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;
105 static void set_default_csv2(struct kvm *kvm)
108 * The default is to expose CSV2 == 1 if the HW isn't affected.
109 * Although this is a per-CPU feature, we make it global because
110 * asymmetric systems are just a nuisance.
112 * Userspace can override this as long as it doesn't promise
115 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED)
116 kvm->arch.pfr0_csv2 = 1;
120 * kvm_arch_init_vm - initializes a VM data structure
121 * @kvm: pointer to the KVM struct
123 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
127 ret = kvm_arm_setup_stage2(kvm, type);
131 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu);
135 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
137 goto out_free_stage2_pgd;
139 kvm_vgic_early_init(kvm);
141 /* The maximum number of VCPUs is limited by the host's GIC model */
142 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
144 set_default_csv2(kvm);
148 kvm_free_stage2_pgd(&kvm->arch.mmu);
152 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
154 return VM_FAULT_SIGBUS;
159 * kvm_arch_destroy_vm - destroy the VM data structure
160 * @kvm: pointer to the KVM struct
162 void kvm_arch_destroy_vm(struct kvm *kvm)
166 bitmap_free(kvm->arch.pmu_filter);
168 kvm_vgic_destroy(kvm);
170 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
172 kvm_vcpu_destroy(kvm->vcpus[i]);
173 kvm->vcpus[i] = NULL;
176 atomic_set(&kvm->online_vcpus, 0);
179 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
183 case KVM_CAP_IRQCHIP:
186 case KVM_CAP_IOEVENTFD:
187 case KVM_CAP_DEVICE_CTRL:
188 case KVM_CAP_USER_MEMORY:
189 case KVM_CAP_SYNC_MMU:
190 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
191 case KVM_CAP_ONE_REG:
192 case KVM_CAP_ARM_PSCI:
193 case KVM_CAP_ARM_PSCI_0_2:
194 case KVM_CAP_READONLY_MEM:
195 case KVM_CAP_MP_STATE:
196 case KVM_CAP_IMMEDIATE_EXIT:
197 case KVM_CAP_VCPU_EVENTS:
198 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
199 case KVM_CAP_ARM_NISV_TO_USER:
200 case KVM_CAP_ARM_INJECT_EXT_DABT:
203 case KVM_CAP_ARM_SET_DEVICE_ADDR:
206 case KVM_CAP_NR_VCPUS:
207 r = num_online_cpus();
209 case KVM_CAP_MAX_VCPUS:
210 case KVM_CAP_MAX_VCPU_ID:
212 r = kvm->arch.max_vcpus;
214 r = kvm_arm_default_max_vcpus();
216 case KVM_CAP_MSI_DEVID:
220 r = kvm->arch.vgic.msis_require_devid;
222 case KVM_CAP_ARM_USER_IRQ:
224 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
225 * (bump this number if adding more devices)
229 case KVM_CAP_STEAL_TIME:
230 r = kvm_arm_pvtime_supported();
233 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
239 long kvm_arch_dev_ioctl(struct file *filp,
240 unsigned int ioctl, unsigned long arg)
245 struct kvm *kvm_arch_alloc_vm(void)
248 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
250 return vzalloc(sizeof(struct kvm));
253 void kvm_arch_free_vm(struct kvm *kvm)
261 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
263 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
266 if (id >= kvm->arch.max_vcpus)
272 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
276 /* Force users to call KVM_ARM_VCPU_INIT */
277 vcpu->arch.target = -1;
278 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
280 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
282 /* Set up the timer */
283 kvm_timer_vcpu_init(vcpu);
285 kvm_pmu_vcpu_init(vcpu);
287 kvm_arm_reset_debug_ptr(vcpu);
289 kvm_arm_pvtime_vcpu_init(&vcpu->arch);
291 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
293 err = kvm_vgic_vcpu_init(vcpu);
297 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
300 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
304 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
306 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
307 static_branch_dec(&userspace_irqchip_in_use);
309 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
310 kvm_timer_vcpu_terminate(vcpu);
311 kvm_pmu_vcpu_destroy(vcpu);
313 kvm_arm_vcpu_destroy(vcpu);
316 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
318 return kvm_timer_is_pending(vcpu);
321 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
324 * If we're about to block (most likely because we've just hit a
325 * WFI), we need to sync back the state of the GIC CPU interface
326 * so that we have the latest PMR and group enables. This ensures
327 * that kvm_arch_vcpu_runnable has up-to-date data to decide
328 * whether we have pending interrupts.
