for purpose anymore by making the vcpu
975 * invalid. The VMM can try and fix it by issuing a
976 * KVM_ARM_VCPU_INIT if it really wants to.
978 vcpu->arch.target = -1;
979 ret = ARM_EXCEPTION_IL;
982 ret = handle_exit(vcpu, ret);
985 /* Tell userspace about in-kernel device output levels */
986 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
987 kvm_timer_update_run(vcpu);
988 kvm_pmu_update_run(vcpu);
991 kvm_sigset_deactivate(vcpu);
995 * In the unlikely event that we are returning to userspace
996 * with pending exceptions or PC adjustment, commit these
997 * adjustments in order to give userspace a consistent view of
998 * the vcpu state. Note that this relies on __kvm_adjust_pc()
999 * being preempt-safe on VHE.
1001 if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
1002 KVM_ARM64_INCREMENT_PC)))
1003 kvm_call_hyp(__kvm_adjust_pc, vcpu);
1009 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1015 if (number == KVM_ARM_IRQ_CPU_IRQ)
1016 bit_index = __ffs(HCR_VI);
1017 else /* KVM_ARM_IRQ_CPU_FIQ */
1018 bit_index = __ffs(HCR_VF);
1020 hcr = vcpu_hcr(vcpu);
1022 set = test_and_set_bit(bit_index, hcr);
1024 set = test_and_clear_bit(bit_index, hcr);
1027 * If we didn't change anything, no need to wake up or kick other CPUs
1033 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1034 * trigger a world-switch round on the running physical CPU to set the
1035 * virtual IRQ/FIQ fields in the HCR appropriately.
1037 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1038 kvm_vcpu_kick(vcpu);
1043 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1046 u32 irq = irq_level->irq;
1047 unsigned int irq_type, vcpu_idx, irq_num;
1048 int nrcpus = atomic_read(&kvm->online_vcpus);
1049 struct kvm_vcpu *vcpu = NULL;
1050 bool level = irq_level->level;
1052 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1053 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1054 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1055 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1057 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1060 case KVM_ARM_IRQ_TYPE_CPU:
1061 if (irqchip_in_kernel(kvm))
1064 if (vcpu_idx >= nrcpus)
1067 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1071 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1074 return vcpu_interrupt_line(vcpu, irq_num, level);
1075 case KVM_ARM_IRQ_TYPE_PPI:
1076 if (!irqchip_in_kernel(kvm))
1079 if (vcpu_idx >= nrcpus)
1082 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1086 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1089 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1090 case KVM_ARM_IRQ_TYPE_SPI:
1091 if (!irqchip_in_kernel(kvm))
1094 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1097 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1103 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1104 const struct kvm_vcpu_init *init)
1106 unsigned int i, ret;
1107 u32 phys_target = kvm_target_cpu();
1109 if (init->target != phys_target)
1113 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1114 * use the same target.
1116 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1119 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1120 for (i = 0; i < sizeof(init->features) * 8; i++) {
1121 bool set = (init->features[i / 32] & (1 << (i % 32)));
1123 if (set && i >= KVM_VCPU_MAX_FEATURES)
1127 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1128 * use the same feature set.
1130 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1131 test_bit(i, vcpu->arch.features) != set)
1135 set_bit(i, vcpu->arch.features);
1138 vcpu->arch.target = phys_target;
1140 /* Now we know what it is, we can reset it. */
1141 ret = kvm_reset_vcpu(vcpu);
1143 vcpu->arch.target = -1;
1144 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1150 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1151 struct kvm_vcpu_init *init)
1155 ret = kvm_vcpu_set_target(vcpu, init);
1160 * Ensure a rebooted VM will fault in RAM pages and detect if the
1161 * guest MMU is turned off and flush the caches as needed.
1163 * S2FWB enforces all memory accesses to RAM being cacheable,
1164 * ensuring that the data side is always coherent. We still
1165 * need to invalidate the I-cache though, as FWB does *not*
1166 * imply CTR_EL0.DIC.
1168 if (vcpu_has_run_once(vcpu)) {
1169 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1170 stage2_unmap_vm(vcpu->kvm);
1172 icache_inval_all_pou();
1175 vcpu_reset_hcr(vcpu);
1176 vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1179 * Handle the "start in power-off" case.
