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 = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
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)
180 bitmap_free(kvm->arch.pmu_filter);
182 kvm_vgic_destroy(kvm);
184 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
186 kvm_vcpu_destroy(kvm->vcpus[i]);
187 kvm->vcpus[i] = NULL;
190 atomic_set(&kvm->online_vcpus, 0);
193 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
197 case KVM_CAP_IRQCHIP:
200 case KVM_CAP_IOEVENTFD:
201 case KVM_CAP_DEVICE_CTRL:
202 case KVM_CAP_USER_MEMORY:
203 case KVM_CAP_SYNC_MMU:
204 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
205 case KVM_CAP_ONE_REG:
206 case KVM_CAP_ARM_PSCI:
207 case KVM_CAP_ARM_PSCI_0_2:
208 case KVM_CAP_READONLY_MEM:
209 case KVM_CAP_MP_STATE:
210 case KVM_CAP_IMMEDIATE_EXIT:
211 case KVM_CAP_VCPU_EVENTS:
212 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
213 case KVM_CAP_ARM_NISV_TO_USER:
214 case KVM_CAP_ARM_INJECT_EXT_DABT:
215 case KVM_CAP_SET_GUEST_DEBUG:
216 case KVM_CAP_VCPU_ATTRIBUTES:
217 case KVM_CAP_PTP_KVM:
220 case KVM_CAP_SET_GUEST_DEBUG2:
221 return KVM_GUESTDBG_VALID_MASK;
222 case KVM_CAP_ARM_SET_DEVICE_ADDR:
225 case KVM_CAP_NR_VCPUS:
226 r = num_online_cpus();
228 case KVM_CAP_MAX_VCPUS:
229 case KVM_CAP_MAX_VCPU_ID:
231 r = kvm->arch.max_vcpus;
233 r = kvm_arm_default_max_vcpus();
235 case KVM_CAP_MSI_DEVID:
239 r = kvm->arch.vgic.msis_require_devid;
241 case KVM_CAP_ARM_USER_IRQ:
243 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
244 * (bump this number if adding more devices)
248 case KVM_CAP_ARM_MTE:
249 r = system_supports_mte();
251 case KVM_CAP_STEAL_TIME:
252 r = kvm_arm_pvtime_supported();
254 case KVM_CAP_ARM_EL1_32BIT:
255 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1);
257 case KVM_CAP_GUEST_DEBUG_HW_BPS:
260 case KVM_CAP_GUEST_DEBUG_HW_WPS:
263 case KVM_CAP_ARM_PMU_V3:
264 r = kvm_arm_support_pmu_v3();
266 case KVM_CAP_ARM_INJECT_SERROR_ESR:
267 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN);
269 case KVM_CAP_ARM_VM_IPA_SIZE:
270 r = get_kvm_ipa_limit();
272 case KVM_CAP_ARM_SVE:
273 r = system_supports_sve();
275 case KVM_CAP_ARM_PTRAUTH_ADDRESS:
276 case KVM_CAP_ARM_PTRAUTH_GENERIC:
277 r = system_has_full_ptr_auth();
286 long kvm_arch_dev_ioctl(struct file *filp,
287 unsigned int ioctl, unsigned long arg)
292 struct kvm *kvm_arch_alloc_vm(void)
294 size_t sz = sizeof(struct kvm);
297 return kzalloc(sz, GFP_KERNEL_ACCOUNT);
299 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
302 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
304 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
307 if (id >= kvm->arch.max_vcpus)
313 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
317 /* Force users to call KVM_ARM_VCPU_INIT */
318 vcpu->arch.target = -1;
319 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
321 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
323 /* Set up the timer */
324 kvm_timer_vcpu_init(vcpu);
326 kvm_pmu_vcpu_init(vcpu);
328 kvm_arm_reset_debug_ptr(vcpu);
330 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
332 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
334 err = kvm_vgic_vcpu_init(vcpu);
338 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
341 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
345 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
347 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
348 static_branch_dec(&userspace_irqchip_in_use);
350 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
351 kvm_timer_vcpu_terminate(vcpu);
352 kvm_pmu_vcpu_destroy(vcpu);
354 kvm_arm_vcpu_destroy(vcpu);
357 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
359 return kvm_timer_is_pending(vcpu);
362 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
365 * If we're about to block (most likely because we've just hit a
366 * WFI), we need to sync back the state of the GIC CPU interface
367 * so that we have the latest PMR and group enables. This ensures
368 * that kvm_arch_vcpu_runnable has up-to-date data to decide
369 * whether we have pending interrupts.
371 * For the same reason, we want to tell GICv4 that we need
372 * doorbells to be signalled, should an interrupt become pending.
375 kvm_vgic_vmcr_sync(vcpu);
376 vgic_v4_put(vcpu, true);
380 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
387 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
389 struct kvm_s2_mmu *mmu;
392 mmu = vcpu->arch.hw_mmu;
393 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
396 * We guarantee that both TLBs and I-cache are private to each
397 * vcpu. If detecting that a vcpu from the same VM has
398 * previously run on the same physical CPU, call into the
399 * hypervisor code to nuke the relevant contexts.
