1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
4 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
8 #include <linux/cpu_pm.h>
9 #include <linux/errno.h>
10 #include <linux/err.h>
11 #include <linux/kvm_host.h>
12 #include <linux/list.h>
13 #include <linux/module.h>
14 #include <linux/vmalloc.h>
16 #include <linux/mman.h>
17 #include <linux/sched.h>
18 #include <linux/kvm.h>
19 #include <linux/kvm_irqfd.h>
20 #include <linux/irqbypass.h>
21 #include <linux/sched/stat.h>
22 #include <trace/events/kvm.h>
24 #define CREATE_TRACE_POINTS
25 #include "trace_arm.h"
27 #include <linux/uaccess.h>
28 #include <asm/ptrace.h>
30 #include <asm/tlbflush.h>
31 #include <asm/cacheflush.h>
32 #include <asm/cpufeature.h>
34 #include <asm/kvm_arm.h>
35 #include <asm/kvm_asm.h>
36 #include <asm/kvm_mmu.h>
37 #include <asm/kvm_emulate.h>
38 #include <asm/kvm_coproc.h>
39 #include <asm/sections.h>
41 #include <kvm/arm_hypercalls.h>
42 #include <kvm/arm_pmu.h>
43 #include <kvm/arm_psci.h>
46 __asm__(".arch_extension virt");
49 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
51 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS];
54 /* The VMID used in the VTTBR */
55 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
56 static u32 kvm_next_vmid;
57 static DEFINE_SPINLOCK(kvm_vmid_lock);
59 static bool vgic_present;
61 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
62 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
64 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
66 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
69 int kvm_arch_hardware_setup(void *opaque)
74 int kvm_arch_check_processor_compat(void *opaque)
79 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
80 struct kvm_enable_cap *cap)
88 case KVM_CAP_ARM_NISV_TO_USER:
90 kvm->arch.return_nisv_io_abort_to_user = true;
100 static int kvm_arm_default_max_vcpus(void)
102 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
106 * kvm_arch_init_vm - initializes a VM data structure
107 * @kvm: pointer to the KVM struct
109 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
113 ret = kvm_arm_setup_stage2(kvm, type);
117 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
121 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
123 goto out_free_stage2_pgd;
125 kvm_vgic_early_init(kvm);
127 /* The maximum number of VCPUs is limited by the host's GIC model */
128 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
132 kvm_free_stage2_pgd(&kvm->arch.mmu);
136 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
138 return VM_FAULT_SIGBUS;
143 * kvm_arch_destroy_vm - destroy the VM data structure
144 * @kvm: pointer to the KVM struct
146 void kvm_arch_destroy_vm(struct kvm *kvm)
150 bitmap_free(kvm->arch.pmu_filter);
152 kvm_vgic_destroy(kvm);
154 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
156 kvm_vcpu_destroy(kvm->vcpus[i]);
157 kvm->vcpus[i] = NULL;
160 atomic_set(&kvm->online_vcpus, 0);
163 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
167 case KVM_CAP_IRQCHIP:
170 case KVM_CAP_IOEVENTFD:
171 case KVM_CAP_DEVICE_CTRL:
172 case KVM_CAP_USER_MEMORY:
173 case KVM_CAP_SYNC_MMU:
174 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
175 case KVM_CAP_ONE_REG:
176 case KVM_CAP_ARM_PSCI:
177 case KVM_CAP_ARM_PSCI_0_2:
178 case KVM_CAP_READONLY_MEM:
179 case KVM_CAP_MP_STATE:
180 case KVM_CAP_IMMEDIATE_EXIT:
181 case KVM_CAP_VCPU_EVENTS:
182 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
183 case KVM_CAP_ARM_NISV_TO_USER:
184 case KVM_CAP_ARM_INJECT_EXT_DABT:
187 case KVM_CAP_ARM_SET_DEVICE_ADDR:
190 case KVM_CAP_NR_VCPUS:
191 r = num_online_cpus();
193 case KVM_CAP_MAX_VCPUS:
194 case KVM_CAP_MAX_VCPU_ID:
196 r = kvm->arch.max_vcpus;
198 r = kvm_arm_default_max_vcpus();
200 case KVM_CAP_MSI_DEVID:
204 r = kvm->arch.vgic.msis_require_devid;
206 case KVM_CAP_ARM_USER_IRQ:
208 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
209 * (bump this number if adding more devices)
213 case KVM_CAP_STEAL_TIME:
214 r = kvm_arm_pvtime_supported();