330 * For the same reason, we want to tell GICv4 that we need
331 * doorbells to be signalled, should an interrupt become pending.
334 kvm_vgic_vmcr_sync(vcpu);
335 vgic_v4_put(vcpu, true);
339 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
346 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
348 struct kvm_s2_mmu *mmu;
351 mmu = vcpu->arch.hw_mmu;
352 last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
355 * We might get preempted before the vCPU actually runs, but
356 * over-invalidation doesn't affect correctness.
358 if (*last_ran != vcpu->vcpu_id) {
359 kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu);
360 *last_ran = vcpu->vcpu_id;
366 kvm_timer_vcpu_load(vcpu);
368 kvm_vcpu_load_sysregs_vhe(vcpu);
369 kvm_arch_vcpu_load_fp(vcpu);
370 kvm_vcpu_pmu_restore_guest(vcpu);
371 if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
372 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
374 if (single_task_running())
375 vcpu_clear_wfx_traps(vcpu);
377 vcpu_set_wfx_traps(vcpu);
379 if (vcpu_has_ptrauth(vcpu))
380 vcpu_ptrauth_disable(vcpu);
383 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
385 kvm_arch_vcpu_put_fp(vcpu);
387 kvm_vcpu_put_sysregs_vhe(vcpu);
388 kvm_timer_vcpu_put(vcpu);
390 kvm_vcpu_pmu_restore_host(vcpu);
395 static void vcpu_power_off(struct kvm_vcpu *vcpu)
397 vcpu->arch.power_off = true;
398 kvm_make_request(KVM_REQ_SLEEP, vcpu);
402 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
403 struct kvm_mp_state *mp_state)
405 if (vcpu->arch.power_off)
406 mp_state->mp_state = KVM_MP_STATE_STOPPED;
408 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
413 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
414 struct kvm_mp_state *mp_state)
418 switch (mp_state->mp_state) {
419 case KVM_MP_STATE_RUNNABLE:
420 vcpu->arch.power_off = false;
422 case KVM_MP_STATE_STOPPED:
423 vcpu_power_off(vcpu);
433 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
434 * @v: The VCPU pointer
436 * If the guest CPU is not waiting for interrupts or an interrupt line is
437 * asserted, the CPU is by definition runnable.
439 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
441 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
442 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
443 && !v->arch.power_off && !v->arch.pause);
446 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
448 return vcpu_mode_priv(vcpu);
451 /* Just ensure a guest exit from a particular CPU */
452 static void exit_vm_noop(void *info)
456 void force_vm_exit(const cpumask_t *mask)
459 smp_call_function_many(mask, exit_vm_noop, NULL, true);
464 * need_new_vmid_gen - check that the VMID is still valid
465 * @vmid: The VMID to check
467 * return true if there is a new generation of VMIDs being used
469 * The hardware supports a limited set of values with the value zero reserved
470 * for the host, so we check if an assigned value belongs to a previous
471 * generation, which requires us to assign a new value. If we're the first to
472 * use a VMID for the new generation, we must flush necessary caches and TLBs
475 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
477 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
478 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
479 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
483 * update_vmid - Update the vmid with a valid VMID for the current generation
484 * @vmid: The stage-2 VMID information struct
486 static void update_vmid(struct kvm_vmid *vmid)
488 if (!need_new_vmid_gen(vmid))
491 spin_lock(&kvm_vmid_lock);
494 * We need to re-check the vmid_gen here to ensure that if another vcpu
495 * already allocated a valid vmid for this vm, then this vcpu should
498 if (!need_new_vmid_gen(vmid)) {
499 spin_unlock(&kvm_vmid_lock);
503 /* First user of a new VMID generation? */
504 if (unlikely(kvm_next_vmid == 0)) {
505 atomic64_inc(&kvm_vmid_gen);
509 * On SMP we know no other CPUs can use this CPU's or each
510 * other's VMID after force_vm_exit returns since the
511 * kvm_vmid_lock blocks them from reentry to the guest.
