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)
210 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
216 long kvm_arch_dev_ioctl(struct file *filp,
217 unsigned int ioctl, unsigned long arg)
222 struct kvm *kvm_arch_alloc_vm(void)
225 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
227 return vzalloc(sizeof(struct kvm));
230 void kvm_arch_free_vm(struct kvm *kvm)
238 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
240 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
243 if (id >= kvm->arch.max_vcpus)
249 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
253 /* Force users to call KVM_ARM_VCPU_INIT */
254 vcpu->arch.target = -1;
255 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
257 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
259 /* Set up the timer */
260 kvm_timer_vcpu_init(vcpu);
262 kvm_pmu_vcpu_init(vcpu);
264 kvm_arm_reset_debug_ptr(vcpu);
266 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
268 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
270 err = kvm_vgic_vcpu_init(vcpu);
274 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
277 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
281 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
283 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
284 static_branch_dec(&userspace_irqchip_in_use);
286 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
287 kvm_timer_vcpu_terminate(vcpu);
288 kvm_pmu_vcpu_destroy(vcpu);
290 kvm_arm_vcpu_destroy(vcpu);
293 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
295 return kvm_timer_is_pending(vcpu);
298 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
301 * If we're about to block (most likely because we've just hit a
302 * WFI), we need to sync back the state of the GIC CPU interface
303 * so that we have the latest PMR and group enables. This ensures
304 * that kvm_arch_vcpu_runnable has up-to-date data to decide
305 * whether we have pending interrupts.
307 * For the same reason, we want to tell GICv4 that we need
308 * doorbells to be signalled, should an interrupt become pending.
311 kvm_vgic_vmcr_sync(vcpu);
312 vgic_v4_put(vcpu, true);
316 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
323 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
325 struct kvm_s2_mmu *mmu;
328 mmu = vcpu->arch.hw_mmu;
329 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
332 * We might get preempted before the vCPU actually runs, but
333 * over-invalidation doesn't affect correctness.
335 if (*last_ran != vcpu->vcpu_id) {
336 kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu);
337 *last_ran = vcpu->vcpu_id;
343 kvm_timer_vcpu_load(vcpu);
345 kvm_vcpu_load_sysregs_vhe(vcpu);
346 kvm_arch_vcpu_load_fp(vcpu);
347 kvm_vcpu_pmu_restore_guest(vcpu);
348 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
349 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
351 if (single_task_running())
352 vcpu_clear_wfx_traps(vcpu);
354 vcpu_set_wfx_traps(vcpu);
356 if (vcpu_has_ptrauth(vcpu))
357 vcpu_ptrauth_disable(vcpu);
360 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
362 kvm_arch_vcpu_put_fp(vcpu);
364 kvm_vcpu_put_sysregs_vhe(vcpu);
365 kvm_timer_vcpu_put(vcpu);
367 kvm_vcpu_pmu_restore_host(vcpu);
372 static void vcpu_power_off(struct kvm_vcpu *vcpu)
374 vcpu->arch.power_off = true;
375 kvm_make_request(KVM_REQ_SLEEP, vcpu);
379 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
380 struct kvm_mp_state *mp_state)
382 if (vcpu->arch.power_off)
383 mp_state->mp_state = KVM_MP_STATE_STOPPED;
385 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
390 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
391 struct kvm_mp_state *mp_state)
395 switch (mp_state->mp_state) {
396 case KVM_MP_STATE_RUNNABLE:
397 vcpu->arch.power_off = false;
399 case KVM_MP_STATE_STOPPED:
400 vcpu_power_off(vcpu);
410 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
411 * @v: The VCPU pointer
413 * If the guest CPU is not waiting for interrupts or an interrupt line is
414 * asserted, the CPU is by definition runnable.