217 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
223 long kvm_arch_dev_ioctl(struct file *filp,
224 unsigned int ioctl, unsigned long arg)
229 struct kvm *kvm_arch_alloc_vm(void)
232 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
234 return vzalloc(sizeof(struct kvm));
237 void kvm_arch_free_vm(struct kvm *kvm)
245 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
247 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
250 if (id >= kvm->arch.max_vcpus)
256 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
260 /* Force users to call KVM_ARM_VCPU_INIT */
261 vcpu->arch.target = -1;
262 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
264 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
266 /* Set up the timer */
267 kvm_timer_vcpu_init(vcpu);
269 kvm_pmu_vcpu_init(vcpu);
271 kvm_arm_reset_debug_ptr(vcpu);
273 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
275 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
277 err = kvm_vgic_vcpu_init(vcpu);
281 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
284 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
288 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
290 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
291 static_branch_dec(&userspace_irqchip_in_use);
293 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
294 kvm_timer_vcpu_terminate(vcpu);
295 kvm_pmu_vcpu_destroy(vcpu);
297 kvm_arm_vcpu_destroy(vcpu);
300 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
302 return kvm_timer_is_pending(vcpu);
305 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
308 * If we're about to block (most likely because we've just hit a
309 * WFI), we need to sync back the state of the GIC CPU interface
310 * so that we have the latest PMR and group enables. This ensures
311 * that kvm_arch_vcpu_runnable has up-to-date data to decide
312 * whether we have pending interrupts.
314 * For the same reason, we want to tell GICv4 that we need
315 * doorbells to be signalled, should an interrupt become pending.
318 kvm_vgic_vmcr_sync(vcpu);
319 vgic_v4_put(vcpu, true);
323 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
330 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
332 struct kvm_s2_mmu *mmu;
335 mmu = vcpu->arch.hw_mmu;
336 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
339 * We might get preempted before the vCPU actually runs, but
340 * over-invalidation doesn't affect correctness.
342 if (*last_ran != vcpu->vcpu_id) {
343 kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu);
344 *last_ran = vcpu->vcpu_id;
350 kvm_timer_vcpu_load(vcpu);
352 kvm_vcpu_load_sysregs_vhe(vcpu);
353 kvm_arch_vcpu_load_fp(vcpu);
354 kvm_vcpu_pmu_restore_guest(vcpu);
355 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
356 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
358 if (single_task_running())
359 vcpu_clear_wfx_traps(vcpu);
361 vcpu_set_wfx_traps(vcpu);
363 if (vcpu_has_ptrauth(vcpu))
364 vcpu_ptrauth_disable(vcpu);
367 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
369 kvm_arch_vcpu_put_fp(vcpu);
371 kvm_vcpu_put_sysregs_vhe(vcpu);
372 kvm_timer_vcpu_put(vcpu);
374 kvm_vcpu_pmu_restore_host(vcpu);
379 static void vcpu_power_off(struct kvm_vcpu *vcpu)
381 vcpu->arch.power_off = true;
382 kvm_make_request(KVM_REQ_SLEEP, vcpu);
386 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
387 struct kvm_mp_state *mp_state)
389 if (vcpu->arch.power_off)
390 mp_state->mp_state = KVM_MP_STATE_STOPPED;
392 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
397 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
398 struct kvm_mp_state *mp_state)
402 switch (mp_state->mp_state) {
403 case KVM_MP_STATE_RUNNABLE:
404 vcpu->arch.power_off = false;
406 case KVM_MP_STATE_STOPPED:
407 vcpu_power_off(vcpu);
417 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
418 * @v: The VCPU pointer
420 * If the guest CPU is not waiting for interrupts or an interrupt line is
421 * asserted, the CPU is by definition runnable.
