2 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
3 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2, as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
19 #include <linux/bug.h>
20 #include <linux/cpu_pm.h>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.h>
24 #include <linux/list.h>
25 #include <linux/module.h>
26 #include <linux/vmalloc.h>
28 #include <linux/mman.h>
29 #include <linux/sched.h>
30 #include <linux/kvm.h>
31 #include <linux/kvm_irqfd.h>
32 #include <linux/irqbypass.h>
33 #include <linux/sched/stat.h>
34 #include <trace/events/kvm.h>
35 #include <kvm/arm_pmu.h>
36 #include <kvm/arm_psci.h>
38 #define CREATE_TRACE_POINTS
41 #include <linux/uaccess.h>
42 #include <asm/ptrace.h>
44 #include <asm/tlbflush.h>
45 #include <asm/cacheflush.h>
46 #include <asm/cpufeature.h>
48 #include <asm/kvm_arm.h>
49 #include <asm/kvm_asm.h>
50 #include <asm/kvm_mmu.h>
51 #include <asm/kvm_emulate.h>
52 #include <asm/kvm_coproc.h>
53 #include <asm/sections.h>
56 __asm__(".arch_extension virt");
59 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
60 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
62 /* Per-CPU variable containing the currently running vcpu. */
63 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
65 /* The VMID used in the VTTBR */
66 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
67 static u32 kvm_next_vmid;
68 static unsigned int kvm_vmid_bits __read_mostly;
69 static DEFINE_RWLOCK(kvm_vmid_lock);
71 static bool vgic_present;
73 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
75 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
77 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
80 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
83 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
84 * Must be called from non-preemptible context
86 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
88 return __this_cpu_read(kvm_arm_running_vcpu);
92 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
94 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
96 return &kvm_arm_running_vcpu;
99 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
101 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
104 int kvm_arch_hardware_setup(void)
109 void kvm_arch_check_processor_compat(void *rtn)
116 * kvm_arch_init_vm - initializes a VM data structure
117 * @kvm: pointer to the KVM struct
119 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
126 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
127 if (!kvm->arch.last_vcpu_ran)
130 for_each_possible_cpu(cpu)
131 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
133 ret = kvm_alloc_stage2_pgd(kvm);
137 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
139 goto out_free_stage2_pgd;
141 kvm_vgic_early_init(kvm);
143 /* Mark the initial VMID generation invalid */
144 kvm->arch.vmid_gen = 0;
146 /* The maximum number of VCPUs is limited by the host's GIC model */
147 kvm->arch.max_vcpus = vgic_present ?
148 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
152 kvm_free_stage2_pgd(kvm);
154 free_percpu(kvm->arch.last_vcpu_ran);
155 kvm->arch.last_vcpu_ran = NULL;
159 bool kvm_arch_has_vcpu_debugfs(void)
164 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
169 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
171 return VM_FAULT_SIGBUS;
176 * kvm_arch_destroy_vm - destroy the VM data structure
177 * @kvm: pointer to the KVM struct
179 void kvm_arch_destroy_vm(struct kvm *kvm)
183 kvm_vgic_destroy(kvm);
185 free_percpu(kvm->arch.last_vcpu_ran);
186 kvm->arch.last_vcpu_ran = NULL;
188 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
190 kvm_arch_vcpu_free(kvm->vcpus[i]);
191 kvm->vcpus[i] = NULL;
194 atomic_set(&kvm->online_vcpus, 0);
197 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
201 case KVM_CAP_IRQCHIP:
204 case KVM_CAP_IOEVENTFD:
205 case KVM_CAP_DEVICE_CTRL:
206 case KVM_CAP_USER_MEMORY:
207 case KVM_CAP_SYNC_MMU:
208 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
209 case KVM_CAP_ONE_REG:
210 case KVM_CAP_ARM_PSCI:
