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 DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data);
50 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
52 /* The VMID used in the VTTBR */
53 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
54 static u32 kvm_next_vmid;
55 static DEFINE_SPINLOCK(kvm_vmid_lock);
57 static bool vgic_present;
59 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
60 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
62 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
64 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
67 int kvm_arch_hardware_setup(void *opaque)
72 int kvm_arch_check_processor_compat(void *opaque)
77 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
78 struct kvm_enable_cap *cap)
86 case KVM_CAP_ARM_NISV_TO_USER:
88 kvm->arch.return_nisv_io_abort_to_user = true;
98 static int kvm_arm_default_max_vcpus(void)
100 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
104 * kvm_arch_init_vm - initializes a VM data structure
105 * @kvm: pointer to the KVM struct
107 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
111 ret = kvm_arm_setup_stage2(kvm, type);
115 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
119 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
121 goto out_free_stage2_pgd;
123 kvm_vgic_early_init(kvm);
125 /* The maximum number of VCPUs is limited by the host's GIC model */
126 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
130 kvm_free_stage2_pgd(&kvm->arch.mmu);
134 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
136 return VM_FAULT_SIGBUS;
141 * kvm_arch_destroy_vm - destroy the VM data structure
142 * @kvm: pointer to the KVM struct
144 void kvm_arch_destroy_vm(struct kvm *kvm)
148 kvm_vgic_destroy(kvm);
150 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
152 kvm_vcpu_destroy(kvm->vcpus[i]);
153 kvm->vcpus[i] = NULL;
156 atomic_set(&kvm->online_vcpus, 0);
159 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
163 case KVM_CAP_IRQCHIP:
166 case KVM_CAP_IOEVENTFD:
167 case KVM_CAP_DEVICE_CTRL:
168 case KVM_CAP_USER_MEMORY:
169 case KVM_CAP_SYNC_MMU:
170 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
171 case KVM_CAP_ONE_REG:
172 case KVM_CAP_ARM_PSCI:
173 case KVM_CAP_ARM_PSCI_0_2:
174 case KVM_CAP_READONLY_MEM:
175 case KVM_CAP_MP_STATE:
176 case KVM_CAP_IMMEDIATE_EXIT:
177 case KVM_CAP_VCPU_EVENTS:
178 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
179 case KVM_CAP_ARM_NISV_TO_USER:
180 case KVM_CAP_ARM_INJECT_EXT_DABT:
183 case KVM_CAP_ARM_SET_DEVICE_ADDR:
186 case KVM_CAP_NR_VCPUS:
187 r = num_online_cpus();
189 case KVM_CAP_MAX_VCPUS:
190 case KVM_CAP_MAX_VCPU_ID:
192 r = kvm->arch.max_vcpus;
194 r = kvm_arm_default_max_vcpus();
196 case KVM_CAP_MSI_DEVID:
200 r = kvm->arch.vgic.msis_require_devid;
202 case KVM_CAP_ARM_USER_IRQ:
204 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
205 * (bump this number if adding more devices)
209 case KVM_CAP_STEAL_TIME:
210 r = kvm_arm_pvtime_supported();
213 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
219 long kvm_arch_dev_ioctl(struct file *filp,
220 unsigned int ioctl, unsigned long arg)
225 struct kvm *kvm_arch_alloc_vm(void)
228 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
230 return vzalloc(sizeof(struct kvm));
233 void kvm_arch_free_vm(struct kvm *kvm)
241 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
243 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
246 if (id >= kvm->arch.max_vcpus)
252 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
256 /* Force users to call KVM_ARM_VCPU_INIT */
257 vcpu->arch.target = -1;
258 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
260 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
262 /* Set up the timer */
263 kvm_timer_vcpu_init(vcpu);
265 kvm_pmu_vcpu_init(vcpu);
267 kvm_arm_reset_debug_ptr(vcpu);
269 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
271 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
273 err = kvm_vgic_vcpu_init(vcpu);
277 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
280 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
284 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
286 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
287 static_branch_dec(&userspace_irqchip_in_use);
289 kvm_mmu_free_memory_caches(vcpu);
290 kvm_timer_vcpu_terminate(vcpu);
291 kvm_pmu_vcpu_destroy(vcpu);
293 kvm_arm_vcpu_destroy(vcpu);
296 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
298 return kvm_timer_is_pending(vcpu);
301 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
304 * If we're about to block (most likely because we've just hit a
305 * WFI), we need to sync back the state of the GIC CPU interface
306 * so that we have the latest PMR and group enables. This ensures
307 * that kvm_arch_vcpu_runnable has up-to-date data to decide
308 * whether we have pending interrupts.