513 force_vm_exit(cpu_all_mask);
515 * Now broadcast TLB + ICACHE invalidation over the inner
516 * shareable domain to make sure all data structures are
519 kvm_call_hyp(__kvm_flush_vm_context);
522 vmid->vmid = kvm_next_vmid;
524 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
527 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
529 spin_unlock(&kvm_vmid_lock);
532 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
534 struct kvm *kvm = vcpu->kvm;
537 if (likely(vcpu->arch.has_run_once))
540 if (!kvm_arm_vcpu_is_finalized(vcpu))
543 vcpu->arch.has_run_once = true;
545 if (likely(irqchip_in_kernel(kvm))) {
547 * Map the VGIC hardware resources before running a vcpu the
548 * first time on this VM.
550 if (unlikely(!vgic_ready(kvm))) {
551 ret = kvm_vgic_map_resources(kvm);
557 * Tell the rest of the code that there are userspace irqchip
560 static_branch_inc(&userspace_irqchip_in_use);
563 ret = kvm_timer_enable(vcpu);
567 ret = kvm_arm_pmu_v3_enable(vcpu);
572 bool kvm_arch_intc_initialized(struct kvm *kvm)
574 return vgic_initialized(kvm);
577 void kvm_arm_halt_guest(struct kvm *kvm)
580 struct kvm_vcpu *vcpu;
582 kvm_for_each_vcpu(i, vcpu, kvm)
583 vcpu->arch.pause = true;
584 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
587 void kvm_arm_resume_guest(struct kvm *kvm)
590 struct kvm_vcpu *vcpu;
592 kvm_for_each_vcpu(i, vcpu, kvm) {
593 vcpu->arch.pause = false;
594 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu));
598 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
600 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
602 rcuwait_wait_event(wait,
603 (!vcpu->arch.power_off) &&(!vcpu->arch.pause),
606 if (vcpu->arch.power_off || vcpu->arch.pause) {
607 /* Awaken to handle a signal, request we sleep again later. */
608 kvm_make_request(KVM_REQ_SLEEP, vcpu);
612 * Make sure we will observe a potential reset request if we've
613 * observed a change to the power state. Pairs with the smp_wmb() in
614 * kvm_psci_vcpu_on().
619 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
621 return vcpu->arch.target >= 0;
624 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
626 if (kvm_request_pending(vcpu)) {
627 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
628 vcpu_req_sleep(vcpu);
630 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
631 kvm_reset_vcpu(vcpu);
634 * Clear IRQ_PENDING requests that were made to guarantee
635 * that a VCPU sees new virtual interrupts.
637 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
639 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
640 kvm_update_stolen_time(vcpu);
642 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
643 /* The distributor enable bits were changed */
645 vgic_v4_put(vcpu, false);
653 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
654 * @vcpu: The VCPU pointer
656 * This function is called through the VCPU_RUN ioctl called from user space. It
657 * will execute VM code in a loop until the time slice for the process is used
658 * or some emulation is needed from user space in which case the function will
659 * return with return value 0 and with the kvm_run structure filled in with the
660 * required data for the requested emulation.
662 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
664 struct kvm_run *run = vcpu->run;
667 if (unlikely(!kvm_vcpu_initialized(vcpu)))
670 ret = kvm_vcpu_first_run_init(vcpu);
674 if (run->exit_reason == KVM_EXIT_MMIO) {
675 ret = kvm_handle_mmio_return(vcpu);
680 if (run->immediate_exit)
685 kvm_sigset_activate(vcpu);
688 run->exit_reason = KVM_EXIT_UNKNOWN;
691 * Check conditions before entering the guest
695 update_vmid(&vcpu->arch.hw_mmu->vmid);
697 check_vcpu_requests(vcpu);
700 * Preparing the interrupts to be injected also
701 * involves poking the GIC, which must be done in a
702 * non-preemptible context.
706 kvm_pmu_flush_hwstate(vcpu);
710 kvm_vgic_flush_hwstate(vcpu);
713 * Exit if we have a signal pending so that we can deliver the
714 * signal to user space.
716 if (signal_pending(current)) {
718 run->exit_reason = KVM_EXIT_INTR;
722 * If we're using a userspace irqchip, then check if we need
723 * to tell a userspace irqchip about timer or PMU level
724 * changes and if so, exit to userspace (the actual level
725 * state gets updated in kvm_timer_update_run and
726 * kvm_pmu_update_run below).