401 * We might get preempted before the vCPU actually runs, but
402 * over-invalidation doesn't affect correctness.
404 if (*last_ran != vcpu->vcpu_id) {
405 kvm_call_hyp(__kvm_flush_cpu_context, mmu);
406 *last_ran = vcpu->vcpu_id;
412 kvm_timer_vcpu_load(vcpu);
414 kvm_vcpu_load_sysregs_vhe(vcpu);
415 kvm_arch_vcpu_load_fp(vcpu);
416 kvm_vcpu_pmu_restore_guest(vcpu);
417 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
418 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
420 if (single_task_running())
421 vcpu_clear_wfx_traps(vcpu);
423 vcpu_set_wfx_traps(vcpu);
425 if (vcpu_has_ptrauth(vcpu))
426 vcpu_ptrauth_disable(vcpu);
427 kvm_arch_vcpu_load_debug_state_flags(vcpu);
430 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
432 kvm_arch_vcpu_put_debug_state_flags(vcpu);
433 kvm_arch_vcpu_put_fp(vcpu);
435 kvm_vcpu_put_sysregs_vhe(vcpu);
436 kvm_timer_vcpu_put(vcpu);
438 kvm_vcpu_pmu_restore_host(vcpu);
443 static void vcpu_power_off(struct kvm_vcpu *vcpu)
445 vcpu->arch.power_off = true;
446 kvm_make_request(KVM_REQ_SLEEP, vcpu);
450 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
451 struct kvm_mp_state *mp_state)
453 if (vcpu->arch.power_off)
454 mp_state->mp_state = KVM_MP_STATE_STOPPED;
456 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
461 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
462 struct kvm_mp_state *mp_state)
466 switch (mp_state->mp_state) {
467 case KVM_MP_STATE_RUNNABLE:
468 vcpu->arch.power_off = false;
470 case KVM_MP_STATE_STOPPED:
471 vcpu_power_off(vcpu);
481 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
482 * @v: The VCPU pointer
484 * If the guest CPU is not waiting for interrupts or an interrupt line is
485 * asserted, the CPU is by definition runnable.
487 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
489 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
490 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
491 && !v->arch.power_off && !v->arch.pause);
494 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
496 return vcpu_mode_priv(vcpu);
499 /* Just ensure a guest exit from a particular CPU */
500 static void exit_vm_noop(void *info)
504 void force_vm_exit(const cpumask_t *mask)
507 smp_call_function_many(mask, exit_vm_noop, NULL, true);
512 * need_new_vmid_gen - check that the VMID is still valid
513 * @vmid: The VMID to check
515 * return true if there is a new generation of VMIDs being used
517 * The hardware supports a limited set of values with the value zero reserved
518 * for the host, so we check if an assigned value belongs to a previous
519 * generation, which requires us to assign a new value. If we're the first to
520 * use a VMID for the new generation, we must flush necessary caches and TLBs
523 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
525 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
526 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
527 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
531 * update_vmid - Update the vmid with a valid VMID for the current generation
532 * @vmid: The stage-2 VMID information struct
534 static void update_vmid(struct kvm_vmid *vmid)
536 if (!need_new_vmid_gen(vmid))
539 spin_lock(&kvm_vmid_lock);
542 * We need to re-check the vmid_gen here to ensure that if another vcpu
543 * already allocated a valid vmid for this vm, then this vcpu should
546 if (!need_new_vmid_gen(vmid)) {
547 spin_unlock(&kvm_vmid_lock);
551 /* First user of a new VMID generation? */
552 if (unlikely(kvm_next_vmid == 0)) {
553 atomic64_inc(&kvm_vmid_gen);
557 * On SMP we know no other CPUs can use this CPU's or each
558 * other's VMID after force_vm_exit returns since the
559 * kvm_vmid_lock blocks them from reentry to the guest.
561 force_vm_exit(cpu_all_mask);
563 * Now broadcast TLB + ICACHE invalidation over the inner
564 * shareable domain to make sure all data structures are
567 kvm_call_hyp(__kvm_flush_vm_context);
570 WRITE_ONCE(vmid->vmid, kvm_next_vmid);
572 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
575 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
577 spin_unlock(&kvm_vmid_lock);
580 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
582 struct kvm *kvm = vcpu->kvm;
585 if (likely(vcpu->arch.has_run_once))
588 if (!kvm_arm_vcpu_is_finalized(vcpu))
591 vcpu->arch.has_run_once = true;
593 kvm_arm_vcpu_init_debug(vcpu);
595 if (likely(irqchip_in_kernel(kvm))) {
597 * Map the VGIC hardware resources before running a vcpu the
598 * first time on this VM.