310 * For the same reason, we want to tell GICv4 that we need
311 * doorbells to be signalled, should an interrupt become pending.
314 kvm_vgic_vmcr_sync(vcpu);
315 vgic_v4_put(vcpu, true);
319 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
326 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
328 struct kvm_s2_mmu *mmu;
331 mmu = vcpu->arch.hw_mmu;
332 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
335 * We might get preempted before the vCPU actually runs, but
336 * over-invalidation doesn't affect correctness.
338 if (*last_ran != vcpu->vcpu_id) {
339 kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu);
340 *last_ran = vcpu->vcpu_id;
346 kvm_timer_vcpu_load(vcpu);
348 kvm_vcpu_load_sysregs_vhe(vcpu);
349 kvm_arch_vcpu_load_fp(vcpu);
350 kvm_vcpu_pmu_restore_guest(vcpu);
351 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
352 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
354 if (single_task_running())
355 vcpu_clear_wfx_traps(vcpu);
357 vcpu_set_wfx_traps(vcpu);
359 if (vcpu_has_ptrauth(vcpu))
360 vcpu_ptrauth_disable(vcpu);
363 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
365 kvm_arch_vcpu_put_fp(vcpu);
367 kvm_vcpu_put_sysregs_vhe(vcpu);
368 kvm_timer_vcpu_put(vcpu);
370 kvm_vcpu_pmu_restore_host(vcpu);
375 static void vcpu_power_off(struct kvm_vcpu *vcpu)
377 vcpu->arch.power_off = true;
378 kvm_make_request(KVM_REQ_SLEEP, vcpu);
382 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
383 struct kvm_mp_state *mp_state)
385 if (vcpu->arch.power_off)
386 mp_state->mp_state = KVM_MP_STATE_STOPPED;
388 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
393 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
394 struct kvm_mp_state *mp_state)
398 switch (mp_state->mp_state) {
399 case KVM_MP_STATE_RUNNABLE:
400 vcpu->arch.power_off = false;
402 case KVM_MP_STATE_STOPPED:
403 vcpu_power_off(vcpu);
413 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
414 * @v: The VCPU pointer
416 * If the guest CPU is not waiting for interrupts or an interrupt line is
417 * asserted, the CPU is by definition runnable.