423 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
425 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
426 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
427 && !v->arch.power_off && !v->arch.pause);
430 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
432 return vcpu_mode_priv(vcpu);
435 /* Just ensure a guest exit from a particular CPU */
436 static void exit_vm_noop(void *info)
440 void force_vm_exit(const cpumask_t *mask)
443 smp_call_function_many(mask, exit_vm_noop, NULL, true);
448 * need_new_vmid_gen - check that the VMID is still valid
449 * @vmid: The VMID to check
451 * return true if there is a new generation of VMIDs being used
453 * The hardware supports a limited set of values with the value zero reserved
454 * for the host, so we check if an assigned value belongs to a previous
455 * generation, which requires us to assign a new value. If we're the first to
456 * use a VMID for the new generation, we must flush necessary caches and TLBs
459 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
461 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
462 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
463 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
467 * update_vmid - Update the vmid with a valid VMID for the current generation
468 * @vmid: The stage-2 VMID information struct
470 static void update_vmid(struct kvm_vmid *vmid)
472 if (!need_new_vmid_gen(vmid))
475 spin_lock(&kvm_vmid_lock);
478 * We need to re-check the vmid_gen here to ensure that if another vcpu
479 * already allocated a valid vmid for this vm, then this vcpu should
482 if (!need_new_vmid_gen(vmid)) {
483 spin_unlock(&kvm_vmid_lock);
487 /* First user of a new VMID generation? */
488 if (unlikely(kvm_next_vmid == 0)) {
489 atomic64_inc(&kvm_vmid_gen);
493 * On SMP we know no other CPUs can use this CPU's or each
494 * other's VMID after force_vm_exit returns since the
495 * kvm_vmid_lock blocks them from reentry to the guest.
497 force_vm_exit(cpu_all_mask);
499 * Now broadcast TLB + ICACHE invalidation over the inner
500 * shareable domain to make sure all data structures are
503 kvm_call_hyp(__kvm_flush_vm_context);
506 vmid->vmid = kvm_next_vmid;
508 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
511 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
513 spin_unlock(&kvm_vmid_lock);
516 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
518 struct kvm *kvm = vcpu->kvm;
521 if (likely(vcpu->arch.has_run_once))
524 if (!kvm_arm_vcpu_is_finalized(vcpu))
527 vcpu->arch.has_run_once = true;
529 if (likely(irqchip_in_kernel(kvm))) {
531 * Map the VGIC hardware resources before running a vcpu the
532 * first time on this VM.
534 if (unlikely(!vgic_ready(kvm))) {
535 ret = kvm_vgic_map_resources(kvm);
541 * Tell the rest of the code that there are userspace irqchip
544 static_branch_inc(&userspace_irqchip_in_use);
547 ret = kvm_timer_enable(vcpu);
551 ret = kvm_arm_pmu_v3_enable(vcpu);
556 bool kvm_arch_intc_initialized(struct kvm *kvm)
558 return vgic_initialized(kvm);
561 void kvm_arm_halt_guest(struct kvm *kvm)
564 struct kvm_vcpu *vcpu;
566 kvm_for_each_vcpu(i, vcpu, kvm)
567 vcpu->arch.pause = true;
568 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
571 void kvm_arm_resume_guest(struct kvm *kvm)
574 struct kvm_vcpu *vcpu;
576 kvm_for_each_vcpu(i, vcpu, kvm) {
577 vcpu->arch.pause = false;
578 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
582 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
584 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
586 rcuwait_wait_event(wait,
587 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
590 if (vcpu->arch.power_off || vcpu->arch.pause) {
591 /* Awaken to handle a signal, request we sleep again later. */
592 kvm_make_request(KVM_REQ_SLEEP, vcpu);
596 * Make sure we will observe a potential reset request if we've
597 * observed a change to the power state. Pairs with the smp_wmb() in
598 * kvm_psci_vcpu_on().
603 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
605 return vcpu->arch.target >= 0;
608 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
610 if (kvm_request_pending(vcpu)) {
611 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
612 vcpu_req_sleep(vcpu);
614 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
615 kvm_reset_vcpu(vcpu);
618 * Clear IRQ_PENDING requests that were made to guarantee
619 * that a VCPU sees new virtual interrupts.
621 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
623 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
624 kvm_update_stolen_time(vcpu);
626 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
627 /* The distributor enable bits were changed */
629 vgic_v4_put(vcpu, false);
637 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
638 * @vcpu: The VCPU pointer
640 * This function is called through the VCPU_RUN ioctl called from user space. It
641 * will execute VM code in a loop until the time slice for the process is used
642 * or some emulation is needed from user space in which case the function will
643 * return with return value 0 and with the kvm_run structure filled in with the
644 * required data for the requested emulation.