416 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
418 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
419 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
420 && !v->arch.power_off && !v->arch.pause);
423 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
425 return vcpu_mode_priv(vcpu);
428 /* Just ensure a guest exit from a particular CPU */
429 static void exit_vm_noop(void *info)
433 void force_vm_exit(const cpumask_t *mask)
436 smp_call_function_many(mask, exit_vm_noop, NULL, true);
441 * need_new_vmid_gen - check that the VMID is still valid
442 * @vmid: The VMID to check
444 * return true if there is a new generation of VMIDs being used
446 * The hardware supports a limited set of values with the value zero reserved
447 * for the host, so we check if an assigned value belongs to a previous
448 * generation, which requires us to assign a new value. If we're the first to
449 * use a VMID for the new generation, we must flush necessary caches and TLBs
452 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
454 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
455 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
456 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
460 * update_vmid - Update the vmid with a valid VMID for the current generation
461 * @vmid: The stage-2 VMID information struct
463 static void update_vmid(struct kvm_vmid *vmid)
465 if (!need_new_vmid_gen(vmid))
468 spin_lock(&kvm_vmid_lock);
471 * We need to re-check the vmid_gen here to ensure that if another vcpu
472 * already allocated a valid vmid for this vm, then this vcpu should
475 if (!need_new_vmid_gen(vmid)) {
476 spin_unlock(&kvm_vmid_lock);
480 /* First user of a new VMID generation? */
481 if (unlikely(kvm_next_vmid == 0)) {
482 atomic64_inc(&kvm_vmid_gen);
486 * On SMP we know no other CPUs can use this CPU's or each
487 * other's VMID after force_vm_exit returns since the
488 * kvm_vmid_lock blocks them from reentry to the guest.
490 force_vm_exit(cpu_all_mask);
492 * Now broadcast TLB + ICACHE invalidation over the inner
493 * shareable domain to make sure all data structures are
496 kvm_call_hyp(__kvm_flush_vm_context);
499 vmid->vmid = kvm_next_vmid;
501 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
504 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
506 spin_unlock(&kvm_vmid_lock);
509 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
511 struct kvm *kvm = vcpu->kvm;
514 if (likely(vcpu->arch.has_run_once))
517 if (!kvm_arm_vcpu_is_finalized(vcpu))
520 vcpu->arch.has_run_once = true;
522 if (likely(irqchip_in_kernel(kvm))) {
524 * Map the VGIC hardware resources before running a vcpu the
525 * first time on this VM.
527 if (unlikely(!vgic_ready(kvm))) {
528 ret = kvm_vgic_map_resources(kvm);
534 * Tell the rest of the code that there are userspace irqchip
537 static_branch_inc(&userspace_irqchip_in_use);
540 ret = kvm_timer_enable(vcpu);
544 ret = kvm_arm_pmu_v3_enable(vcpu);
549 bool kvm_arch_intc_initialized(struct kvm *kvm)
551 return vgic_initialized(kvm);
554 void kvm_arm_halt_guest(struct kvm *kvm)
557 struct kvm_vcpu *vcpu;
559 kvm_for_each_vcpu(i, vcpu, kvm)
560 vcpu->arch.pause = true;
561 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
564 void kvm_arm_resume_guest(struct kvm *kvm)
567 struct kvm_vcpu *vcpu;
569 kvm_for_each_vcpu(i, vcpu, kvm) {
570 vcpu->arch.pause = false;
571 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
575 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
577 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
579 rcuwait_wait_event(wait,
580 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
583 if (vcpu->arch.power_off || vcpu->arch.pause) {
584 /* Awaken to handle a signal, request we sleep again later. */
585 kvm_make_request(KVM_REQ_SLEEP, vcpu);
589 * Make sure we will observe a potential reset request if we've
590 * observed a change to the power state. Pairs with the smp_wmb() in
591 * kvm_psci_vcpu_on().
596 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
598 return vcpu->arch.target >= 0;
601 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
603 if (kvm_request_pending(vcpu)) {
604 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
605 vcpu_req_sleep(vcpu);
607 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
608 kvm_reset_vcpu(vcpu);
611 * Clear IRQ_PENDING requests that were made to guarantee
612 * that a VCPU sees new virtual interrupts.