419 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
421 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
422 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
423 && !v->arch.power_off && !v->arch.pause);
426 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
428 return vcpu_mode_priv(vcpu);
431 /* Just ensure a guest exit from a particular CPU */
432 static void exit_vm_noop(void *info)
436 void force_vm_exit(const cpumask_t *mask)
439 smp_call_function_many(mask, exit_vm_noop, NULL, true);
444 * need_new_vmid_gen - check that the VMID is still valid
445 * @vmid: The VMID to check
447 * return true if there is a new generation of VMIDs being used
449 * The hardware supports a limited set of values with the value zero reserved
450 * for the host, so we check if an assigned value belongs to a previous
451 * generation, which requires us to assign a new value. If we're the first to
452 * use a VMID for the new generation, we must flush necessary caches and TLBs
455 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
457 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
458 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
459 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
463 * update_vmid - Update the vmid with a valid VMID for the current generation
464 * @vmid: The stage-2 VMID information struct
466 static void update_vmid(struct kvm_vmid *vmid)
468 if (!need_new_vmid_gen(vmid))
471 spin_lock(&kvm_vmid_lock);
474 * We need to re-check the vmid_gen here to ensure that if another vcpu
475 * already allocated a valid vmid for this vm, then this vcpu should
478 if (!need_new_vmid_gen(vmid)) {
479 spin_unlock(&kvm_vmid_lock);
483 /* First user of a new VMID generation? */
484 if (unlikely(kvm_next_vmid == 0)) {
485 atomic64_inc(&kvm_vmid_gen);
489 * On SMP we know no other CPUs can use this CPU's or each
490 * other's VMID after force_vm_exit returns since the
491 * kvm_vmid_lock blocks them from reentry to the guest.
493 force_vm_exit(cpu_all_mask);
495 * Now broadcast TLB + ICACHE invalidation over the inner
496 * shareable domain to make sure all data structures are
499 kvm_call_hyp(__kvm_flush_vm_context);
502 vmid->vmid = kvm_next_vmid;
504 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
507 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
509 spin_unlock(&kvm_vmid_lock);
512 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
514 struct kvm *kvm = vcpu->kvm;
517 if (likely(vcpu->arch.has_run_once))
520 if (!kvm_arm_vcpu_is_finalized(vcpu))
523 vcpu->arch.has_run_once = true;
525 if (likely(irqchip_in_kernel(kvm))) {
527 * Map the VGIC hardware resources before running a vcpu the
528 * first time on this VM.
530 if (unlikely(!vgic_ready(kvm))) {
531 ret = kvm_vgic_map_resources(kvm);
537 * Tell the rest of the code that there are userspace irqchip
540 static_branch_inc(&userspace_irqchip_in_use);
543 ret = kvm_timer_enable(vcpu);
547 ret = kvm_arm_pmu_v3_enable(vcpu);
552 bool kvm_arch_intc_initialized(struct kvm *kvm)
554 return vgic_initialized(kvm);
557 void kvm_arm_halt_guest(struct kvm *kvm)
560 struct kvm_vcpu *vcpu;
562 kvm_for_each_vcpu(i, vcpu, kvm)
563 vcpu->arch.pause = true;
564 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
567 void kvm_arm_resume_guest(struct kvm *kvm)
570 struct kvm_vcpu *vcpu;
572 kvm_for_each_vcpu(i, vcpu, kvm) {
573 vcpu->arch.pause = false;
574 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
578 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
580 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
582 rcuwait_wait_event(wait,
583 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
586 if (vcpu->arch.power_off || vcpu->arch.pause) {
587 /* Awaken to handle a signal, request we sleep again later. */
588 kvm_make_request(KVM_REQ_SLEEP, vcpu);
592 * Make sure we will observe a potential reset request if we've
593 * observed a change to the power state. Pairs with the smp_wmb() in
594 * kvm_psci_vcpu_on().
599 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
601 return vcpu->arch.target >= 0;
604 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
606 if (kvm_request_pending(vcpu)) {
607 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
608 vcpu_req_sleep(vcpu);
610 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
611 kvm_reset_vcpu(vcpu);
614 * Clear IRQ_PENDING requests that were made to guarantee
615 * that a VCPU sees new virtual interrupts.
617 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
619 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
620 kvm_update_stolen_time(vcpu);
622 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
623 /* The distributor enable bits were changed */
625 vgic_v4_put(vcpu, false);
633 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
634 * @vcpu: The VCPU pointer
636 * This function is called through the VCPU_RUN ioctl called from user space. It
637 * will execute VM code in a loop until the time slice for the process is used
638 * or some emulation is needed from user space in which case the function will
639 * return with return value 0 and with the kvm_run structure filled in with the
640 * required data for the requested emulation.