728 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
729 if (kvm_timer_should_notify_user(vcpu) ||
730 kvm_pmu_should_notify_user(vcpu)) {
732 run->exit_reason = KVM_EXIT_INTR;
737 * Ensure we set mode to IN_GUEST_MODE after we disable
738 * interrupts and before the final VCPU requests check.
739 * See the comment in kvm_vcpu_exiting_guest_mode() and
740 * Documentation/virt/kvm/vcpu-requests.rst
742 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
744 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
745 kvm_request_pending(vcpu)) {
746 vcpu->mode = OUTSIDE_GUEST_MODE;
747 isb(); /* Ensure work in x_flush_hwstate is committed */
748 kvm_pmu_sync_hwstate(vcpu);
749 if (static_branch_unlikely(&userspace_irqchip_in_use))
750 kvm_timer_sync_user(vcpu);
751 kvm_vgic_sync_hwstate(vcpu);
757 kvm_arm_setup_debug(vcpu);
759 /**************************************************************
762 trace_kvm_entry(*vcpu_pc(vcpu));
763 guest_enter_irqoff();
765 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
767 vcpu->mode = OUTSIDE_GUEST_MODE;
771 *************************************************************/
773 kvm_arm_clear_debug(vcpu);
776 * We must sync the PMU state before the vgic state so
777 * that the vgic can properly sample the updated state of the
780 kvm_pmu_sync_hwstate(vcpu);
783 * Sync the vgic state before syncing the timer state because
784 * the timer code needs to know if the virtual timer
785 * interrupts are active.
787 kvm_vgic_sync_hwstate(vcpu);
790 * Sync the timer hardware state before enabling interrupts as
791 * we don't want vtimer interrupts to race with syncing the
792 * timer virtual interrupt state.
794 if (static_branch_unlikely(&userspace_irqchip_in_use))
795 kvm_timer_sync_user(vcpu);
797 kvm_arch_vcpu_ctxsync_fp(vcpu);
800 * We may have taken a host interrupt in HYP mode (ie
801 * while executing the guest). This interrupt is still
802 * pending, as we haven't serviced it yet!
804 * We're now back in SVC mode, with interrupts
805 * disabled. Enabling the interrupts now will have
806 * the effect of taking the interrupt again, in SVC
812 * We do local_irq_enable() before calling guest_exit() so
813 * that if a timer interrupt hits while running the guest we
814 * account that tick as being spent in the guest. We enable
815 * preemption after calling guest_exit() so that if we get
816 * preempted we make sure ticks after that is not counted as
820 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
822 /* Exit types that need handling before we can be preempted */
823 handle_exit_early(vcpu, ret);
828 * The ARMv8 architecture doesn't give the hypervisor
829 * a mechanism to prevent a guest from dropping to AArch32 EL0
830 * if implemented by the CPU. If we spot the guest in such
831 * state and that we decided it wasn't supposed to do so (like
832 * with the asymmetric AArch32 case), return to userspace with
835 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) {
837 * As we have caught the guest red-handed, decide that
838 * it isn't fit for purpose anymore by making the vcpu
839 * invalid. The VMM can try and fix it by issuing a
840 * KVM_ARM_VCPU_INIT if it really wants to.
842 vcpu->arch.target = -1;
843 ret = ARM_EXCEPTION_IL;
846 ret = handle_exit(vcpu, ret);
849 /* Tell userspace about in-kernel device output levels */
850 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
851 kvm_timer_update_run(vcpu);
852 kvm_pmu_update_run(vcpu);
855 kvm_sigset_deactivate(vcpu);
861 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
867 if (number == KVM_ARM_IRQ_CPU_IRQ)
868 bit_index = __ffs(HCR_VI);
869 else /* KVM_ARM_IRQ_CPU_FIQ */
870 bit_index = __ffs(HCR_VF);
872 hcr = vcpu_hcr(vcpu);
874 set = test_and_set_bit(bit_index, hcr);
876 set = test_and_clear_bit(bit_index, hcr);
879 * If we didn't change anything, no need to wake up or kick other CPUs
885 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
886 * trigger a world-switch round on the running physical CPU to set the
887 * virtual IRQ/FIQ fields in the HCR appropriately.