614 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
616 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
617 kvm_update_stolen_time(vcpu);
619 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
620 /* The distributor enable bits were changed */
622 vgic_v4_put(vcpu, false);
630 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
631 * @vcpu: The VCPU pointer
633 * This function is called through the VCPU_RUN ioctl called from user space. It
634 * will execute VM code in a loop until the time slice for the process is used
635 * or some emulation is needed from user space in which case the function will
636 * return with return value 0 and with the kvm_run structure filled in with the
637 * required data for the requested emulation.
639 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
641 struct kvm_run *run = vcpu->run;
644 if (unlikely(!kvm_vcpu_initialized(vcpu)))
647 ret = kvm_vcpu_first_run_init(vcpu);
651 if (run->exit_reason == KVM_EXIT_MMIO) {
652 ret = kvm_handle_mmio_return(vcpu);
657 if (run->immediate_exit)
662 kvm_sigset_activate(vcpu);
665 run->exit_reason = KVM_EXIT_UNKNOWN;
668 * Check conditions before entering the guest
672 update_vmid(&vcpu->arch.hw_mmu->vmid);
674 check_vcpu_requests(vcpu);
677 * Preparing the interrupts to be injected also
678 * involves poking the GIC, which must be done in a
679 * non-preemptible context.
683 kvm_pmu_flush_hwstate(vcpu);
687 kvm_vgic_flush_hwstate(vcpu);
690 * Exit if we have a signal pending so that we can deliver the
691 * signal to user space.
693 if (signal_pending(current)) {
695 run->exit_reason = KVM_EXIT_INTR;
699 * If we're using a userspace irqchip, then check if we need
700 * to tell a userspace irqchip about timer or PMU level
701 * changes and if so, exit to userspace (the actual level
702 * state gets updated in kvm_timer_update_run and
703 * kvm_pmu_update_run below).
705 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
706 if (kvm_timer_should_notify_user(vcpu) ||
707 kvm_pmu_should_notify_user(vcpu)) {
709 run->exit_reason = KVM_EXIT_INTR;
714 * Ensure we set mode to IN_GUEST_MODE after we disable
715 * interrupts and before the final VCPU requests check.
716 * See the comment in kvm_vcpu_exiting_guest_mode() and
717 * Documentation/virt/kvm/vcpu-requests.rst
719 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
721 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
722 kvm_request_pending(vcpu)) {
723 vcpu->mode = OUTSIDE_GUEST_MODE;
724 isb(); /* Ensure work in x_flush_hwstate is committed */
725 kvm_pmu_sync_hwstate(vcpu);
726 if (static_branch_unlikely(&userspace_irqchip_in_use))
727 kvm_timer_sync_user(vcpu);
728 kvm_vgic_sync_hwstate(vcpu);
734 kvm_arm_setup_debug(vcpu);
736 /**************************************************************
739 trace_kvm_entry(*vcpu_pc(vcpu));
740 guest_enter_irqoff();
742 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
744 vcpu->mode = OUTSIDE_GUEST_MODE;
748 *************************************************************/
750 kvm_arm_clear_debug(vcpu);
753 * We must sync the PMU state before the vgic state so
754 * that the vgic can properly sample the updated state of the
757 kvm_pmu_sync_hwstate(vcpu);
760 * Sync the vgic state before syncing the timer state because
761 * the timer code needs to know if the virtual timer
762 * interrupts are active.
764 kvm_vgic_sync_hwstate(vcpu);
767 * Sync the timer hardware state before enabling interrupts as
768 * we don't want vtimer interrupts to race with syncing the
769 * timer virtual interrupt state.
771 if (static_branch_unlikely(&userspace_irqchip_in_use))
772 kvm_timer_sync_user(vcpu);
774 kvm_arch_vcpu_ctxsync_fp(vcpu);
777 * We may have taken a host interrupt in HYP mode (ie
778 * while executing the guest). This interrupt is still
779 * pending, as we haven't serviced it yet!