642 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
644 struct kvm_run *run = vcpu->run;
647 if (unlikely(!kvm_vcpu_initialized(vcpu)))
650 ret = kvm_vcpu_first_run_init(vcpu);
654 if (run->exit_reason == KVM_EXIT_MMIO) {
655 ret = kvm_handle_mmio_return(vcpu);
660 if (run->immediate_exit)
665 kvm_sigset_activate(vcpu);
668 run->exit_reason = KVM_EXIT_UNKNOWN;
671 * Check conditions before entering the guest
675 update_vmid(&vcpu->arch.hw_mmu->vmid);
677 check_vcpu_requests(vcpu);
680 * Preparing the interrupts to be injected also
681 * involves poking the GIC, which must be done in a
682 * non-preemptible context.
686 kvm_pmu_flush_hwstate(vcpu);
690 kvm_vgic_flush_hwstate(vcpu);
693 * Exit if we have a signal pending so that we can deliver the
694 * signal to user space.
696 if (signal_pending(current)) {
698 run->exit_reason = KVM_EXIT_INTR;
702 * If we're using a userspace irqchip, then check if we need
703 * to tell a userspace irqchip about timer or PMU level
704 * changes and if so, exit to userspace (the actual level
705 * state gets updated in kvm_timer_update_run and
706 * kvm_pmu_update_run below).
708 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
709 if (kvm_timer_should_notify_user(vcpu) ||
710 kvm_pmu_should_notify_user(vcpu)) {
712 run->exit_reason = KVM_EXIT_INTR;
717 * Ensure we set mode to IN_GUEST_MODE after we disable
718 * interrupts and before the final VCPU requests check.
719 * See the comment in kvm_vcpu_exiting_guest_mode() and
720 * Documentation/virt/kvm/vcpu-requests.rst
722 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
724 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
725 kvm_request_pending(vcpu)) {
726 vcpu->mode = OUTSIDE_GUEST_MODE;
727 isb(); /* Ensure work in x_flush_hwstate is committed */
728 kvm_pmu_sync_hwstate(vcpu);
729 if (static_branch_unlikely(&userspace_irqchip_in_use))
730 kvm_timer_sync_user(vcpu);
731 kvm_vgic_sync_hwstate(vcpu);
737 kvm_arm_setup_debug(vcpu);
739 /**************************************************************
742 trace_kvm_entry(*vcpu_pc(vcpu));
743 guest_enter_irqoff();
745 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
747 vcpu->mode = OUTSIDE_GUEST_MODE;
751 *************************************************************/
753 kvm_arm_clear_debug(vcpu);
756 * We must sync the PMU state before the vgic state so
757 * that the vgic can properly sample the updated state of the
760 kvm_pmu_sync_hwstate(vcpu);
763 * Sync the vgic state before syncing the timer state because
764 * the timer code needs to know if the virtual timer
765 * interrupts are active.
767 kvm_vgic_sync_hwstate(vcpu);
770 * Sync the timer hardware state before enabling interrupts as
771 * we don't want vtimer interrupts to race with syncing the
772 * timer virtual interrupt state.
774 if (static_branch_unlikely(&userspace_irqchip_in_use))
775 kvm_timer_sync_user(vcpu);
777 kvm_arch_vcpu_ctxsync_fp(vcpu);
780 * We may have taken a host interrupt in HYP mode (ie
781 * while executing the guest). This interrupt is still
782 * pending, as we haven't serviced it yet!