211 case KVM_CAP_ARM_PSCI_0_2:
212 case KVM_CAP_READONLY_MEM:
213 case KVM_CAP_MP_STATE:
214 case KVM_CAP_IMMEDIATE_EXIT:
217 case KVM_CAP_ARM_SET_DEVICE_ADDR:
220 case KVM_CAP_NR_VCPUS:
221 r = num_online_cpus();
223 case KVM_CAP_MAX_VCPUS:
226 case KVM_CAP_NR_MEMSLOTS:
227 r = KVM_USER_MEM_SLOTS;
229 case KVM_CAP_MSI_DEVID:
233 r = kvm->arch.vgic.msis_require_devid;
235 case KVM_CAP_ARM_USER_IRQ:
237 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
238 * (bump this number if adding more devices)
243 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
249 long kvm_arch_dev_ioctl(struct file *filp,
250 unsigned int ioctl, unsigned long arg)
255 struct kvm *kvm_arch_alloc_vm(void)
258 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
260 return vzalloc(sizeof(struct kvm));
263 void kvm_arch_free_vm(struct kvm *kvm)
271 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
274 struct kvm_vcpu *vcpu;
276 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
281 if (id >= kvm->arch.max_vcpus) {
286 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
292 err = kvm_vcpu_init(vcpu, kvm, id);
296 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
302 kvm_vcpu_uninit(vcpu);
304 kmem_cache_free(kvm_vcpu_cache, vcpu);
309 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
313 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
315 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
316 static_branch_dec(&userspace_irqchip_in_use);
318 kvm_mmu_free_memory_caches(vcpu);
319 kvm_timer_vcpu_terminate(vcpu);
320 kvm_pmu_vcpu_destroy(vcpu);
321 kvm_vcpu_uninit(vcpu);
322 kmem_cache_free(kvm_vcpu_cache, vcpu);
325 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
327 kvm_arch_vcpu_free(vcpu);
330 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
332 return kvm_timer_is_pending(vcpu);
335 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
337 kvm_timer_schedule(vcpu);
338 kvm_vgic_v4_enable_doorbell(vcpu);
341 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
343 kvm_timer_unschedule(vcpu);
344 kvm_vgic_v4_disable_doorbell(vcpu);
347 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
349 /* Force users to call KVM_ARM_VCPU_INIT */
350 vcpu->arch.target = -1;
351 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
353 /* Set up the timer */
354 kvm_timer_vcpu_init(vcpu);
356 kvm_arm_reset_debug_ptr(vcpu);
358 return kvm_vgic_vcpu_init(vcpu);
361 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
365 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
368 * We might get preempted before the vCPU actually runs, but
369 * over-invalidation doesn't affect correctness.
371 if (*last_ran != vcpu->vcpu_id) {
372 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
373 *last_ran = vcpu->vcpu_id;
377 vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
379 kvm_arm_set_running_vcpu(vcpu);
381 kvm_timer_vcpu_load(vcpu);
382 kvm_vcpu_load_sysregs(vcpu);
383 kvm_arch_vcpu_load_fp(vcpu);
385 if (single_task_running())
386 vcpu_clear_wfe_traps(vcpu);
388 vcpu_set_wfe_traps(vcpu);
391 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
393 kvm_arch_vcpu_put_fp(vcpu);
394 kvm_vcpu_put_sysregs(vcpu);
395 kvm_timer_vcpu_put(vcpu);
400 kvm_arm_set_running_vcpu(NULL);
403 static void vcpu_power_off(struct kvm_vcpu *vcpu)
405 vcpu->arch.power_off = true;
406 kvm_make_request(KVM_REQ_SLEEP, vcpu);
410 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
411 struct kvm_mp_state *mp_state)
413 if (vcpu->arch.power_off)
414 mp_state->mp_state = KVM_MP_STATE_STOPPED;
416 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
421 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
422 struct kvm_mp_state *mp_state)
426 switch (mp_state->mp_state) {
427 case KVM_MP_STATE_RUNNABLE:
428 vcpu->arch.power_off = false;
430 case KVM_MP_STATE_STOPPED:
431 vcpu_power_off(vcpu);
441 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
442 * @v: The VCPU pointer
444 * If the guest CPU is not waiting for interrupts or an interrupt line is
445 * asserted, the CPU is by definition runnable.