781 * We're now back in SVC mode, with interrupts
782 * disabled. Enabling the interrupts now will have
783 * the effect of taking the interrupt again, in SVC
789 * We do local_irq_enable() before calling guest_exit() so
790 * that if a timer interrupt hits while running the guest we
791 * account that tick as being spent in the guest. We enable
792 * preemption after calling guest_exit() so that if we get
793 * preempted we make sure ticks after that is not counted as
797 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
799 /* Exit types that need handling before we can be preempted */
800 handle_exit_early(vcpu, ret);
804 ret = handle_exit(vcpu, ret);
807 /* Tell userspace about in-kernel device output levels */
808 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
809 kvm_timer_update_run(vcpu);
810 kvm_pmu_update_run(vcpu);
813 kvm_sigset_deactivate(vcpu);
819 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
825 if (number == KVM_ARM_IRQ_CPU_IRQ)
826 bit_index = __ffs(HCR_VI);
827 else /* KVM_ARM_IRQ_CPU_FIQ */
828 bit_index = __ffs(HCR_VF);
830 hcr = vcpu_hcr(vcpu);
832 set = test_and_set_bit(bit_index, hcr);
834 set = test_and_clear_bit(bit_index, hcr);
837 * If we didn't change anything, no need to wake up or kick other CPUs
843 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
844 * trigger a world-switch round on the running physical CPU to set the
845 * virtual IRQ/FIQ fields in the HCR appropriately.
847 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
853 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
856 u32 irq = irq_level->irq;
857 unsigned int irq_type, vcpu_idx, irq_num;
858 int nrcpus = atomic_read(&kvm->online_vcpus);
859 struct kvm_vcpu *vcpu = NULL;
860 bool level = irq_level->level;
862 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
863 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
864 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
865 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
867 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
870 case KVM_ARM_IRQ_TYPE_CPU:
871 if (irqchip_in_kernel(kvm))
874 if (vcpu_idx >= nrcpus)
877 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
881 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
884 return vcpu_interrupt_line(vcpu, irq_num, level);
885 case KVM_ARM_IRQ_TYPE_PPI:
886 if (!irqchip_in_kernel(kvm))
889 if (vcpu_idx >= nrcpus)
892 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
896 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
899 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
900 case KVM_ARM_IRQ_TYPE_SPI:
901 if (!irqchip_in_kernel(kvm))
904 if (irq_num < VGIC_NR_PRIVATE_IRQS)
907 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
913 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
914 const struct kvm_vcpu_init *init)
917 int phys_target = kvm_target_cpu();
919 if (init->target != phys_target)
923 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
924 * use the same target.
926 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
929 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
930 for (i = 0; i < sizeof(init->features) * 8; i++) {
931 bool set = (init->features[i / 32] & (1 << (i % 32)));
933 if (set && i >= KVM_VCPU_MAX_FEATURES)
937 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
938 * use the same feature set.
940 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
941 test_bit(i, vcpu->arch.features) != set)
945 set_bit(i, vcpu->arch.features);
948 vcpu->arch.target = phys_target;
950 /* Now we know what it is, we can reset it. */
951 ret = kvm_reset_vcpu(vcpu);
953 vcpu->arch.target = -1;
954 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
960 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
961 struct kvm_vcpu_init *init)
965 ret = kvm_vcpu_set_target(vcpu, init);
970 * Ensure a rebooted VM will fault in RAM pages and detect if the
971 * guest MMU is turned off and flush the caches as needed.
973 * S2FWB enforces all memory accesses to RAM being cacheable,
974 * ensuring that the data side is always coherent. We still
975 * need to invalidate the I-cache though, as FWB does *not*
978 if (vcpu->arch.has_run_once) {
979 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
980 stage2_unmap_vm(vcpu->kvm);
982 __flush_icache_all();
985 vcpu_reset_hcr(vcpu);
988 * Handle the "start in power-off" case.