889 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
895 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
898 u32 irq = irq_level->irq;
899 unsigned int irq_type, vcpu_idx, irq_num;
900 int nrcpus = atomic_read(&kvm->online_vcpus);
901 struct kvm_vcpu *vcpu = NULL;
902 bool level = irq_level->level;
904 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
905 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
906 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
907 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
909 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
912 case KVM_ARM_IRQ_TYPE_CPU:
913 if (irqchip_in_kernel(kvm))
916 if (vcpu_idx >= nrcpus)
919 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
923 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
926 return vcpu_interrupt_line(vcpu, irq_num, level);
927 case KVM_ARM_IRQ_TYPE_PPI:
928 if (!irqchip_in_kernel(kvm))
931 if (vcpu_idx >= nrcpus)
934 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
938 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
941 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
942 case KVM_ARM_IRQ_TYPE_SPI:
943 if (!irqchip_in_kernel(kvm))
946 if (irq_num < VGIC_NR_PRIVATE_IRQS)
949 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
955 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
956 const struct kvm_vcpu_init *init)
959 int phys_target = kvm_target_cpu();
961 if (init->target != phys_target)
965 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
966 * use the same target.
968 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
971 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
972 for (i = 0; i < sizeof(init->features) * 8; i++) {
973 bool set = (init->features[i / 32] & (1 << (i % 32)));
975 if (set && i >= KVM_VCPU_MAX_FEATURES)
979 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
980 * use the same feature set.
982 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
983 test_bit(i, vcpu->arch.features) != set)
987 set_bit(i, vcpu->arch.features);
990 vcpu->arch.target = phys_target;
992 /* Now we know what it is, we can reset it. */
993 ret = kvm_reset_vcpu(vcpu);
995 vcpu->arch.target = -1;
996 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1002 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1003 struct kvm_vcpu_init *init)
1007 ret = kvm_vcpu_set_target(vcpu, init);
1012 * Ensure a rebooted VM will fault in RAM pages and detect if the
1013 * guest MMU is turned off and flush the caches as needed.
1015 * S2FWB enforces all memory accesses to RAM being cacheable,
1016 * ensuring that the data side is always coherent. We still
1017 * need to invalidate the I-cache though, as FWB does *not*
1018 * imply CTR_EL0.DIC.
1020 if (vcpu->arch.has_run_once) {
1021 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1022 stage2_unmap_vm(vcpu->kvm);
1024 __flush_icache_all();
1027 vcpu_reset_hcr(vcpu);
1030 * Handle the "start in power-off" case.
1032 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1033 vcpu_power_off(vcpu);
1035 vcpu->arch.power_off = false;
1040 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1041 struct kvm_device_attr *attr)
1045 switch (attr->group) {
1047 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1054 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1055 struct kvm_device_attr *attr)
1059 switch (attr->group) {
1061 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1068 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1069 struct kvm_device_attr *attr)
1073 switch (attr->group) {
1075 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1082 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1083 struct kvm_vcpu_events *events)
1085 memset(events, 0, sizeof(*events));
1087 return __kvm_arm_vcpu_get_events(vcpu, events);
1090 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1091 struct kvm_vcpu_events *events)
1095 /* check whether the reserved field is zero */
1096 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1097 if (events->reserved[i])
1100 /* check whether the pad field is zero */
1101 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1102 if (events->exception.pad[i])
1105 return __kvm_arm_vcpu_set_events(vcpu, events);
1108 long kvm_arch_vcpu_ioctl(struct file *filp,
1109 unsigned int ioctl, unsigned long arg)
1111 struct kvm_vcpu *vcpu = filp->private_data;
1112 void __user *argp = (void __user *)arg;
1113 struct kvm_device_attr attr;
1117 case KVM_ARM_VCPU_INIT: {
1118 struct kvm_vcpu_init init;
1121 if (copy_from_user(&init, argp, sizeof(init)))
1124 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1127 case KVM_SET_ONE_REG:
1128 case KVM_GET_ONE_REG: {