600 ret = kvm_vgic_map_resources(kvm);
605 * Tell the rest of the code that there are userspace irqchip
608 static_branch_inc(&userspace_irqchip_in_use);
611 ret = kvm_timer_enable(vcpu);
615 ret = kvm_arm_pmu_v3_enable(vcpu);
618 * Initialize traps for protected VMs.
619 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once
620 * the code is in place for first run initialization at EL2.
622 if (kvm_vm_is_protected(kvm))
623 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, 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_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
723 * @vcpu: The VCPU pointer
724 * @ret: Pointer to write optional return code
726 * Returns: true if the VCPU needs to return to a preemptible + interruptible
727 * and skip guest entry.
729 * This function disambiguates between two different types of exits: exits to a
730 * preemptible + interruptible kernel context and exits to userspace. For an
731 * exit to userspace, this function will write the return code to ret and return
732 * true. For an exit to preemptible + interruptible kernel context (i.e. check
733 * for pending work and re-enter), return true without writing to ret.
735 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
737 struct kvm_run *run = vcpu->run;
740 * If we're using a userspace irqchip, then check if we need
741 * to tell a userspace irqchip about timer or PMU level
742 * changes and if so, exit to userspace (the actual level
743 * state gets updated in kvm_timer_update_run and
744 * kvm_pmu_update_run below).
746 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
747 if (kvm_timer_should_notify_user(vcpu) ||
748 kvm_pmu_should_notify_user(vcpu)) {
750 run->exit_reason = KVM_EXIT_INTR;
755 return kvm_request_pending(vcpu) ||
756 need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
757 xfer_to_guest_mode_work_pending();
761 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
762 * @vcpu: The VCPU pointer
764 * This function is called through the VCPU_RUN ioctl called from user space. It
765 * will execute VM code in a loop until the time slice for the process is used
766 * or some emulation is needed from user space in which case the function will
767 * return with return value 0 and with the kvm_run structure filled in with the
768 * required data for the requested emulation.
770 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
772 struct kvm_run *run = vcpu->run;
775 if (unlikely(!kvm_vcpu_initialized(vcpu)))
778 ret = kvm_vcpu_first_run_init(vcpu);
782 if (run->exit_reason == KVM_EXIT_MMIO) {
783 ret = kvm_handle_mmio_return(vcpu);
790 if (run->immediate_exit) {
795 kvm_sigset_activate(vcpu);
798 run->exit_reason = KVM_EXIT_UNKNOWN;
801 * Check conditions before entering the guest
803 ret = xfer_to_guest_mode_handle_work(vcpu);
807 update_vmid(&vcpu->arch.hw_mmu->vmid);
809 check_vcpu_requests(vcpu);
812 * Preparing the interrupts to be injected also
813 * involves poking the GIC, which must be done in a
814 * non-preemptible context.
818 kvm_pmu_flush_hwstate(vcpu);
822 kvm_vgic_flush_hwstate(vcpu);
825 * Ensure we set mode to IN_GUEST_MODE after we disable
826 * interrupts and before the final VCPU requests check.
827 * See the comment in kvm_vcpu_exiting_guest_mode() and
828 * Documentation/virt/kvm/vcpu-requests.rst
830 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
832 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
833 vcpu->mode = OUTSIDE_GUEST_MODE;
834 isb(); /* Ensure work in x_flush_hwstate is committed */
835 kvm_pmu_sync_hwstate(vcpu);
836 if (static_branch_unlikely(&userspace_irqchip_in_use))
837 kvm_timer_sync_user(vcpu);
838 kvm_vgic_sync_hwstate(vcpu);
844 kvm_arm_setup_debug(vcpu);
846 /**************************************************************
849 trace_kvm_entry(*vcpu_pc(vcpu));
850 guest_enter_irqoff();
852 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
854 vcpu->mode = OUTSIDE_GUEST_MODE;
858 *************************************************************/
860 kvm_arm_clear_debug(vcpu);
863 * We must sync the PMU state before the vgic state so
864 * that the vgic can properly sample the updated state of the
867 kvm_pmu_sync_hwstate(vcpu);
870 * Sync the vgic state before syncing the timer state because
871 * the timer code needs to know if the virtual timer
872 * interrupts are active.
874 kvm_vgic_sync_hwstate(vcpu);
877 * Sync the timer hardware state before enabling interrupts as
878 * we don't want vtimer interrupts to race with syncing the
879 * timer virtual interrupt state.
881 if (static_branch_unlikely(&userspace_irqchip_in_use))
882 kvm_timer_sync_user(vcpu);
884 kvm_arch_vcpu_ctxsync_fp(vcpu);
887 * We may have taken a host interrupt in HYP mode (ie
888 * while executing the guest). This interrupt is still
889 * pending, as we haven't serviced it yet!