646 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
648 struct kvm_run *run = vcpu->run;
651 if (unlikely(!kvm_vcpu_initialized(vcpu)))
654 ret = kvm_vcpu_first_run_init(vcpu);
658 if (run->exit_reason == KVM_EXIT_MMIO) {
659 ret = kvm_handle_mmio_return(vcpu);
664 if (run->immediate_exit)
669 kvm_sigset_activate(vcpu);
672 run->exit_reason = KVM_EXIT_UNKNOWN;
675 * Check conditions before entering the guest
679 update_vmid(&vcpu->arch.hw_mmu->vmid);
681 check_vcpu_requests(vcpu);
684 * Preparing the interrupts to be injected also
685 * involves poking the GIC, which must be done in a
686 * non-preemptible context.
690 kvm_pmu_flush_hwstate(vcpu);
694 kvm_vgic_flush_hwstate(vcpu);
697 * Exit if we have a signal pending so that we can deliver the
698 * signal to user space.
700 if (signal_pending(current)) {
702 run->exit_reason = KVM_EXIT_INTR;
706 * If we're using a userspace irqchip, then check if we need
707 * to tell a userspace irqchip about timer or PMU level
708 * changes and if so, exit to userspace (the actual level
709 * state gets updated in kvm_timer_update_run and
710 * kvm_pmu_update_run below).
712 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
713 if (kvm_timer_should_notify_user(vcpu) ||
714 kvm_pmu_should_notify_user(vcpu)) {
716 run->exit_reason = KVM_EXIT_INTR;
721 * Ensure we set mode to IN_GUEST_MODE after we disable
722 * interrupts and before the final VCPU requests check.
723 * See the comment in kvm_vcpu_exiting_guest_mode() and
724 * Documentation/virt/kvm/vcpu-requests.rst
726 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
728 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
729 kvm_request_pending(vcpu)) {
730 vcpu->mode = OUTSIDE_GUEST_MODE;
731 isb(); /* Ensure work in x_flush_hwstate is committed */
732 kvm_pmu_sync_hwstate(vcpu);
733 if (static_branch_unlikely(&userspace_irqchip_in_use))
734 kvm_timer_sync_user(vcpu);
735 kvm_vgic_sync_hwstate(vcpu);
741 kvm_arm_setup_debug(vcpu);
743 /**************************************************************
746 trace_kvm_entry(*vcpu_pc(vcpu));
747 guest_enter_irqoff();
749 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
751 vcpu->mode = OUTSIDE_GUEST_MODE;
755 *************************************************************/
757 kvm_arm_clear_debug(vcpu);
760 * We must sync the PMU state before the vgic state so
761 * that the vgic can properly sample the updated state of the
764 kvm_pmu_sync_hwstate(vcpu);
767 * Sync the vgic state before syncing the timer state because
768 * the timer code needs to know if the virtual timer
769 * interrupts are active.
771 kvm_vgic_sync_hwstate(vcpu);
774 * Sync the timer hardware state before enabling interrupts as
775 * we don't want vtimer interrupts to race with syncing the
776 * timer virtual interrupt state.
778 if (static_branch_unlikely(&userspace_irqchip_in_use))
779 kvm_timer_sync_user(vcpu);
781 kvm_arch_vcpu_ctxsync_fp(vcpu);
784 * We may have taken a host interrupt in HYP mode (ie
785 * while executing the guest). This interrupt is still
786 * pending, as we haven't serviced it yet!
788 * We're now back in SVC mode, with interrupts
789 * disabled. Enabling the interrupts now will have
790 * the effect of taking the interrupt again, in SVC
796 * We do local_irq_enable() before calling guest_exit() so
797 * that if a timer interrupt hits while running the guest we
798 * account that tick as being spent in the guest. We enable
799 * preemption after calling guest_exit() so that if we get
800 * preempted we make sure ticks after that is not counted as
804 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
806 /* Exit types that need handling before we can be preempted */
807 handle_exit_early(vcpu, ret);
812 * The ARMv8 architecture doesn't give the hypervisor
813 * a mechanism to prevent a guest from dropping to AArch32 EL0
814 * if implemented by the CPU. If we spot the guest in such
815 * state and that we decided it wasn't supposed to do so (like
816 * with the asymmetric AArch32 case), return to userspace with
819 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
821 * As we have caught the guest red-handed, decide that
822 * it isn't fit for purpose anymore by making the vcpu
823 * invalid. The VMM can try and fix it by issuing a
824 * KVM_ARM_VCPU_INIT if it really wants to.