1129 struct kvm_one_reg reg;
1132 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1136 if (copy_from_user(®, argp, sizeof(reg)))
1139 if (ioctl == KVM_SET_ONE_REG)
1140 r = kvm_arm_set_reg(vcpu, ®);
1142 r = kvm_arm_get_reg(vcpu, ®);
1145 case KVM_GET_REG_LIST: {
1146 struct kvm_reg_list __user *user_list = argp;
1147 struct kvm_reg_list reg_list;
1151 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1155 if (!kvm_arm_vcpu_is_finalized(vcpu))
1159 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1162 reg_list.n = kvm_arm_num_regs(vcpu);
1163 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1168 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1171 case KVM_SET_DEVICE_ATTR: {
1173 if (copy_from_user(&attr, argp, sizeof(attr)))
1175 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1178 case KVM_GET_DEVICE_ATTR: {
1180 if (copy_from_user(&attr, argp, sizeof(attr)))
1182 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1185 case KVM_HAS_DEVICE_ATTR: {
1187 if (copy_from_user(&attr, argp, sizeof(attr)))
1189 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1192 case KVM_GET_VCPU_EVENTS: {
1193 struct kvm_vcpu_events events;
1195 if (kvm_arm_vcpu_get_events(vcpu, &events))
1198 if (copy_to_user(argp, &events, sizeof(events)))
1203 case KVM_SET_VCPU_EVENTS: {
1204 struct kvm_vcpu_events events;
1206 if (copy_from_user(&events, argp, sizeof(events)))
1209 return kvm_arm_vcpu_set_events(vcpu, &events);
1211 case KVM_ARM_VCPU_FINALIZE: {
1214 if (!kvm_vcpu_initialized(vcpu))
1217 if (get_user(what, (const int __user *)argp))
1220 return kvm_arm_vcpu_finalize(vcpu, what);
1229 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1234 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1235 struct kvm_memory_slot *memslot)
1237 kvm_flush_remote_tlbs(kvm);
1240 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1241 struct kvm_arm_device_addr *dev_addr)
1243 unsigned long dev_id, type;
1245 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1246 KVM_ARM_DEVICE_ID_SHIFT;
1247 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1248 KVM_ARM_DEVICE_TYPE_SHIFT;
1251 case KVM_ARM_DEVICE_VGIC_V2:
1254 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1260 long kvm_arch_vm_ioctl(struct file *filp,
1261 unsigned int ioctl, unsigned long arg)
1263 struct kvm *kvm = filp->private_data;
1264 void __user *argp = (void __user *)arg;
1267 case KVM_CREATE_IRQCHIP: {
1271 mutex_lock(&kvm->lock);
1272 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1273 mutex_unlock(&kvm->lock);
1276 case KVM_ARM_SET_DEVICE_ADDR: {
1277 struct kvm_arm_device_addr dev_addr;
1279 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1281 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1283 case KVM_ARM_PREFERRED_TARGET: {
1285 struct kvm_vcpu_init init;
1287 err = kvm_vcpu_preferred_target(&init);
1291 if (copy_to_user(argp, &init, sizeof(init)))
1301 static unsigned long nvhe_percpu_size(void)
1303 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1304 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1307 static unsigned long nvhe_percpu_order(void)
1309 unsigned long size = nvhe_percpu_size();
1311 return size ? get_order(size) : 0;
1314 static int kvm_map_vectors(void)
1317 * SV2 = ARM64_SPECTRE_V2
1318 * HEL2 = ARM64_HARDEN_EL2_VECTORS
1320 * !SV2 + !HEL2 -> use direct vectors
1321 * SV2 + !HEL2 -> use hardened vectors in place
1322 * !SV2 + HEL2 -> allocate one vector slot and use exec mapping
1323 * SV2 + HEL2 -> use hardened vectors and use exec mapping
1325 if (cpus_have_const_cap(ARM64_SPECTRE_V2)) {
1326 __kvm_bp_vect_base = kvm_ksym_ref(__bp_harden_hyp_vecs);
1327 __kvm_bp_vect_base = kern_hyp_va(__kvm_bp_vect_base);
1330 if (cpus_have_const_cap(ARM64_HARDEN_EL2_VECTORS)) {
1331 phys_addr_t vect_pa = __pa_symbol(__bp_harden_hyp_vecs);
1332 unsigned long size = __BP_HARDEN_HYP_VECS_SZ;
1335 * Always allocate a spare vector slot, as we don't
1336 * know yet which CPUs have a BP hardening slot that
1339 __kvm_harden_el2_vector_slot = atomic_inc_return(&arm64_el2_vector_last_slot);
1340 BUG_ON(__kvm_harden_el2_vector_slot >= BP_HARDEN_EL2_SLOTS);
1341 return create_hyp_exec_mappings(vect_pa, size,
1342 &__kvm_bp_vect_base);
1348 static void cpu_init_hyp_mode(void)
1350 phys_addr_t pgd_ptr;
1351 unsigned long hyp_stack_ptr;
1352 unsigned long vector_ptr;
1353 unsigned long tpidr_el2;
1354 struct arm_smccc_res res;
1356 /* Switch from the HYP stub to our own HYP init vector */
1357 __hyp_set_vectors(kvm_get_idmap_vector());
1360 * Calculate the raw per-cpu offset without a translation from the
1361 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1362 * so that we can use adr_l to access per-cpu variables in EL2.