1181 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1182 vcpu_power_off(vcpu);
1184 vcpu->arch.power_off = false;
1189 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1190 struct kvm_device_attr *attr)
1194 switch (attr->group) {
1196 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1203 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1204 struct kvm_device_attr *attr)
1208 switch (attr->group) {
1210 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1217 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1218 struct kvm_device_attr *attr)
1222 switch (attr->group) {
1224 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1231 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1232 struct kvm_vcpu_events *events)
1234 memset(events, 0, sizeof(*events));
1236 return __kvm_arm_vcpu_get_events(vcpu, events);
1239 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1240 struct kvm_vcpu_events *events)
1244 /* check whether the reserved field is zero */
1245 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1246 if (events->reserved[i])
1249 /* check whether the pad field is zero */
1250 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1251 if (events->exception.pad[i])
1254 return __kvm_arm_vcpu_set_events(vcpu, events);
1257 long kvm_arch_vcpu_ioctl(struct file *filp,
1258 unsigned int ioctl, unsigned long arg)
1260 struct kvm_vcpu *vcpu = filp->private_data;
1261 void __user *argp = (void __user *)arg;
1262 struct kvm_device_attr attr;
1266 case KVM_ARM_VCPU_INIT: {
1267 struct kvm_vcpu_init init;
1270 if (copy_from_user(&init, argp, sizeof(init)))
1273 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1276 case KVM_SET_ONE_REG:
1277 case KVM_GET_ONE_REG: {
1278 struct kvm_one_reg reg;
1281 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1285 if (copy_from_user(®, argp, sizeof(reg)))
1289 * We could owe a reset due to PSCI. Handle the pending reset
1290 * here to ensure userspace register accesses are ordered after
1293 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1294 kvm_reset_vcpu(vcpu);
1296 if (ioctl == KVM_SET_ONE_REG)
1297 r = kvm_arm_set_reg(vcpu, ®);
1299 r = kvm_arm_get_reg(vcpu, ®);
1302 case KVM_GET_REG_LIST: {
1303 struct kvm_reg_list __user *user_list = argp;
1304 struct kvm_reg_list reg_list;
1308 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1312 if (!kvm_arm_vcpu_is_finalized(vcpu))
1316 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1319 reg_list.n = kvm_arm_num_regs(vcpu);
1320 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1325 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1328 case KVM_SET_DEVICE_ATTR: {
1330 if (copy_from_user(&attr, argp, sizeof(attr)))
1332 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1335 case KVM_GET_DEVICE_ATTR: {
1337 if (copy_from_user(&attr, argp, sizeof(attr)))
1339 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1342 case KVM_HAS_DEVICE_ATTR: {
1344 if (copy_from_user(&attr, argp, sizeof(attr)))
1346 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1349 case KVM_GET_VCPU_EVENTS: {
1350 struct kvm_vcpu_events events;
1352 if (kvm_arm_vcpu_get_events(vcpu, &events))
1355 if (copy_to_user(argp, &events, sizeof(events)))
1360 case KVM_SET_VCPU_EVENTS: {
1361 struct kvm_vcpu_events events;
1363 if (copy_from_user(&events, argp, sizeof(events)))
1366 return kvm_arm_vcpu_set_events(vcpu, &events);
1368 case KVM_ARM_VCPU_FINALIZE: {
1371 if (!kvm_vcpu_initialized(vcpu))
1374 if (get_user(what, (const int __user *)argp))
1377 return kvm_arm_vcpu_finalize(vcpu, what);
1386 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1391 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1392 const struct kvm_memory_slot *memslot)
1394 kvm_flush_remote_tlbs(kvm);
1397 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1398 struct kvm_arm_device_addr *dev_addr)
1400 unsigned long dev_id, type;
1402 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1403 KVM_ARM_DEVICE_ID_SHIFT;
1404 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1405 KVM_ARM_DEVICE_TYPE_SHIFT;
1408 case KVM_ARM_DEVICE_VGIC_V2:
1411 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1417 long kvm_arch_vm_ioctl(struct file *filp,
1418 unsigned int ioctl, unsigned long arg)
1420 struct kvm *kvm = filp->private_data;
1421 void __user *argp = (void __user *)arg;
1424 case KVM_CREATE_IRQCHIP: {
1428 mutex_lock(&kvm->lock);
1429 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1430 mutex_unlock(&kvm->lock);
1433 case KVM_ARM_SET_DEVICE_ADDR: {
1434 struct kvm_arm_device_addr dev_addr;
1436 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1438 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1440 case KVM_ARM_PREFERRED_TARGET: {
1441 struct kvm_vcpu_init init;
1443 kvm_vcpu_preferred_target(&init);
1445 if (copy_to_user(argp, &init, sizeof(init)))
1450 case KVM_ARM_MTE_COPY_TAGS: {
1451 struct kvm_arm_copy_mte_tags copy_tags;
1453 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1455 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1462 static unsigned long nvhe_percpu_size(void)
1464 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1465 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1468 static unsigned long nvhe_percpu_order(void)
1470 unsigned long size = nvhe_percpu_size();
1472 return size ? get_order(size) : 0;
1475 /* A lookup table holding the hypervisor VA for each vector slot */
1476 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1478 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1480 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1483 static int kvm_init_vector_slots(void)
1488 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1489 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1491 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1492 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1494 if (kvm_system_needs_idmapped_vectors() && !has_vhe()) {
1495 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1496 __BP_HARDEN_HYP_VECS_SZ, &base);
1501 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1502 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1506 static void cpu_prepare_hyp_mode(int cpu)
1508 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1512 * Calculate the raw per-cpu offset without a translation from the
1513 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1514 * so that we can use adr_l to access per-cpu variables in EL2.