826 vcpu->arch.target = -1;
827 ret = ARM_EXCEPTION_IL;
830 ret = handle_exit(vcpu, ret);
833 /* Tell userspace about in-kernel device output levels */
834 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
835 kvm_timer_update_run(vcpu);
836 kvm_pmu_update_run(vcpu);
839 kvm_sigset_deactivate(vcpu);
845 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
851 if (number == KVM_ARM_IRQ_CPU_IRQ)
852 bit_index = __ffs(HCR_VI);
853 else /* KVM_ARM_IRQ_CPU_FIQ */
854 bit_index = __ffs(HCR_VF);
856 hcr = vcpu_hcr(vcpu);
858 set = test_and_set_bit(bit_index, hcr);
860 set = test_and_clear_bit(bit_index, hcr);
863 * If we didn't change anything, no need to wake up or kick other CPUs
869 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
870 * trigger a world-switch round on the running physical CPU to set the
871 * virtual IRQ/FIQ fields in the HCR appropriately.
873 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
879 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
882 u32 irq = irq_level->irq;
883 unsigned int irq_type, vcpu_idx, irq_num;
884 int nrcpus = atomic_read(&kvm->online_vcpus);
885 struct kvm_vcpu *vcpu = NULL;
886 bool level = irq_level->level;
888 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
889 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
890 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
891 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
893 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
896 case KVM_ARM_IRQ_TYPE_CPU:
897 if (irqchip_in_kernel(kvm))
900 if (vcpu_idx >= nrcpus)
903 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
907 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
910 return vcpu_interrupt_line(vcpu, irq_num, level);
911 case KVM_ARM_IRQ_TYPE_PPI:
912 if (!irqchip_in_kernel(kvm))
915 if (vcpu_idx >= nrcpus)
918 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
922 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
925 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
926 case KVM_ARM_IRQ_TYPE_SPI:
927 if (!irqchip_in_kernel(kvm))
930 if (irq_num < VGIC_NR_PRIVATE_IRQS)
933 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
939 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
940 const struct kvm_vcpu_init *init)
943 int phys_target = kvm_target_cpu();
945 if (init->target != phys_target)
949 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
950 * use the same target.
952 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
955 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
956 for (i = 0; i < sizeof(init->features) * 8; i++) {
957 bool set = (init->features[i / 32] & (1 << (i % 32)));
959 if (set && i >= KVM_VCPU_MAX_FEATURES)
963 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
964 * use the same feature set.
966 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
967 test_bit(i, vcpu->arch.features) != set)
971 set_bit(i, vcpu->arch.features);
974 vcpu->arch.target = phys_target;
976 /* Now we know what it is, we can reset it. */
977 ret = kvm_reset_vcpu(vcpu);
979 vcpu->arch.target = -1;
980 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
986 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
987 struct kvm_vcpu_init *init)
991 ret = kvm_vcpu_set_target(vcpu, init);
996 * Ensure a rebooted VM will fault in RAM pages and detect if the
997 * guest MMU is turned off and flush the caches as needed.
999 * S2FWB enforces all memory accesses to RAM being cacheable,
1000 * ensuring that the data side is always coherent. We still
1001 * need to invalidate the I-cache though, as FWB does *not*
1002 * imply CTR_EL0.DIC.
1004 if (vcpu->arch.has_run_once) {
1005 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1006 stage2_unmap_vm(vcpu->kvm);
1008 __flush_icache_all();
1011 vcpu_reset_hcr(vcpu);
1014 * Handle the "start in power-off" case.