891 * We're now back in SVC mode, with interrupts
892 * disabled. Enabling the interrupts now will have
893 * the effect of taking the interrupt again, in SVC
899 * We do local_irq_enable() before calling guest_exit() so
900 * that if a timer interrupt hits while running the guest we
901 * account that tick as being spent in the guest. We enable
902 * preemption after calling guest_exit() so that if we get
903 * preempted we make sure ticks after that is not counted as
907 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
909 /* Exit types that need handling before we can be preempted */
910 handle_exit_early(vcpu, ret);
915 * The ARMv8 architecture doesn't give the hypervisor
916 * a mechanism to prevent a guest from dropping to AArch32 EL0
917 * if implemented by the CPU. If we spot the guest in such
918 * state and that we decided it wasn't supposed to do so (like
919 * with the asymmetric AArch32 case), return to userspace with
922 if (vcpu_mode_is_bad_32bit(vcpu)) {
924 * As we have caught the guest red-handed, decide that
925 * it isn't fit for purpose anymore by making the vcpu
926 * invalid. The VMM can try and fix it by issuing a
927 * KVM_ARM_VCPU_INIT if it really wants to.
929 vcpu->arch.target = -1;
930 ret = ARM_EXCEPTION_IL;
933 ret = handle_exit(vcpu, ret);
936 /* Tell userspace about in-kernel device output levels */
937 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
938 kvm_timer_update_run(vcpu);
939 kvm_pmu_update_run(vcpu);
942 kvm_sigset_deactivate(vcpu);
946 * In the unlikely event that we are returning to userspace
947 * with pending exceptions or PC adjustment, commit these
948 * adjustments in order to give userspace a consistent view of
949 * the vcpu state. Note that this relies on __kvm_adjust_pc()
950 * being preempt-safe on VHE.
952 if (unlikely(vcpu->arch.flags & (KVM_ARM64_PENDING_EXCEPTION |
953 KVM_ARM64_INCREMENT_PC)))
954 kvm_call_hyp(__kvm_adjust_pc, vcpu);
960 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
966 if (number == KVM_ARM_IRQ_CPU_IRQ)
967 bit_index = __ffs(HCR_VI);
968 else /* KVM_ARM_IRQ_CPU_FIQ */
969 bit_index = __ffs(HCR_VF);
971 hcr = vcpu_hcr(vcpu);
973 set = test_and_set_bit(bit_index, hcr);
975 set = test_and_clear_bit(bit_index, hcr);
978 * If we didn't change anything, no need to wake up or kick other CPUs
984 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
985 * trigger a world-switch round on the running physical CPU to set the
986 * virtual IRQ/FIQ fields in the HCR appropriately.
988 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
994 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
997 u32 irq = irq_level->irq;
998 unsigned int irq_type, vcpu_idx, irq_num;
999 int nrcpus = atomic_read(&kvm->online_vcpus);
1000 struct kvm_vcpu *vcpu = NULL;
1001 bool level = irq_level->level;
1003 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1004 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1005 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1006 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1008 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
1011 case KVM_ARM_IRQ_TYPE_CPU:
1012 if (irqchip_in_kernel(kvm))
1015 if (vcpu_idx >= nrcpus)
1018 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1022 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1025 return vcpu_interrupt_line(vcpu, irq_num, level);
1026 case KVM_ARM_IRQ_TYPE_PPI:
1027 if (!irqchip_in_kernel(kvm))
1030 if (vcpu_idx >= nrcpus)
1033 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
1037 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1040 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
1041 case KVM_ARM_IRQ_TYPE_SPI:
1042 if (!irqchip_in_kernel(kvm))
1045 if (irq_num < VGIC_NR_PRIVATE_IRQS)
1048 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
1054 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1055 const struct kvm_vcpu_init *init)
1057 unsigned int i, ret;
1058 u32 phys_target = kvm_target_cpu();
1060 if (init->target != phys_target)
1064 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1065 * use the same target.
1067 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
1070 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
1071 for (i = 0; i < sizeof(init->features) * 8; i++) {
1072 bool set = (init->features[i / 32] & (1 << (i % 32)));
1074 if (set && i >= KVM_VCPU_MAX_FEATURES)
1078 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1079 * use the same feature set.
1081 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1082 test_bit(i, vcpu->arch.features) != set)
1086 set_bit(i, vcpu->arch.features);
1089 vcpu->arch.target = phys_target;
1091 /* Now we know what it is, we can reset it. */
1092 ret = kvm_reset_vcpu(vcpu);
1094 vcpu->arch.target = -1;
1095 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1101 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1102 struct kvm_vcpu_init *init)
1106 ret = kvm_vcpu_set_target(vcpu, init);
1111 * Ensure a rebooted VM will fault in RAM pages and detect if the
1112 * guest MMU is turned off and flush the caches as needed.
1114 * S2FWB enforces all memory accesses to RAM being cacheable,
1115 * ensuring that the data side is always coherent. We still
1116 * need to invalidate the I-cache though, as FWB does *not*
1117 * imply CTR_EL0.DIC.