990 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
991 vcpu_power_off(vcpu);
993 vcpu->arch.power_off = false;
998 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
999 struct kvm_device_attr *attr)
1003 switch (attr->group) {
1005 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1012 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1013 struct kvm_device_attr *attr)
1017 switch (attr->group) {
1019 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1026 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1027 struct kvm_device_attr *attr)
1031 switch (attr->group) {
1033 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1040 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1041 struct kvm_vcpu_events *events)
1043 memset(events, 0, sizeof(*events));
1045 return __kvm_arm_vcpu_get_events(vcpu, events);
1048 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1049 struct kvm_vcpu_events *events)
1053 /* check whether the reserved field is zero */
1054 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1055 if (events->reserved[i])
1058 /* check whether the pad field is zero */
1059 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1060 if (events->exception.pad[i])
1063 return __kvm_arm_vcpu_set_events(vcpu, events);
1066 long kvm_arch_vcpu_ioctl(struct file *filp,
1067 unsigned int ioctl, unsigned long arg)
1069 struct kvm_vcpu *vcpu = filp->private_data;
1070 void __user *argp = (void __user *)arg;
1071 struct kvm_device_attr attr;
1075 case KVM_ARM_VCPU_INIT: {
1076 struct kvm_vcpu_init init;
1079 if (copy_from_user(&init, argp, sizeof(init)))
1082 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1085 case KVM_SET_ONE_REG:
1086 case KVM_GET_ONE_REG: {
1087 struct kvm_one_reg reg;
1090 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1094 if (copy_from_user(®, argp, sizeof(reg)))
1097 if (ioctl == KVM_SET_ONE_REG)
1098 r = kvm_arm_set_reg(vcpu, ®);
1100 r = kvm_arm_get_reg(vcpu, ®);
1103 case KVM_GET_REG_LIST: {
1104 struct kvm_reg_list __user *user_list = argp;
1105 struct kvm_reg_list reg_list;
1109 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1113 if (!kvm_arm_vcpu_is_finalized(vcpu))
1117 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1120 reg_list.n = kvm_arm_num_regs(vcpu);
1121 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1126 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1129 case KVM_SET_DEVICE_ATTR: {
1131 if (copy_from_user(&attr, argp, sizeof(attr)))
1133 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1136 case KVM_GET_DEVICE_ATTR: {
1138 if (copy_from_user(&attr, argp, sizeof(attr)))
1140 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1143 case KVM_HAS_DEVICE_ATTR: {
1145 if (copy_from_user(&attr, argp, sizeof(attr)))
1147 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1150 case KVM_GET_VCPU_EVENTS: {
1151 struct kvm_vcpu_events events;
1153 if (kvm_arm_vcpu_get_events(vcpu, &events))
1156 if (copy_to_user(argp, &events, sizeof(events)))
1161 case KVM_SET_VCPU_EVENTS: {
1162 struct kvm_vcpu_events events;
1164 if (copy_from_user(&events, argp, sizeof(events)))
1167 return kvm_arm_vcpu_set_events(vcpu, &events);
1169 case KVM_ARM_VCPU_FINALIZE: {
1172 if (!kvm_vcpu_initialized(vcpu))
1175 if (get_user(what, (const int __user *)argp))
1178 return kvm_arm_vcpu_finalize(vcpu, what);
1187 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1192 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1193 struct kvm_memory_slot *memslot)
1195 kvm_flush_remote_tlbs(kvm);
1198 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1199 struct kvm_arm_device_addr *dev_addr)
1201 unsigned long dev_id, type;
1203 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1204 KVM_ARM_DEVICE_ID_SHIFT;
1205 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1206 KVM_ARM_DEVICE_TYPE_SHIFT;
1209 case KVM_ARM_DEVICE_VGIC_V2:
1212 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1218 long kvm_arch_vm_ioctl(struct file *filp,
1219 unsigned int ioctl, unsigned long arg)
1221 struct kvm *kvm = filp->private_data;
1222 void __user *argp = (void __user *)arg;
1225 case KVM_CREATE_IRQCHIP: {
1229 mutex_lock(&kvm->lock);
1230 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1231 mutex_unlock(&kvm->lock);
1234 case KVM_ARM_SET_DEVICE_ADDR: {
1235 struct kvm_arm_device_addr dev_addr;
1237 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1239 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1241 case KVM_ARM_PREFERRED_TARGET: {
1243 struct kvm_vcpu_init init;
1245 err = kvm_vcpu_preferred_target(&init);
1249 if (copy_to_user(argp, &init, sizeof(init)))
1259 static void cpu_init_hyp_mode(void)
1261 phys_addr_t pgd_ptr;
1262 unsigned long hyp_stack_ptr;
1263 unsigned long vector_ptr;
1264 unsigned long tpidr_el2;
1266 /* Switch from the HYP stub to our own HYP init vector */
1267 __hyp_set_vectors(kvm_get_idmap_vector());
1270 * Calculate the raw per-cpu offset without a translation from the
1271 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1272 * so that we can use adr_l to access per-cpu variables in EL2.