1016 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1017 vcpu_power_off(vcpu);
1019 vcpu->arch.power_off = false;
1024 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1025 struct kvm_device_attr *attr)
1029 switch (attr->group) {
1031 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1038 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1039 struct kvm_device_attr *attr)
1043 switch (attr->group) {
1045 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1052 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1053 struct kvm_device_attr *attr)
1057 switch (attr->group) {
1059 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1066 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1067 struct kvm_vcpu_events *events)
1069 memset(events, 0, sizeof(*events));
1071 return __kvm_arm_vcpu_get_events(vcpu, events);
1074 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1075 struct kvm_vcpu_events *events)
1079 /* check whether the reserved field is zero */
1080 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1081 if (events->reserved[i])
1084 /* check whether the pad field is zero */
1085 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1086 if (events->exception.pad[i])
1089 return __kvm_arm_vcpu_set_events(vcpu, events);
1092 long kvm_arch_vcpu_ioctl(struct file *filp,
1093 unsigned int ioctl, unsigned long arg)
1095 struct kvm_vcpu *vcpu = filp->private_data;
1096 void __user *argp = (void __user *)arg;
1097 struct kvm_device_attr attr;
1101 case KVM_ARM_VCPU_INIT: {
1102 struct kvm_vcpu_init init;
1105 if (copy_from_user(&init, argp, sizeof(init)))
1108 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1111 case KVM_SET_ONE_REG:
1112 case KVM_GET_ONE_REG: {
1113 struct kvm_one_reg reg;
1116 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1120 if (copy_from_user(®, argp, sizeof(reg)))
1123 if (ioctl == KVM_SET_ONE_REG)
1124 r = kvm_arm_set_reg(vcpu, ®);
1126 r = kvm_arm_get_reg(vcpu, ®);
1129 case KVM_GET_REG_LIST: {
1130 struct kvm_reg_list __user *user_list = argp;
1131 struct kvm_reg_list reg_list;
1135 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1139 if (!kvm_arm_vcpu_is_finalized(vcpu))
1143 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1146 reg_list.n = kvm_arm_num_regs(vcpu);
1147 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1152 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1155 case KVM_SET_DEVICE_ATTR: {
1157 if (copy_from_user(&attr, argp, sizeof(attr)))
1159 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1162 case KVM_GET_DEVICE_ATTR: {
1164 if (copy_from_user(&attr, argp, sizeof(attr)))
1166 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1169 case KVM_HAS_DEVICE_ATTR: {
1171 if (copy_from_user(&attr, argp, sizeof(attr)))
1173 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1176 case KVM_GET_VCPU_EVENTS: {
1177 struct kvm_vcpu_events events;
1179 if (kvm_arm_vcpu_get_events(vcpu, &events))
1182 if (copy_to_user(argp, &events, sizeof(events)))
1187 case KVM_SET_VCPU_EVENTS: {
1188 struct kvm_vcpu_events events;
1190 if (copy_from_user(&events, argp, sizeof(events)))
1193 return kvm_arm_vcpu_set_events(vcpu, &events);
1195 case KVM_ARM_VCPU_FINALIZE: {
1198 if (!kvm_vcpu_initialized(vcpu))
1201 if (get_user(what, (const int __user *)argp))
1204 return kvm_arm_vcpu_finalize(vcpu, what);
1213 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1218 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1219 struct kvm_memory_slot *memslot)
1221 kvm_flush_remote_tlbs(kvm);
1224 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1225 struct kvm_arm_device_addr *dev_addr)
1227 unsigned long dev_id, type;
1229 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1230 KVM_ARM_DEVICE_ID_SHIFT;
1231 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1232 KVM_ARM_DEVICE_TYPE_SHIFT;
1235 case KVM_ARM_DEVICE_VGIC_V2:
1238 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1244 long kvm_arch_vm_ioctl(struct file *filp,
1245 unsigned int ioctl, unsigned long arg)
1247 struct kvm *kvm = filp->private_data;
1248 void __user *argp = (void __user *)arg;
1251 case KVM_CREATE_IRQCHIP: {
1255 mutex_lock(&kvm->lock);
1256 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1257 mutex_unlock(&kvm->lock);
1260 case KVM_ARM_SET_DEVICE_ADDR: {
1261 struct kvm_arm_device_addr dev_addr;
1263 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1265 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1267 case KVM_ARM_PREFERRED_TARGET: {
1269 struct kvm_vcpu_init init;
1271 err = kvm_vcpu_preferred_target(&init);
1275 if (copy_to_user(argp, &init, sizeof(init)))
1285 static unsigned long nvhe_percpu_size(void)
1287 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1288 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1291 static unsigned long nvhe_percpu_order(void)