1119 if (vcpu->arch.has_run_once) {
1120 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1121 stage2_unmap_vm(vcpu->kvm);
1123 icache_inval_all_pou();
1126 vcpu_reset_hcr(vcpu);
1127 vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT;
1130 * Handle the "start in power-off" case.
1132 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1133 vcpu_power_off(vcpu);
1135 vcpu->arch.power_off = false;
1140 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1141 struct kvm_device_attr *attr)
1145 switch (attr->group) {
1147 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1154 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1155 struct kvm_device_attr *attr)
1159 switch (attr->group) {
1161 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1168 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1169 struct kvm_device_attr *attr)
1173 switch (attr->group) {
1175 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1182 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1183 struct kvm_vcpu_events *events)
1185 memset(events, 0, sizeof(*events));
1187 return __kvm_arm_vcpu_get_events(vcpu, events);
1190 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1191 struct kvm_vcpu_events *events)
1195 /* check whether the reserved field is zero */
1196 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1197 if (events->reserved[i])
1200 /* check whether the pad field is zero */
1201 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1202 if (events->exception.pad[i])
1205 return __kvm_arm_vcpu_set_events(vcpu, events);
1208 long kvm_arch_vcpu_ioctl(struct file *filp,
1209 unsigned int ioctl, unsigned long arg)
1211 struct kvm_vcpu *vcpu = filp->private_data;
1212 void __user *argp = (void __user *)arg;
1213 struct kvm_device_attr attr;
1217 case KVM_ARM_VCPU_INIT: {
1218 struct kvm_vcpu_init init;
1221 if (copy_from_user(&init, argp, sizeof(init)))
1224 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1227 case KVM_SET_ONE_REG:
1228 case KVM_GET_ONE_REG: {
1229 struct kvm_one_reg reg;
1232 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1236 if (copy_from_user(®, argp, sizeof(reg)))
1240 * We could owe a reset due to PSCI. Handle the pending reset
1241 * here to ensure userspace register accesses are ordered after
1244 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1245 kvm_reset_vcpu(vcpu);
1247 if (ioctl == KVM_SET_ONE_REG)
1248 r = kvm_arm_set_reg(vcpu, ®);
1250 r = kvm_arm_get_reg(vcpu, ®);
1253 case KVM_GET_REG_LIST: {
1254 struct kvm_reg_list __user *user_list = argp;
1255 struct kvm_reg_list reg_list;
1259 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1263 if (!kvm_arm_vcpu_is_finalized(vcpu))
1267 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1270 reg_list.n = kvm_arm_num_regs(vcpu);
1271 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1276 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1279 case KVM_SET_DEVICE_ATTR: {
1281 if (copy_from_user(&attr, argp, sizeof(attr)))
1283 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1286 case KVM_GET_DEVICE_ATTR: {
1288 if (copy_from_user(&attr, argp, sizeof(attr)))
1290 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1293 case KVM_HAS_DEVICE_ATTR: {
1295 if (copy_from_user(&attr, argp, sizeof(attr)))
1297 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1300 case KVM_GET_VCPU_EVENTS: {
1301 struct kvm_vcpu_events events;
1303 if (kvm_arm_vcpu_get_events(vcpu, &events))
1306 if (copy_to_user(argp, &events, sizeof(events)))
1311 case KVM_SET_VCPU_EVENTS: {
1312 struct kvm_vcpu_events events;
1314 if (copy_from_user(&events, argp, sizeof(events)))
1317 return kvm_arm_vcpu_set_events(vcpu, &events);
1319 case KVM_ARM_VCPU_FINALIZE: {
1322 if (!kvm_vcpu_initialized(vcpu))
1325 if (get_user(what, (const int __user *)argp))
1328 return kvm_arm_vcpu_finalize(vcpu, what);
1337 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1342 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1343 const struct kvm_memory_slot *memslot)
1345 kvm_flush_remote_tlbs(kvm);
1348 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1349 struct kvm_arm_device_addr *dev_addr)
1351 unsigned long dev_id, type;
1353 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1354 KVM_ARM_DEVICE_ID_SHIFT;
1355 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1356 KVM_ARM_DEVICE_TYPE_SHIFT;
1359 case KVM_ARM_DEVICE_VGIC_V2:
1362 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1368 long kvm_arch_vm_ioctl(struct file *filp,
1369 unsigned int ioctl, unsigned long arg)
1371 struct kvm *kvm = filp->private_data;
1372 void __user *argp = (void __user *)arg;
1375 case KVM_CREATE_IRQCHIP: {
1379 mutex_lock(&kvm->lock);
1380 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1381 mutex_unlock(&kvm->lock);
1384 case KVM_ARM_SET_DEVICE_ADDR: {
1385 struct kvm_arm_device_addr dev_addr;
1387 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1389 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1391 case KVM_ARM_PREFERRED_TARGET: {
1392 struct kvm_vcpu_init init;
1394 kvm_vcpu_preferred_target(&init);
1396 if (copy_to_user(argp, &init, sizeof(init)))
1401 case KVM_ARM_MTE_COPY_TAGS: {
1402 struct kvm_arm_copy_mte_tags copy_tags;
1404 if (copy_from_user(©_tags, argp, sizeof(copy_tags)))
1406 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags);
1413 static unsigned long nvhe_percpu_size(void)
1415 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1416 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1419 static unsigned long nvhe_percpu_order(void)
1421 unsigned long size = nvhe_percpu_size();
1423 return size ? get_order(size) : 0;
1426 /* A lookup table holding the hypervisor VA for each vector slot */
1427 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1429 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1431 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1434 static int kvm_init_vector_slots(void)
1439 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1440 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1442 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1443 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1445 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A))
1449 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1450 __BP_HARDEN_HYP_VECS_SZ, &base);
1455 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1456 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1460 static void cpu_prepare_hyp_mode(int cpu)
1462 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1466 * Calculate the raw per-cpu offset without a translation from the
1467 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1468 * so that we can use adr_l to access per-cpu variables in EL2.