1274 tpidr_el2 = ((unsigned long)this_cpu_ptr(&kvm_host_data) -
1275 (unsigned long)kvm_ksym_ref(&kvm_host_data));
1277 pgd_ptr = kvm_mmu_get_httbr();
1278 hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE;
1279 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1282 * Call initialization code, and switch to the full blown HYP code.
1283 * If the cpucaps haven't been finalized yet, something has gone very
1284 * wrong, and hyp will crash and burn when it uses any
1285 * cpus_have_const_cap() wrapper.
1287 BUG_ON(!system_capabilities_finalized());
1288 __kvm_call_hyp((void *)pgd_ptr, hyp_stack_ptr, vector_ptr, tpidr_el2);
1291 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1294 if (this_cpu_has_cap(ARM64_SSBS) &&
1295 arm64_get_ssbd_state() == ARM64_SSBD_FORCE_DISABLE) {
1296 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1300 static void cpu_hyp_reset(void)
1302 if (!is_kernel_in_hyp_mode())
1303 __hyp_reset_vectors();
1306 static void cpu_hyp_reinit(void)
1308 kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1312 if (is_kernel_in_hyp_mode())
1313 kvm_timer_init_vhe();
1315 cpu_init_hyp_mode();
1317 kvm_arm_init_debug();
1320 kvm_vgic_init_cpu_hardware();
1323 static void _kvm_arch_hardware_enable(void *discard)
1325 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1327 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1331 int kvm_arch_hardware_enable(void)
1333 _kvm_arch_hardware_enable(NULL);
1337 static void _kvm_arch_hardware_disable(void *discard)
1339 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1341 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1345 void kvm_arch_hardware_disable(void)
1347 _kvm_arch_hardware_disable(NULL);
1350 #ifdef CONFIG_CPU_PM
1351 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1356 * kvm_arm_hardware_enabled is left with its old value over
1357 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1362 if (__this_cpu_read(kvm_arm_hardware_enabled))
1364 * don't update kvm_arm_hardware_enabled here
1365 * so that the hardware will be re-enabled
1366 * when we resume. See below.
1371 case CPU_PM_ENTER_FAILED:
1373 if (__this_cpu_read(kvm_arm_hardware_enabled))
1374 /* The hardware was enabled before suspend. */
1384 static struct notifier_block hyp_init_cpu_pm_nb = {
1385 .notifier_call = hyp_init_cpu_pm_notifier,
1388 static void __init hyp_cpu_pm_init(void)
1390 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1392 static void __init hyp_cpu_pm_exit(void)
1394 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1397 static inline void hyp_cpu_pm_init(void)
1400 static inline void hyp_cpu_pm_exit(void)
1405 static int init_common_resources(void)
1407 return kvm_set_ipa_limit();
1410 static int init_subsystems(void)
1415 * Enable hardware so that subsystem initialisation can access EL2.