1293 unsigned long size = nvhe_percpu_size();
1295 return size ? get_order(size) : 0;
1298 static int kvm_map_vectors(void)
1301 * SV2 = ARM64_SPECTRE_V2
1302 * HEL2 = ARM64_HARDEN_EL2_VECTORS
1304 * !SV2 + !HEL2 -> use direct vectors
1305 * SV2 + !HEL2 -> use hardened vectors in place
1306 * !SV2 + HEL2 -> allocate one vector slot and use exec mapping
1307 * SV2 + HEL2 -> use hardened vectors and use exec mapping
1309 if (cpus_have_const_cap(ARM64_SPECTRE_V2)) {
1310 __kvm_bp_vect_base = kvm_ksym_ref(__bp_harden_hyp_vecs);
1311 __kvm_bp_vect_base = kern_hyp_va(__kvm_bp_vect_base);
1314 if (cpus_have_const_cap(ARM64_HARDEN_EL2_VECTORS)) {
1315 phys_addr_t vect_pa = __pa_symbol(__bp_harden_hyp_vecs);
1316 unsigned long size = __BP_HARDEN_HYP_VECS_SZ;
1319 * Always allocate a spare vector slot, as we don't
1320 * know yet which CPUs have a BP hardening slot that
1323 __kvm_harden_el2_vector_slot = atomic_inc_return(&arm64_el2_vector_last_slot);
1324 BUG_ON(__kvm_harden_el2_vector_slot >= BP_HARDEN_EL2_SLOTS);
1325 return create_hyp_exec_mappings(vect_pa, size,
1326 &__kvm_bp_vect_base);
1332 static void cpu_init_hyp_mode(void)
1334 phys_addr_t pgd_ptr;
1335 unsigned long hyp_stack_ptr;
1336 unsigned long vector_ptr;
1337 unsigned long tpidr_el2;
1338 struct arm_smccc_res res;
1340 /* Switch from the HYP stub to our own HYP init vector */
1341 __hyp_set_vectors(kvm_get_idmap_vector());
1344 * Calculate the raw per-cpu offset without a translation from the
1345 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1346 * so that we can use adr_l to access per-cpu variables in EL2.
1348 tpidr_el2 = (unsigned long)this_cpu_ptr_nvhe_sym(__per_cpu_start) -
1349 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1351 pgd_ptr = kvm_mmu_get_httbr();
1352 hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE;
1353 hyp_stack_ptr = kern_hyp_va(hyp_stack_ptr);
1354 vector_ptr = (unsigned long)kern_hyp_va(kvm_ksym_ref(__kvm_hyp_host_vector));
1357 * Call initialization code, and switch to the full blown HYP code.
1358 * If the cpucaps haven't been finalized yet, something has gone very
1359 * wrong, and hyp will crash and burn when it uses any
1360 * cpus_have_const_cap() wrapper.
1362 BUG_ON(!system_capabilities_finalized());
1363 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init),
1364 pgd_ptr, tpidr_el2, hyp_stack_ptr, vector_ptr, &res);
1365 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1368 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1371 if (this_cpu_has_cap(ARM64_SSBS) &&
1372 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1373 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1377 static void cpu_hyp_reset(void)
1379 if (!is_kernel_in_hyp_mode())
1380 __hyp_reset_vectors();
1383 static void cpu_hyp_reinit(void)
1385 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1389 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)kvm_get_hyp_vector();
1391 if (is_kernel_in_hyp_mode())
1392 kvm_timer_init_vhe();
1394 cpu_init_hyp_mode();
1396 kvm_arm_init_debug();
1399 kvm_vgic_init_cpu_hardware();
1402 static void _kvm_arch_hardware_enable(void *discard)
1404 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1406 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1410 int kvm_arch_hardware_enable(void)
1412 _kvm_arch_hardware_enable(NULL);
1416 static void _kvm_arch_hardware_disable(void *discard)
1418 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1420 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1424 void kvm_arch_hardware_disable(void)
1426 _kvm_arch_hardware_disable(NULL);
1429 #ifdef CONFIG_CPU_PM
1430 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1435 * kvm_arm_hardware_enabled is left with its old value over
1436 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1441 if (__this_cpu_read(kvm_arm_hardware_enabled))
1443 * don't update kvm_arm_hardware_enabled here
1444 * so that the hardware will be re-enabled
1445 * when we resume. See below.
1450 case CPU_PM_ENTER_FAILED:
1452 if (__this_cpu_read(kvm_arm_hardware_enabled))
1453 /* The hardware was enabled before suspend. */
1463 static struct notifier_block hyp_init_cpu_pm_nb = {
1464 .notifier_call = hyp_init_cpu_pm_notifier,
1467 static void __init hyp_cpu_pm_init(void)
1469 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1471 static void __init hyp_cpu_pm_exit(void)
1473 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1476 static inline void hyp_cpu_pm_init(void)
1479 static inline void hyp_cpu_pm_exit(void)
1484 static int init_common_resources(void)
1486 return kvm_set_ipa_limit();
1489 static int init_subsystems(void)
1494 * Enable hardware so that subsystem initialisation can access EL2.