1469 * Also drop the KASAN tag which gets in the way...
1471 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
1472 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1474 params->mair_el2 = read_sysreg(mair_el1);
1477 * The ID map may be configured to use an extended virtual address
1478 * range. This is only the case if system RAM is out of range for the
1479 * currently configured page size and VA_BITS, in which case we will
1480 * also need the extended virtual range for the HYP ID map, or we won't
1481 * be able to enable the EL2 MMU.
1483 * However, at EL2, there is only one TTBR register, and we can't switch
1484 * between translation tables *and* update TCR_EL2.T0SZ at the same
1485 * time. Bottom line: we need to use the extended range with *both* our
1486 * translation tables.
1488 * So use the same T0SZ value we use for the ID map.
1490 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1;
1491 tcr &= ~TCR_T0SZ_MASK;
1492 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET;
1493 params->tcr_el2 = tcr;
1495 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE);
1496 params->pgd_pa = kvm_mmu_get_httbr();
1497 if (is_protected_kvm_enabled())
1498 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
1500 params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
1501 params->vttbr = params->vtcr = 0;
1504 * Flush the init params from the data cache because the struct will
1505 * be read while the MMU is off.
1507 kvm_flush_dcache_to_poc(params, sizeof(*params));
1510 static void hyp_install_host_vector(void)
1512 struct kvm_nvhe_init_params *params;
1513 struct arm_smccc_res res;
1515 /* Switch from the HYP stub to our own HYP init vector */
1516 __hyp_set_vectors(kvm_get_idmap_vector());
1519 * Call initialization code, and switch to the full blown HYP code.
1520 * If the cpucaps haven't been finalized yet, something has gone very
1521 * wrong, and hyp will crash and burn when it uses any
1522 * cpus_have_const_cap() wrapper.
1524 BUG_ON(!system_capabilities_finalized());
1525 params = this_cpu_ptr_nvhe_sym(kvm_init_params);
1526 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
1527 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1530 static void cpu_init_hyp_mode(void)
1532 hyp_install_host_vector();
1535 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1538 if (this_cpu_has_cap(ARM64_SSBS) &&
1539 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1540 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1544 static void cpu_hyp_reset(void)
1546 if (!is_kernel_in_hyp_mode())
1547 __hyp_reset_vectors();
1551 * EL2 vectors can be mapped and rerouted in a number of ways,
1552 * depending on the kernel configuration and CPU present:
1554 * - If the CPU is affected by Spectre-v2, the hardening sequence is
1555 * placed in one of the vector slots, which is executed before jumping
1556 * to the real vectors.
1558 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
1559 * containing the hardening sequence is mapped next to the idmap page,
1560 * and executed before jumping to the real vectors.
1562 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
1563 * empty slot is selected, mapped next to the idmap page, and
1564 * executed before jumping to the real vectors.
1566 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
1567 * VHE, as we don't have hypervisor-specific mappings. If the system
1568 * is VHE and yet selects this capability, it will be ignored.
1570 static void cpu_set_hyp_vector(void)
1572 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
1573 void *vector = hyp_spectre_vector_selector[data->slot];
1575 if (!is_protected_kvm_enabled())
1576 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
1578 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
1581 static void cpu_hyp_init_context(void)
1583 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1585 if (!is_kernel_in_hyp_mode())
1586 cpu_init_hyp_mode();
1589 static void cpu_hyp_init_features(void)
1591 cpu_set_hyp_vector();
1592 kvm_arm_init_debug();
1594 if (is_kernel_in_hyp_mode())
1595 kvm_timer_init_vhe();
1598 kvm_vgic_init_cpu_hardware();
1601 static void cpu_hyp_reinit(void)
1604 cpu_hyp_init_context();
1605 cpu_hyp_init_features();
1608 static void _kvm_arch_hardware_enable(void *discard)
1610 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1612 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1616 int kvm_arch_hardware_enable(void)
1618 _kvm_arch_hardware_enable(NULL);
1622 static void _kvm_arch_hardware_disable(void *discard)
1624 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1626 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1630 void kvm_arch_hardware_disable(void)
1632 if (!is_protected_kvm_enabled())
1633 _kvm_arch_hardware_disable(NULL);
1636 #ifdef CONFIG_CPU_PM
1637 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1642 * kvm_arm_hardware_enabled is left with its old value over
1643 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1648 if (__this_cpu_read(kvm_arm_hardware_enabled))
1650 * don't update kvm_arm_hardware_enabled here
1651 * so that the hardware will be re-enabled
1652 * when we resume. See below.