784 * We're now back in SVC mode, with interrupts
785 * disabled. Enabling the interrupts now will have
786 * the effect of taking the interrupt again, in SVC
792 * We do local_irq_enable() before calling guest_exit() so
793 * that if a timer interrupt hits while running the guest we
794 * account that tick as being spent in the guest. We enable
795 * preemption after calling guest_exit() so that if we get
796 * preempted we make sure ticks after that is not counted as
800 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
802 /* Exit types that need handling before we can be preempted */
803 handle_exit_early(vcpu, ret);
807 ret = handle_exit(vcpu, ret);
810 /* Tell userspace about in-kernel device output levels */
811 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
812 kvm_timer_update_run(vcpu);
813 kvm_pmu_update_run(vcpu);
816 kvm_sigset_deactivate(vcpu);
822 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
828 if (number == KVM_ARM_IRQ_CPU_IRQ)
829 bit_index = __ffs(HCR_VI);
830 else /* KVM_ARM_IRQ_CPU_FIQ */
831 bit_index = __ffs(HCR_VF);
833 hcr = vcpu_hcr(vcpu);
835 set = test_and_set_bit(bit_index, hcr);
837 set = test_and_clear_bit(bit_index, hcr);
840 * If we didn't change anything, no need to wake up or kick other CPUs
846 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
847 * trigger a world-switch round on the running physical CPU to set the
848 * virtual IRQ/FIQ fields in the HCR appropriately.
850 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
856 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
859 u32 irq = irq_level->irq;
860 unsigned int irq_type, vcpu_idx, irq_num;
861 int nrcpus = atomic_read(&kvm->online_vcpus);
862 struct kvm_vcpu *vcpu = NULL;
863 bool level = irq_level->level;
865 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
866 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
867 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
868 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
870 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
873 case KVM_ARM_IRQ_TYPE_CPU:
874 if (irqchip_in_kernel(kvm))
877 if (vcpu_idx >= nrcpus)
880 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
884 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
887 return vcpu_interrupt_line(vcpu, irq_num, level);
888 case KVM_ARM_IRQ_TYPE_PPI:
889 if (!irqchip_in_kernel(kvm))
892 if (vcpu_idx >= nrcpus)
895 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
899 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
902 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
903 case KVM_ARM_IRQ_TYPE_SPI:
904 if (!irqchip_in_kernel(kvm))
907 if (irq_num < VGIC_NR_PRIVATE_IRQS)
910 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
916 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
917 const struct kvm_vcpu_init *init)
920 int phys_target = kvm_target_cpu();
922 if (init->target != phys_target)
926 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
927 * use the same target.
929 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
932 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
933 for (i = 0; i < sizeof(init->features) * 8; i++) {
934 bool set = (init->features[i / 32] & (1 << (i % 32)));
936 if (set && i >= KVM_VCPU_MAX_FEATURES)
940 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
941 * use the same feature set.
943 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
944 test_bit(i, vcpu->arch.features) != set)
948 set_bit(i, vcpu->arch.features);
951 vcpu->arch.target = phys_target;
953 /* Now we know what it is, we can reset it. */
954 ret = kvm_reset_vcpu(vcpu);
956 vcpu->arch.target = -1;
957 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
963 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
964 struct kvm_vcpu_init *init)
968 ret = kvm_vcpu_set_target(vcpu, init);
973 * Ensure a rebooted VM will fault in RAM pages and detect if the
974 * guest MMU is turned off and flush the caches as needed.
976 * S2FWB enforces all memory accesses to RAM being cacheable,
977 * ensuring that the data side is always coherent. We still
978 * need to invalidate the I-cache though, as FWB does *not*
981 if (vcpu->arch.has_run_once) {
982 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
983 stage2_unmap_vm(vcpu->kvm);
985 __flush_icache_all();
988 vcpu_reset_hcr(vcpu);
991 * Handle the "start in power-off" case.