1417 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1420 * Register CPU lower-power notifier
1425 * Init HYP view of VGIC
1427 err = kvm_vgic_hyp_init();
1430 vgic_present = true;
1434 vgic_present = false;
1442 * Init HYP architected timer support
1444 err = kvm_timer_hyp_init(vgic_present);
1449 kvm_coproc_table_init();
1452 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1457 static void teardown_hyp_mode(void)
1462 for_each_possible_cpu(cpu)
1463 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1467 * Inits Hyp-mode on all online CPUs
1469 static int init_hyp_mode(void)
1475 * Allocate Hyp PGD and setup Hyp identity mapping
1477 err = kvm_mmu_init();
1482 * Allocate stack pages for Hypervisor-mode
1484 for_each_possible_cpu(cpu) {
1485 unsigned long stack_page;
1487 stack_page = __get_free_page(GFP_KERNEL);
1493 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1497 * Map the Hyp-code called directly from the host
1499 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1500 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1502 kvm_err("Cannot map world-switch code\n");
1506 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1507 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1509 kvm_err("Cannot map rodata section\n");
1513 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1514 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1516 kvm_err("Cannot map bss section\n");
1520 err = kvm_map_vectors();
1522 kvm_err("Cannot map vectors\n");
1527 * Map the Hyp stack pages
1529 for_each_possible_cpu(cpu) {
1530 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1531 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1535 kvm_err("Cannot map hyp stack\n");
1540 for_each_possible_cpu(cpu) {
1541 kvm_host_data_t *cpu_data;
1543 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1544 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1547 kvm_err("Cannot map host CPU state: %d\n", err);
1552 err = hyp_map_aux_data();
1554 kvm_err("Cannot map host auxiliary data: %d\n", err);
1559 teardown_hyp_mode();
1560 kvm_err("error initializing Hyp mode: %d\n", err);
1564 static void check_kvm_target_cpu(void *ret)
1566 *(int *)ret = kvm_target_cpu();
1569 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1571 struct kvm_vcpu *vcpu;
1574 mpidr &= MPIDR_HWID_BITMASK;
1575 kvm_for_each_vcpu(i, vcpu, kvm) {
1576 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1582 bool kvm_arch_has_irq_bypass(void)
1587 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1588 struct irq_bypass_producer *prod)
1590 struct kvm_kernel_irqfd *irqfd =
1591 container_of(cons, struct kvm_kernel_irqfd, consumer);
1593 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1596 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1597 struct irq_bypass_producer *prod)
1599 struct kvm_kernel_irqfd *irqfd =
1600 container_of(cons, struct kvm_kernel_irqfd, consumer);
1602 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1606 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1608 struct kvm_kernel_irqfd *irqfd =
1609 container_of(cons, struct kvm_kernel_irqfd, consumer);
1611 kvm_arm_halt_guest(irqfd->kvm);
1614 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1616 struct kvm_kernel_irqfd *irqfd =
1617 container_of(cons, struct kvm_kernel_irqfd, consumer);
1619 kvm_arm_resume_guest(irqfd->kvm);
1623 * Initialize Hyp-mode and memory mappings on all CPUs.
1625 int kvm_arch_init(void *opaque)
1631 if (!is_hyp_mode_available()) {
1632 kvm_info("HYP mode not available\n");
1636 in_hyp_mode = is_kernel_in_hyp_mode();
1638 if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1639 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1643 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE))
1644 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
1645 "Only trusted guests should be used on this system.\n");
1647 for_each_online_cpu(cpu) {
1648 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1650 kvm_err("Error, CPU %d not supported!\n", cpu);
1655 err = init_common_resources();
1659 err = kvm_arm_init_sve();
1664 err = init_hyp_mode();
1669 err = init_subsystems();
1674 kvm_info("VHE mode initialized successfully\n");
1676 kvm_info("Hyp mode initialized successfully\n");
1683 teardown_hyp_mode();
1688 /* NOP: Compiling as a module not supported */
1689 void kvm_arch_exit(void)
1691 kvm_perf_teardown();
1694 static int arm_init(void)
1696 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1700 module_init(arm_init);