1657 case CPU_PM_ENTER_FAILED:
1659 if (__this_cpu_read(kvm_arm_hardware_enabled))
1660 /* The hardware was enabled before suspend. */
1670 static struct notifier_block hyp_init_cpu_pm_nb = {
1671 .notifier_call = hyp_init_cpu_pm_notifier,
1674 static void hyp_cpu_pm_init(void)
1676 if (!is_protected_kvm_enabled())
1677 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1679 static void hyp_cpu_pm_exit(void)
1681 if (!is_protected_kvm_enabled())
1682 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1685 static inline void hyp_cpu_pm_init(void)
1688 static inline void hyp_cpu_pm_exit(void)
1693 static void init_cpu_logical_map(void)
1698 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
1699 * Only copy the set of online CPUs whose features have been chacked
1700 * against the finalized system capabilities. The hypervisor will not
1701 * allow any other CPUs from the `possible` set to boot.
1703 for_each_online_cpu(cpu)
1704 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
1707 #define init_psci_0_1_impl_state(config, what) \
1708 config.psci_0_1_ ## what ## _implemented = psci_ops.what
1710 static bool init_psci_relay(void)
1713 * If PSCI has not been initialized, protected KVM cannot install
1714 * itself on newly booted CPUs.
1716 if (!psci_ops.get_version) {
1717 kvm_err("Cannot initialize protected mode without PSCI\n");
1721 kvm_host_psci_config.version = psci_ops.get_version();
1723 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
1724 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
1725 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
1726 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
1727 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
1728 init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
1733 static int init_subsystems(void)
1738 * Enable hardware so that subsystem initialisation can access EL2.
1740 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1743 * Register CPU lower-power notifier
1748 * Init HYP view of VGIC
1750 err = kvm_vgic_hyp_init();
1753 vgic_present = true;
1757 vgic_present = false;
1765 * Init HYP architected timer support
1767 err = kvm_timer_hyp_init(vgic_present);
1772 kvm_sys_reg_table_init();
1775 if (err || !is_protected_kvm_enabled())
1776 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1781 static void teardown_hyp_mode(void)
1786 for_each_possible_cpu(cpu) {
1787 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1788 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1792 static int do_pkvm_init(u32 hyp_va_bits)
1794 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base);
1798 cpu_hyp_init_context();
1799 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
1800 num_possible_cpus(), kern_hyp_va(per_cpu_base),
1802 cpu_hyp_init_features();
1805 * The stub hypercalls are now disabled, so set our local flag to
1806 * prevent a later re-init attempt in kvm_arch_hardware_enable().
1808 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1814 static int kvm_hyp_init_protection(u32 hyp_va_bits)
1816 void *addr = phys_to_virt(hyp_mem_base);
1819 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
1820 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
1821 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
1822 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
1823 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
1824 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
1825 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
1827 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
1831 ret = do_pkvm_init(hyp_va_bits);
1841 * Inits Hyp-mode on all online CPUs
1843 static int init_hyp_mode(void)
1850 * The protected Hyp-mode cannot be initialized if the memory pool
1851 * allocation has failed.
1853 if (is_protected_kvm_enabled() && !hyp_mem_base)
1857 * Allocate Hyp PGD and setup Hyp identity mapping
1859 err = kvm_mmu_init(&hyp_va_bits);
1864 * Allocate stack pages for Hypervisor-mode
1866 for_each_possible_cpu(cpu) {
1867 unsigned long stack_page;
1869 stack_page = __get_free_page(GFP_KERNEL);
1875 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1879 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1881 for_each_possible_cpu(cpu) {
1885 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1891 page_addr = page_address(page);
1892 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1893 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1897 * Map the Hyp-code called directly from the host
1899 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1900 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1902 kvm_err("Cannot map world-switch code\n");
1906 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
1907 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
1909 kvm_err("Cannot map .hyp.rodata section\n");
1913 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1914 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1916 kvm_err("Cannot map rodata section\n");
1921 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
1922 * section thanks to an assertion in the linker script. Map it RW and
1923 * the rest of .bss RO.