1515 * Also drop the KASAN tag which gets in the way...
1517 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1518 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1520 params->mair_el2 = read_sysreg(mair_el1);
1523 * The ID map may be configured to use an extended virtual address
1524 * range. This is only the case if system RAM is out of range for the
1525 * currently configured page size and VA_BITS, in which case we will
1526 * also need the extended virtual range for the HYP ID map, or we won't
1527 * be able to enable the EL2 MMU.
1529 * However, at EL2, there is only one TTBR register, and we can't switch
1530 * between translation tables *and* update TCR_EL2.T0SZ at the same
1531 * time. Bottom line: we need to use the extended range with *both* our
1532 * translation tables.
1534 * So use the same T0SZ value we use for the ID map.
1536 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1537 tcr &= ~TCR_T0SZ_MASK;
1538 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1539 params->tcr_el2 = tcr;
1541 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1542 params->pgd_pa = kvm_mmu_get_httbr();
1543 if (is_protected_kvm_enabled())
1544 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1546 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1547 params->vttbr = params->vtcr = 0;
1550 * Flush the init params from the data cache because the struct will
1551 * be read while the MMU is off.
1553 kvm_flush_dcache_to_poc(params, sizeof(*params));
1556 static void hyp_install_host_vector(void)
1558 struct kvm_nvhe_init_params *params;
1559 struct arm_smccc_res res;
1561 /* Switch from the HYP stub to our own HYP init vector */
1562 __hyp_set_vectors(kvm_get_idmap_vector());
1565 * Call initialization code, and switch to the full blown HYP code.
1566 * If the cpucaps haven't been finalized yet, something has gone very
1567 * wrong, and hyp will crash and burn when it uses any
1568 * cpus_have_const_cap() wrapper.
1570 BUG_ON(!system_capabilities_finalized());
1571 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1572 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1573 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1576 static void cpu_init_hyp_mode(void)
1578 hyp_install_host_vector();
1581 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1584 if (this_cpu_has_cap(ARM64_SSBS) &&
1585 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1586 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1590 static void cpu_hyp_reset(void)
1592 if (!is_kernel_in_hyp_mode())
1593 __hyp_reset_vectors();
1597 * EL2 vectors can be mapped and rerouted in a number of ways,
1598 * depending on the kernel configuration and CPU present:
1600 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1601 * placed in one of the vector slots, which is executed before jumping
1602 * to the real vectors.
1604 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1605 * containing the hardening sequence is mapped next to the idmap page,
1606 * and executed before jumping to the real vectors.
1608 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1609 * empty slot is selected, mapped next to the idmap page, and
1610 * executed before jumping to the real vectors.
1612 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1613 * VHE, as we don't have hypervisor-specific mappings. If the system
1614 * is VHE and yet selects this capability, it will be ignored.
1616 static void cpu_set_hyp_vector(void)
1618 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1619 void *vector = hyp_spectre_vector_selector[data->slot];
1621 if (!is_protected_kvm_enabled())
1622 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1624 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1627 static void cpu_hyp_init_context(void)
1629 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1631 if (!is_kernel_in_hyp_mode())
1632 cpu_init_hyp_mode();
1635 static void cpu_hyp_init_features(void)
1637 cpu_set_hyp_vector();
1638 kvm_arm_init_debug();
1640 if (is_kernel_in_hyp_mode())
1641 kvm_timer_init_vhe();
1644 kvm_vgic_init_cpu_hardware();
1647 static void cpu_hyp_reinit(void)
1650 cpu_hyp_init_context();
1651 cpu_hyp_init_features();
1654 static void _kvm_arch_hardware_enable(void *discard)
1656 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1658 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1662 int kvm_arch_hardware_enable(void)
1664 _kvm_arch_hardware_enable(NULL);
1668 static void _kvm_arch_hardware_disable(void *discard)
1670 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1672 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1676 void kvm_arch_hardware_disable(void)
1678 if (!is_protected_kvm_enabled())
1679 _kvm_arch_hardware_disable(NULL);
1682 #ifdef CONFIG_CPU_PM
1683 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1688 * kvm_arm_hardware_enabled is left with its old value over
1689 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1694 if (__this_cpu_read(kvm_arm_hardware_enabled))
1696 * don't update kvm_arm_hardware_enabled here
1697 * so that the hardware will be re-enabled
1698 * when we resume. See below.