1364 tpidr_el2 = (unsigned long)this_cpu_ptr_nvhe_sym(__per_cpu_start) -
1365 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
1367 pgd_ptr = kvm_mmu_get_httbr();
1368 hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE;
1369 hyp_stack_ptr = kern_hyp_va(hyp_stack_ptr);
1370 vector_ptr = (unsigned long)kern_hyp_va(kvm_ksym_ref(__kvm_hyp_host_vector));
1373 * Call initialization code, and switch to the full blown HYP code.
1374 * If the cpucaps haven't been finalized yet, something has gone very
1375 * wrong, and hyp will crash and burn when it uses any
1376 * cpus_have_const_cap() wrapper.
1378 BUG_ON(!system_capabilities_finalized());
1379 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init),
1380 pgd_ptr, tpidr_el2, hyp_stack_ptr, vector_ptr, &res);
1381 WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
1384 * Disabling SSBD on a non-VHE system requires us to enable SSBS
1387 if (this_cpu_has_cap(ARM64_SSBS) &&
1388 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
1389 kvm_call_hyp_nvhe(__kvm_enable_ssbs);
1393 static void cpu_hyp_reset(void)
1395 if (!is_kernel_in_hyp_mode())
1396 __hyp_reset_vectors();
1399 static void cpu_hyp_reinit(void)
1401 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt);
1405 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)kvm_get_hyp_vector();
1407 if (is_kernel_in_hyp_mode())
1408 kvm_timer_init_vhe();
1410 cpu_init_hyp_mode();
1412 kvm_arm_init_debug();
1415 kvm_vgic_init_cpu_hardware();
1418 static void _kvm_arch_hardware_enable(void *discard)
1420 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1422 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1426 int kvm_arch_hardware_enable(void)
1428 _kvm_arch_hardware_enable(NULL);
1432 static void _kvm_arch_hardware_disable(void *discard)
1434 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1436 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1440 void kvm_arch_hardware_disable(void)
1442 _kvm_arch_hardware_disable(NULL);
1445 #ifdef CONFIG_CPU_PM
1446 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1451 * kvm_arm_hardware_enabled is left with its old value over
1452 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1457 if (__this_cpu_read(kvm_arm_hardware_enabled))
1459 * don't update kvm_arm_hardware_enabled here
1460 * so that the hardware will be re-enabled
1461 * when we resume. See below.
1466 case CPU_PM_ENTER_FAILED:
1468 if (__this_cpu_read(kvm_arm_hardware_enabled))
1469 /* The hardware was enabled before suspend. */
1479 static struct notifier_block hyp_init_cpu_pm_nb = {
1480 .notifier_call = hyp_init_cpu_pm_notifier,
1483 static void __init hyp_cpu_pm_init(void)
1485 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1487 static void __init hyp_cpu_pm_exit(void)
1489 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1492 static inline void hyp_cpu_pm_init(void)
1495 static inline void hyp_cpu_pm_exit(void)
1500 static int init_common_resources(void)
1502 return kvm_set_ipa_limit();
1505 static int init_subsystems(void)
1510 * Enable hardware so that subsystem initialisation can access EL2.