993 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
994 vcpu_power_off(vcpu);
996 vcpu->arch.power_off = false;
1001 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1002 struct kvm_device_attr *attr)
1006 switch (attr->group) {
1008 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1015 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1016 struct kvm_device_attr *attr)
1020 switch (attr->group) {
1022 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1029 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1030 struct kvm_device_attr *attr)
1034 switch (attr->group) {
1036 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1043 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1044 struct kvm_vcpu_events *events)
1046 memset(events, 0, sizeof(*events));
1048 return __kvm_arm_vcpu_get_events(vcpu, events);
1051 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1052 struct kvm_vcpu_events *events)
1056 /* check whether the reserved field is zero */
1057 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1058 if (events->reserved[i])
1061 /* check whether the pad field is zero */
1062 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1063 if (events->exception.pad[i])
1066 return __kvm_arm_vcpu_set_events(vcpu, events);
1069 long kvm_arch_vcpu_ioctl(struct file *filp,
1070 unsigned int ioctl, unsigned long arg)
1072 struct kvm_vcpu *vcpu = filp->private_data;
1073 void __user *argp = (void __user *)arg;
1074 struct kvm_device_attr attr;
1078 case KVM_ARM_VCPU_INIT: {
1079 struct kvm_vcpu_init init;
1082 if (copy_from_user(&init, argp, sizeof(init)))
1085 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1088 case KVM_SET_ONE_REG:
1089 case KVM_GET_ONE_REG: {
1090 struct kvm_one_reg reg;
1093 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1097 if (copy_from_user(®, argp, sizeof(reg)))
1100 if (ioctl == KVM_SET_ONE_REG)
1101 r = kvm_arm_set_reg(vcpu, ®);
1103 r = kvm_arm_get_reg(vcpu, ®);
1106 case KVM_GET_REG_LIST: {
1107 struct kvm_reg_list __user *user_list = argp;
1108 struct kvm_reg_list reg_list;
1112 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1116 if (!kvm_arm_vcpu_is_finalized(vcpu))
1120 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1123 reg_list.n = kvm_arm_num_regs(vcpu);
1124 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1129 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1132 case KVM_SET_DEVICE_ATTR: {
1134 if (copy_from_user(&attr, argp, sizeof(attr)))
1136 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1139 case KVM_GET_DEVICE_ATTR: {
1141 if (copy_from_user(&attr, argp, sizeof(attr)))
1143 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1146 case KVM_HAS_DEVICE_ATTR: {
1148 if (copy_from_user(&attr, argp, sizeof(attr)))
1150 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1153 case KVM_GET_VCPU_EVENTS: {
1154 struct kvm_vcpu_events events;
1156 if (kvm_arm_vcpu_get_events(vcpu, &events))
1159 if (copy_to_user(argp, &events, sizeof(events)))
1164 case KVM_SET_VCPU_EVENTS: {
1165 struct kvm_vcpu_events events;
1167 if (copy_from_user(&events, argp, sizeof(events)))
1170 return kvm_arm_vcpu_set_events(vcpu, &events);
1172 case KVM_ARM_VCPU_FINALIZE: {
1175 if (!kvm_vcpu_initialized(vcpu))
1178 if (get_user(what, (const int __user *)argp))
1181 return kvm_arm_vcpu_finalize(vcpu, what);
1190 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1195 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1196 struct kvm_memory_slot *memslot)
1198 kvm_flush_remote_tlbs(kvm);
1201 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1202 struct kvm_arm_device_addr *dev_addr)
1204 unsigned long dev_id, type;
1206 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1207 KVM_ARM_DEVICE_ID_SHIFT;
1208 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1209 KVM_ARM_DEVICE_TYPE_SHIFT;
1212 case KVM_ARM_DEVICE_VGIC_V2:
1215 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1221 long kvm_arch_vm_ioctl(struct file *filp,
1222 unsigned int ioctl, unsigned long arg)
1224 struct kvm *kvm = filp->private_data;
1225 void __user *argp = (void __user *)arg;
1228 case KVM_CREATE_IRQCHIP: {
1232 mutex_lock(&kvm->lock);
1233 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1234 mutex_unlock(&kvm->lock);
1237 case KVM_ARM_SET_DEVICE_ADDR: {
1238 struct kvm_arm_device_addr dev_addr;
1240 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1242 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1244 case KVM_ARM_PREFERRED_TARGET: {
1246 struct kvm_vcpu_init init;
1248 err = kvm_vcpu_preferred_target(&init);
1252 if (copy_to_user(argp, &init, sizeof(init)))
1262 static void cpu_init_hyp_mode(void)
1264 phys_addr_t pgd_ptr;
1265 unsigned long hyp_stack_ptr;
1266 unsigned long vector_ptr;
1267 unsigned long tpidr_el2;
1269 /* Switch from the HYP stub to our own HYP init vector */
1270 __hyp_set_vectors(kvm_get_idmap_vector());
1273 * Calculate the raw per-cpu offset without a translation from the
1274 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1275 * so that we can use adr_l to access per-cpu variables in EL2.