1925 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
1926 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
1928 kvm_err("Cannot map hyp bss section: %d\n", err);
1932 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
1933 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1935 kvm_err("Cannot map bss section\n");
1940 * Map the Hyp stack pages
1942 for_each_possible_cpu(cpu) {
1943 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1944 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1948 kvm_err("Cannot map hyp stack\n");
1953 for_each_possible_cpu(cpu) {
1954 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1955 char *percpu_end = percpu_begin + nvhe_percpu_size();
1957 /* Map Hyp percpu pages */
1958 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1960 kvm_err("Cannot map hyp percpu region\n");
1964 /* Prepare the CPU initialization parameters */
1965 cpu_prepare_hyp_mode(cpu);
1968 if (is_protected_kvm_enabled()) {
1969 init_cpu_logical_map();
1971 if (!init_psci_relay()) {
1977 if (is_protected_kvm_enabled()) {
1978 err = kvm_hyp_init_protection(hyp_va_bits);
1980 kvm_err("Failed to init hyp memory protection\n");
1988 teardown_hyp_mode();
1989 kvm_err("error initializing Hyp mode: %d\n", err);
1993 static void _kvm_host_prot_finalize(void *arg)
1997 if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)))
1998 WRITE_ONCE(*err, -EINVAL);
2001 static int pkvm_drop_host_privileges(void)
2006 * Flip the static key upfront as that may no longer be possible
2007 * once the host stage 2 is installed.
2009 static_branch_enable(&kvm_protected_mode_initialized);
2010 on_each_cpu(_kvm_host_prot_finalize, &ret, 1);
2014 static int finalize_hyp_mode(void)
2016 if (!is_protected_kvm_enabled())
2020 * Exclude HYP BSS from kmemleak so that it doesn't get peeked
2021 * at, which would end badly once the section is inaccessible.
2022 * None of other sections should ever be introspected.
2024 kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start);
2025 return pkvm_drop_host_privileges();
2028 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2030 struct kvm_vcpu *vcpu;
2033 mpidr &= MPIDR_HWID_BITMASK;
2034 kvm_for_each_vcpu(i, vcpu, kvm) {
2035 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2041 bool kvm_arch_has_irq_bypass(void)
2046 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2047 struct irq_bypass_producer *prod)
2049 struct kvm_kernel_irqfd *irqfd =
2050 container_of(cons, struct kvm_kernel_irqfd, consumer);
2052 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2055 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2056 struct irq_bypass_producer *prod)
2058 struct kvm_kernel_irqfd *irqfd =
2059 container_of(cons, struct kvm_kernel_irqfd, consumer);
2061 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2065 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2067 struct kvm_kernel_irqfd *irqfd =
2068 container_of(cons, struct kvm_kernel_irqfd, consumer);
2070 kvm_arm_halt_guest(irqfd->kvm);
2073 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2075 struct kvm_kernel_irqfd *irqfd =
2076 container_of(cons, struct kvm_kernel_irqfd, consumer);
2078 kvm_arm_resume_guest(irqfd->kvm);
2082 * Initialize Hyp-mode and memory mappings on all CPUs.
2084 int kvm_arch_init(void *opaque)
2089 if (!is_hyp_mode_available()) {
2090 kvm_info("HYP mode not available\n");
2094 if (kvm_get_mode() == KVM_MODE_NONE) {
2095 kvm_info("KVM disabled from command line\n");
2099 in_hyp_mode = is_kernel_in_hyp_mode();
2101 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2102 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2103 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2104 "Only trusted guests should be used on this system.\n");
2106 err = kvm_set_ipa_limit();
2110 err = kvm_arm_init_sve();
2115 err = init_hyp_mode();
2120 err = kvm_init_vector_slots();
2122 kvm_err("Cannot initialise vector slots\n");
2126 err = init_subsystems();
2131 err = finalize_hyp_mode();
2133 kvm_err("Failed to finalize Hyp protection\n");
2138 if (is_protected_kvm_enabled()) {
2139 kvm_info("Protected nVHE mode initialized successfully\n");
2140 } else if (in_hyp_mode) {
2141 kvm_info("VHE mode initialized successfully\n");
2143 kvm_info("Hyp mode initialized successfully\n");
2151 teardown_hyp_mode();
2156 /* NOP: Compiling as a module not supported */
2157 void kvm_arch_exit(void)
2159 kvm_perf_teardown();
2162 static int __init early_kvm_mode_cfg(char *arg)
2167 if (strcmp(arg, "protected") == 0) {
2168 kvm_mode = KVM_MODE_PROTECTED;
2172 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2173 kvm_mode = KVM_MODE_DEFAULT;
2177 if (strcmp(arg, "none") == 0) {
2178 kvm_mode = KVM_MODE_NONE;
2184 early_param("kvm-arm.mode", early_kvm_mode_cfg);
2186 enum kvm_mode kvm_get_mode(void)
2191 static int arm_init(void)
2193 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2197 module_init(arm_init);