1512 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1515 * Register CPU lower-power notifier
1520 * Init HYP view of VGIC
1522 err = kvm_vgic_hyp_init();
1525 vgic_present = true;
1529 vgic_present = false;
1537 * Init HYP architected timer support
1539 err = kvm_timer_hyp_init(vgic_present);
1544 kvm_coproc_table_init();
1547 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1552 static void teardown_hyp_mode(void)
1557 for_each_possible_cpu(cpu) {
1558 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1559 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order());
1564 * Inits Hyp-mode on all online CPUs
1566 static int init_hyp_mode(void)
1572 * Allocate Hyp PGD and setup Hyp identity mapping
1574 err = kvm_mmu_init();
1579 * Allocate stack pages for Hypervisor-mode
1581 for_each_possible_cpu(cpu) {
1582 unsigned long stack_page;
1584 stack_page = __get_free_page(GFP_KERNEL);
1590 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1594 * Allocate and initialize pages for Hypervisor-mode percpu regions.
1596 for_each_possible_cpu(cpu) {
1600 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
1606 page_addr = page_address(page);
1607 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
1608 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr;
1612 * Map the Hyp-code called directly from the host
1614 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1615 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1617 kvm_err("Cannot map world-switch code\n");
1621 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1622 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1624 kvm_err("Cannot map rodata section\n");
1628 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1629 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1631 kvm_err("Cannot map bss section\n");
1635 err = kvm_map_vectors();
1637 kvm_err("Cannot map vectors\n");
1642 * Map the Hyp stack pages
1644 for_each_possible_cpu(cpu) {
1645 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1646 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1650 kvm_err("Cannot map hyp stack\n");
1656 * Map Hyp percpu pages
1658 for_each_possible_cpu(cpu) {
1659 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu];
1660 char *percpu_end = percpu_begin + nvhe_percpu_size();
1662 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
1665 kvm_err("Cannot map hyp percpu region\n");
1673 teardown_hyp_mode();
1674 kvm_err("error initializing Hyp mode: %d\n", err);
1678 static void check_kvm_target_cpu(void *ret)
1680 *(int *)ret = kvm_target_cpu();
1683 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1685 struct kvm_vcpu *vcpu;
1688 mpidr &= MPIDR_HWID_BITMASK;
1689 kvm_for_each_vcpu(i, vcpu, kvm) {
1690 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1696 bool kvm_arch_has_irq_bypass(void)
1701 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1702 struct irq_bypass_producer *prod)
1704 struct kvm_kernel_irqfd *irqfd =
1705 container_of(cons, struct kvm_kernel_irqfd, consumer);
1707 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1710 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1711 struct irq_bypass_producer *prod)
1713 struct kvm_kernel_irqfd *irqfd =
1714 container_of(cons, struct kvm_kernel_irqfd, consumer);
1716 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1720 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1722 struct kvm_kernel_irqfd *irqfd =
1723 container_of(cons, struct kvm_kernel_irqfd, consumer);
1725 kvm_arm_halt_guest(irqfd->kvm);
1728 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1730 struct kvm_kernel_irqfd *irqfd =
1731 container_of(cons, struct kvm_kernel_irqfd, consumer);
1733 kvm_arm_resume_guest(irqfd->kvm);
1737 * Initialize Hyp-mode and memory mappings on all CPUs.
1739 int kvm_arch_init(void *opaque)
1745 if (!is_hyp_mode_available()) {
1746 kvm_info("HYP mode not available\n");
1750 in_hyp_mode = is_kernel_in_hyp_mode();
1752 if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1753 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1757 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
1758 cpus_have_final_cap(ARM64_WORKAROUND_1508412))
1759 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
1760 "Only trusted guests should be used on this system.\n");
1762 for_each_online_cpu(cpu) {
1763 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1765 kvm_err("Error, CPU %d not supported!\n", cpu);
1770 err = init_common_resources();
1774 err = kvm_arm_init_sve();
1779 err = init_hyp_mode();
1784 err = init_subsystems();
1789 kvm_info("VHE mode initialized successfully\n");
1791 kvm_info("Hyp mode initialized successfully\n");
1798 teardown_hyp_mode();
1803 /* NOP: Compiling as a module not supported */
1804 void kvm_arch_exit(void)
1806 kvm_perf_teardown();
1809 static int arm_init(void)
1811 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1815 module_init(arm_init);