1277 tpidr_el2 = ((unsigned long)this_cpu_ptr(&kvm_host_data) -
1278 (unsigned long)kvm_ksym_ref(&kvm_host_data));
1280 pgd_ptr = kvm_mmu_get_httbr();
1281 hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE;
1282 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1285 * Call initialization code, and switch to the full blown HYP code.
1286 * If the cpucaps haven't been finalized yet, something has gone very
1287 * wrong, and hyp will crash and burn when it uses any
1288 * cpus_have_const_cap() wrapper.
1290 BUG_ON(!system_capabilities_finalized());
1291 __kvm_call_hyp((void *)pgd_ptr, hyp_stack_ptr, vector_ptr, tpidr_el2);
1294 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1297 if (this_cpu_has_cap(ARM64_SSBS) &&
1298 arm64_get_ssbd_state() == ARM64_SSBD_FORCE_DISABLE) {
1299 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1303 static void cpu_hyp_reset(void)
1305 if (!is_kernel_in_hyp_mode())
1306 __hyp_reset_vectors();
1309 static void cpu_hyp_reinit(void)
1311 kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1315 if (is_kernel_in_hyp_mode())
1316 kvm_timer_init_vhe();
1318 cpu_init_hyp_mode();
1320 kvm_arm_init_debug();
1323 kvm_vgic_init_cpu_hardware();
1326 static void _kvm_arch_hardware_enable(void *discard)
1328 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1330 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1334 int kvm_arch_hardware_enable(void)
1336 _kvm_arch_hardware_enable(NULL);
1340 static void _kvm_arch_hardware_disable(void *discard)
1342 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1344 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1348 void kvm_arch_hardware_disable(void)
1350 _kvm_arch_hardware_disable(NULL);
1353 #ifdef CONFIG_CPU_PM
1354 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1359 * kvm_arm_hardware_enabled is left with its old value over
1360 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1365 if (__this_cpu_read(kvm_arm_hardware_enabled))
1367 * don't update kvm_arm_hardware_enabled here
1368 * so that the hardware will be re-enabled
1369 * when we resume. See below.
1374 case CPU_PM_ENTER_FAILED:
1376 if (__this_cpu_read(kvm_arm_hardware_enabled))
1377 /* The hardware was enabled before suspend. */
1387 static struct notifier_block hyp_init_cpu_pm_nb = {
1388 .notifier_call = hyp_init_cpu_pm_notifier,
1391 static void __init hyp_cpu_pm_init(void)
1393 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1395 static void __init hyp_cpu_pm_exit(void)
1397 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1400 static inline void hyp_cpu_pm_init(void)
1403 static inline void hyp_cpu_pm_exit(void)
1408 static int init_common_resources(void)
1410 return kvm_set_ipa_limit();
1413 static int init_subsystems(void)
1418 * Enable hardware so that subsystem initialisation can access EL2.