1496 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1499 * Register CPU lower-power notifier
1504 * Init HYP view of VGIC
1506 err = kvm_vgic_hyp_init();
1509 vgic_present = true;
1513 vgic_present = false;
1521 * Init HYP architected timer support
1523 err = kvm_timer_hyp_init(vgic_present);
1528 kvm_coproc_table_init();
1531 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1536 static void teardown_hyp_mode(void)
1541 for_each_possible_cpu(cpu) {
1542 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1543 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1548 * Inits Hyp-mode on all online CPUs
1550 static int init_hyp_mode(void)
1556 * Allocate Hyp PGD and setup Hyp identity mapping
1558 err = kvm_mmu_init();
1563 * Allocate stack pages for Hypervisor-mode
1565 for_each_possible_cpu(cpu) {
1566 unsigned long stack_page;
1568 stack_page = __get_free_page(GFP_KERNEL);
1574 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1578 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1580 for_each_possible_cpu(cpu) {
1584 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1590 page_addr = page_address(page);
1591 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1592 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1596 * Map the Hyp-code called directly from the host
1598 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1599 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1601 kvm_err("Cannot map world-switch code\n");
1605 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1606 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1608 kvm_err("Cannot map rodata section\n");
1612 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1613 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1615 kvm_err("Cannot map bss section\n");
1619 err = kvm_map_vectors();
1621 kvm_err("Cannot map vectors\n");
1626 * Map the Hyp stack pages
1628 for_each_possible_cpu(cpu) {
1629 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1630 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1634 kvm_err("Cannot map hyp stack\n");
1640 * Map Hyp percpu pages
1642 for_each_possible_cpu(cpu) {
1643 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1644 char *percpu_end = percpu_begin + nvhe_percpu_size();
1646 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1649 kvm_err("Cannot map hyp percpu region\n");
1657 teardown_hyp_mode();
1658 kvm_err("error initializing Hyp mode: %d\n", err);
1662 static void check_kvm_target_cpu(void *ret)
1664 *(int *)ret = kvm_target_cpu();
1667 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1669 struct kvm_vcpu *vcpu;
1672 mpidr &= MPIDR_HWID_BITMASK;
1673 kvm_for_each_vcpu(i, vcpu, kvm) {
1674 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1680 bool kvm_arch_has_irq_bypass(void)
1685 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1686 struct irq_bypass_producer *prod)
1688 struct kvm_kernel_irqfd *irqfd =
1689 container_of(cons, struct kvm_kernel_irqfd, consumer);
1691 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1694 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1695 struct irq_bypass_producer *prod)
1697 struct kvm_kernel_irqfd *irqfd =
1698 container_of(cons, struct kvm_kernel_irqfd, consumer);
1700 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1704 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1706 struct kvm_kernel_irqfd *irqfd =
1707 container_of(cons, struct kvm_kernel_irqfd, consumer);
1709 kvm_arm_halt_guest(irqfd->kvm);
1712 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1714 struct kvm_kernel_irqfd *irqfd =
1715 container_of(cons, struct kvm_kernel_irqfd, consumer);
1717 kvm_arm_resume_guest(irqfd->kvm);
1721 * Initialize Hyp-mode and memory mappings on all CPUs.
1723 int kvm_arch_init(void *opaque)
1729 if (!is_hyp_mode_available()) {
1730 kvm_info("HYP mode not available\n");
1734 in_hyp_mode = is_kernel_in_hyp_mode();
1736 if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1737 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1741 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE))
1742 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
1743 "Only trusted guests should be used on this system.\n");
1745 for_each_online_cpu(cpu) {
1746 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1748 kvm_err("Error, CPU %d not supported!\n", cpu);
1753 err = init_common_resources();
1757 err = kvm_arm_init_sve();
1762 err = init_hyp_mode();
1767 err = init_subsystems();
1772 kvm_info("VHE mode initialized successfully\n");
1774 kvm_info("Hyp mode initialized successfully\n");
1781 teardown_hyp_mode();
1786 /* NOP: Compiling as a module not supported */
1787 void kvm_arch_exit(void)
1789 kvm_perf_teardown();
1792 static int arm_init(void)
1794 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1798 module_init(arm_init);