1703 case CPU_PM_ENTER_FAILED:
1705 if (__this_cpu_read(kvm_arm_hardware_enabled))
1706 /* The hardware was enabled before suspend. */
1716 static struct notifier_block hyp_init_cpu_pm_nb = {
1717 .notifier_call = hyp_init_cpu_pm_notifier,
1720 static void hyp_cpu_pm_init(void)
1722 if (!is_protected_kvm_enabled())
1723 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1725 static void hyp_cpu_pm_exit(void)
1727 if (!is_protected_kvm_enabled())
1728 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1731 static inline void hyp_cpu_pm_init(void)
1734 static inline void hyp_cpu_pm_exit(void)
1739 static void init_cpu_logical_map(void)
1744 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1745 * Only copy the set of online CPUs whose features have been chacked
1746 * against the finalized system capabilities. The hypervisor will not
1747 * allow any other CPUs from the `possible` set to boot.
1749 for_each_online_cpu(cpu)
1750 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1753 #define init_psci_0_1_impl_state(config, what) \
1754 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1756 static bool init_psci_relay(void)
1759 * If PSCI has not been initialized, protected KVM cannot install
1760 * itself on newly booted CPUs.
1762 if (!psci_ops.get_version) {
1763 kvm_err("Cannot initialize protected mode without PSCI\n");
1767 kvm_host_psci_config.version = psci_ops.get_version();
1769 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1770 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1771 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1772 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1773 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1774 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1779 static int init_subsystems(void)
1784 * Enable hardware so that subsystem initialisation can access EL2.
1786 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1789 * Register CPU lower-power notifier
1794 * Init HYP view of VGIC
1796 err = kvm_vgic_hyp_init();
1799 vgic_present = true;
1803 vgic_present = false;
1811 * Init HYP architected timer support
1813 err = kvm_timer_hyp_init(vgic_present);
1817 kvm_register_perf_callbacks(NULL);
1819 kvm_sys_reg_table_init();
1822 if (err || !is_protected_kvm_enabled())
1823 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1828 static void teardown_hyp_mode(void)
1833 for_each_possible_cpu(cpu) {
1834 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1835 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1839 static int do_pkvm_init(u32 hyp_va_bits)
1841 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1845 cpu_hyp_init_context();
1846 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1847 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1849 cpu_hyp_init_features();
1852 * The stub hypercalls are now disabled, so set our local flag to
1853 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1855 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1861 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1863 void *addr = phys_to_virt(hyp_mem_base);
1866 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1867 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1868 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1869 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1870 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
1871 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1872 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1873 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1875 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1879 ret = do_pkvm_init(hyp_va_bits);
1889 * Inits Hyp-mode on all online CPUs
1891 static int init_hyp_mode(void)
1898 * The protected Hyp-mode cannot be initialized if the memory pool
1899 * allocation has failed.
1901 if (is_protected_kvm_enabled() && !hyp_mem_base)
1905 * Allocate Hyp PGD and setup Hyp identity mapping
1907 err = kvm_mmu_init(&hyp_va_bits);
1912 * Allocate stack pages for Hypervisor-mode
1914 for_each_possible_cpu(cpu) {
1915 unsigned long stack_page;
1917 stack_page = __get_free_page(GFP_KERNEL);
1923 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1927 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1929 for_each_possible_cpu(cpu) {
1933 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1939 page_addr = page_address(page);
1940 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1941 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1945 * Map the Hyp-code called directly from the host
1947 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1948 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1950 kvm_err("Cannot map world-switch code\n");
1954 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1955 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1957 kvm_err("Cannot map .hyp.rodata section\n");
1961 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1962 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1964 kvm_err("Cannot map rodata section\n");
1969 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1970 * section thanks to an assertion in the linker script. Map it RW and
1971 * the rest of .bss RO.