1420 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1423 * Register CPU lower-power notifier
1428 * Init HYP view of VGIC
1430 err = kvm_vgic_hyp_init();
1433 vgic_present = true;
1437 vgic_present = false;
1445 * Init HYP architected timer support
1447 err = kvm_timer_hyp_init(vgic_present);
1452 kvm_coproc_table_init();
1455 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1460 static void teardown_hyp_mode(void)
1465 for_each_possible_cpu(cpu)
1466 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1470 * Inits Hyp-mode on all online CPUs
1472 static int init_hyp_mode(void)
1478 * Allocate Hyp PGD and setup Hyp identity mapping
1480 err = kvm_mmu_init();
1485 * Allocate stack pages for Hypervisor-mode
1487 for_each_possible_cpu(cpu) {
1488 unsigned long stack_page;
1490 stack_page = __get_free_page(GFP_KERNEL);
1496 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1500 * Map the Hyp-code called directly from the host
1502 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1503 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1505 kvm_err("Cannot map world-switch code\n");
1509 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1510 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1512 kvm_err("Cannot map rodata section\n");
1516 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1517 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1519 kvm_err("Cannot map bss section\n");
1523 err = kvm_map_vectors();
1525 kvm_err("Cannot map vectors\n");
1530 * Map the Hyp stack pages
1532 for_each_possible_cpu(cpu) {
1533 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1534 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1538 kvm_err("Cannot map hyp stack\n");
1543 for_each_possible_cpu(cpu) {
1544 kvm_host_data_t *cpu_data;
1546 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1547 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1550 kvm_err("Cannot map host CPU state: %d\n", err);
1555 err = hyp_map_aux_data();
1557 kvm_err("Cannot map host auxiliary data: %d\n", err);
1562 teardown_hyp_mode();
1563 kvm_err("error initializing Hyp mode: %d\n", err);
1567 static void check_kvm_target_cpu(void *ret)
1569 *(int *)ret = kvm_target_cpu();
1572 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1574 struct kvm_vcpu *vcpu;
1577 mpidr &= MPIDR_HWID_BITMASK;
1578 kvm_for_each_vcpu(i, vcpu, kvm) {
1579 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1585 bool kvm_arch_has_irq_bypass(void)
1590 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1591 struct irq_bypass_producer *prod)
1593 struct kvm_kernel_irqfd *irqfd =
1594 container_of(cons, struct kvm_kernel_irqfd, consumer);
1596 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1599 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1600 struct irq_bypass_producer *prod)
1602 struct kvm_kernel_irqfd *irqfd =
1603 container_of(cons, struct kvm_kernel_irqfd, consumer);
1605 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1609 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1611 struct kvm_kernel_irqfd *irqfd =
1612 container_of(cons, struct kvm_kernel_irqfd, consumer);
1614 kvm_arm_halt_guest(irqfd->kvm);
1617 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1619 struct kvm_kernel_irqfd *irqfd =
1620 container_of(cons, struct kvm_kernel_irqfd, consumer);
1622 kvm_arm_resume_guest(irqfd->kvm);
1626 * Initialize Hyp-mode and memory mappings on all CPUs.
1628 int kvm_arch_init(void *opaque)
1634 if (!is_hyp_mode_available()) {
1635 kvm_info("HYP mode not available\n");
1639 in_hyp_mode = is_kernel_in_hyp_mode();
1641 if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1642 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1646 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE))
1647 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
1648 "Only trusted guests should be used on this system.\n");
1650 for_each_online_cpu(cpu) {
1651 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1653 kvm_err("Error, CPU %d not supported!\n", cpu);
1658 err = init_common_resources();
1662 err = kvm_arm_init_sve();
1667 err = init_hyp_mode();
1672 err = init_subsystems();
1677 kvm_info("VHE mode initialized successfully\n");
1679 kvm_info("Hyp mode initialized successfully\n");
1686 teardown_hyp_mode();
1691 /* NOP: Compiling as a module not supported */
1692 void kvm_arch_exit(void)
1694 kvm_perf_teardown();
1697 static int arm_init(void)
1699 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1703 module_init(arm_init);