1973 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1974 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1976 kvm_err("Cannot map hyp bss section: %d\n", err);
1980 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1981 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1983 kvm_err("Cannot map bss section\n");
1988 * Map the Hyp stack pages
1990 for_each_possible_cpu(cpu) {
1991 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1992 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1996 kvm_err("Cannot map hyp stack\n");
2001 for_each_possible_cpu(cpu) {
2002 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
2003 char *percpu_end = percpu_begin + nvhe_percpu_size();
2005 /* Map Hyp percpu pages */
2006 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2008 kvm_err("Cannot map hyp percpu region\n");
2012 /* Prepare the CPU initialization parameters */
2013 cpu_prepare_hyp_mode(cpu);
2016 if (is_protected_kvm_enabled()) {
2017 init_cpu_logical_map();
2019 if (!init_psci_relay()) {
2025 if (is_protected_kvm_enabled()) {
2026 err = kvm_hyp_init_protection(hyp_va_bits);
2028 kvm_err("Failed to init hyp memory protection\n");
2036 teardown_hyp_mode();
2037 kvm_err("error initializing Hyp mode: %d\n", err);
2041 static void _kvm_host_prot_finalize(void *arg)
2045 if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
2046 WRITE_ONCE(*err, -EINVAL);
2049 static int pkvm_drop_host_privileges(void)
2054 * Flip the static key upfront as that may no longer be possible
2055 * once the host stage 2 is installed.
2057 static_branch_enable(&kvm_protected_mode_initialized);
2058 on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2062 static int finalize_hyp_mode(void)
2064 if (!is_protected_kvm_enabled())
2068 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2069 * at, which would end badly once the section is inaccessible.
2070 * None of other sections should ever be introspected.
2072 kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2073 return pkvm_drop_host_privileges();
2076 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2078 struct kvm_vcpu *vcpu;
2081 mpidr &= MPIDR_HWID_BITMASK;
2082 kvm_for_each_vcpu(i, vcpu, kvm) {
2083 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2089 bool kvm_arch_has_irq_bypass(void)
2094 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2095 struct irq_bypass_producer *prod)
2097 struct kvm_kernel_irqfd *irqfd =
2098 container_of(cons, struct kvm_kernel_irqfd, consumer);
2100 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2103 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2104 struct irq_bypass_producer *prod)
2106 struct kvm_kernel_irqfd *irqfd =
2107 container_of(cons, struct kvm_kernel_irqfd, consumer);
2109 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2113 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2115 struct kvm_kernel_irqfd *irqfd =
2116 container_of(cons, struct kvm_kernel_irqfd, consumer);
2118 kvm_arm_halt_guest(irqfd->kvm);
2121 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2123 struct kvm_kernel_irqfd *irqfd =
2124 container_of(cons, struct kvm_kernel_irqfd, consumer);
2126 kvm_arm_resume_guest(irqfd->kvm);
2130 * Initialize Hyp-mode and memory mappings on all CPUs.
2132 int kvm_arch_init(void *opaque)
2137 if (!is_hyp_mode_available()) {
2138 kvm_info("HYP mode not available\n");
2142 if (kvm_get_mode() == KVM_MODE_NONE) {
2143 kvm_info("KVM disabled from command line\n");
2147 in_hyp_mode = is_kernel_in_hyp_mode();
2149 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2150 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2151 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2152 "Only trusted guests should be used on this system.\n");
2154 err = kvm_set_ipa_limit();
2158 err = kvm_arm_init_sve();
2163 err = init_hyp_mode();
2168 err = kvm_init_vector_slots();
2170 kvm_err("Cannot initialise vector slots\n");
2174 err = init_subsystems();
2179 err = finalize_hyp_mode();
2181 kvm_err("Failed to finalize Hyp protection\n");
2186 if (is_protected_kvm_enabled()) {
2187 kvm_info("Protected nVHE mode initialized successfully\n");
2188 } else if (in_hyp_mode) {
2189 kvm_info("VHE mode initialized successfully\n");
2191 kvm_info("Hyp mode initialized successfully\n");
2199 teardown_hyp_mode();
2204 /* NOP: Compiling as a module not supported */
2205 void kvm_arch_exit(void)
2207 kvm_unregister_perf_callbacks();
2210 static int __init early_kvm_mode_cfg(char *arg)
2215 if (strcmp(arg, "protected") == 0) {
2216 kvm_mode = KVM_MODE_PROTECTED;
2220 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2221 kvm_mode = KVM_MODE_DEFAULT;
2225 if (strcmp(arg, "none") == 0) {
2226 kvm_mode = KVM_MODE_NONE;
2232 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2234 enum kvm_mode kvm_get_mode(void)
2239 static int arm_init(void)
2241 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2245 module_init(arm_init);