447 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
449 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
450 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
451 && !v->arch.power_off && !v->arch.pause);
454 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
456 return vcpu_mode_priv(vcpu);
459 /* Just ensure a guest exit from a particular CPU */
460 static void exit_vm_noop(void *info)
464 void force_vm_exit(const cpumask_t *mask)
467 smp_call_function_many(mask, exit_vm_noop, NULL, true);
472 * need_new_vmid_gen - check that the VMID is still valid
473 * @kvm: The VM's VMID to check
475 * return true if there is a new generation of VMIDs being used
477 * The hardware supports only 256 values with the value zero reserved for the
478 * host, so we check if an assigned value belongs to a previous generation,
479 * which which requires us to assign a new value. If we're the first to use a
480 * VMID for the new generation, we must flush necessary caches and TLBs on all
483 static bool need_new_vmid_gen(struct kvm *kvm)
485 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
489 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
490 * @kvm The guest that we are about to run
492 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
493 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
496 static void update_vttbr(struct kvm *kvm)
498 phys_addr_t pgd_phys;
502 read_lock(&kvm_vmid_lock);
503 new_gen = need_new_vmid_gen(kvm);
504 read_unlock(&kvm_vmid_lock);
509 write_lock(&kvm_vmid_lock);
512 * We need to re-check the vmid_gen here to ensure that if another vcpu
513 * already allocated a valid vmid for this vm, then this vcpu should
516 if (!need_new_vmid_gen(kvm)) {
517 write_unlock(&kvm_vmid_lock);
521 /* First user of a new VMID generation? */
522 if (unlikely(kvm_next_vmid == 0)) {
523 atomic64_inc(&kvm_vmid_gen);
527 * On SMP we know no other CPUs can use this CPU's or each
528 * other's VMID after force_vm_exit returns since the
529 * kvm_vmid_lock blocks them from reentry to the guest.
531 force_vm_exit(cpu_all_mask);
533 * Now broadcast TLB + ICACHE invalidation over the inner
534 * shareable domain to make sure all data structures are
537 kvm_call_hyp(__kvm_flush_vm_context);
540 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
541 kvm->arch.vmid = kvm_next_vmid;
543 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
545 /* update vttbr to be used with the new vmid */
546 pgd_phys = virt_to_phys(kvm->arch.pgd);
547 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
548 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
549 kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid;
551 write_unlock(&kvm_vmid_lock);
554 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
556 struct kvm *kvm = vcpu->kvm;
559 if (likely(vcpu->arch.has_run_once))
562 vcpu->arch.has_run_once = true;
564 if (likely(irqchip_in_kernel(kvm))) {
566 * Map the VGIC hardware resources before running a vcpu the
567 * first time on this VM.
569 if (unlikely(!vgic_ready(kvm))) {
570 ret = kvm_vgic_map_resources(kvm);
576 * Tell the rest of the code that there are userspace irqchip
579 static_branch_inc(&userspace_irqchip_in_use);
582 ret = kvm_timer_enable(vcpu);
586 ret = kvm_arm_pmu_v3_enable(vcpu);
591 bool kvm_arch_intc_initialized(struct kvm *kvm)
593 return vgic_initialized(kvm);
596 void kvm_arm_halt_guest(struct kvm *kvm)
599 struct kvm_vcpu *vcpu;
601 kvm_for_each_vcpu(i, vcpu, kvm)
602 vcpu->arch.pause = true;
603 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
606 void kvm_arm_resume_guest(struct kvm *kvm)
609 struct kvm_vcpu *vcpu;
611 kvm_for_each_vcpu(i, vcpu, kvm) {
612 vcpu->arch.pause = false;
613 swake_up_one(kvm_arch_vcpu_wq(vcpu));
617 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
619 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
621 swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
622 (!vcpu->arch.pause)));
624 if (vcpu->arch.power_off || vcpu->arch.pause) {
625 /* Awaken to handle a signal, request we sleep again later. */
626 kvm_make_request(KVM_REQ_SLEEP, vcpu);
630 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
632 return vcpu->arch.target >= 0;
635 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
637 if (kvm_request_pending(vcpu)) {
638 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
639 vcpu_req_sleep(vcpu);
642 * Clear IRQ_PENDING requests that were made to guarantee
643 * that a VCPU sees new virtual interrupts.
645 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
650 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
651 * @vcpu: The VCPU pointer
652 * @run: The kvm_run structure pointer used for userspace state exchange
654 * This function is called through the VCPU_RUN ioctl called from user space. It
655 * will execute VM code in a loop until the time slice for the process is used
656 * or some emulation is needed from user space in which case the function will
657 * return with return value 0 and with the kvm_run structure filled in with the
658 * required data for the requested emulation.
660 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
664 if (unlikely(!kvm_vcpu_initialized(vcpu)))
667 ret = kvm_vcpu_first_run_init(vcpu);
671 if (run->exit_reason == KVM_EXIT_MMIO) {
672 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
675 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
679 if (run->immediate_exit)
684 kvm_sigset_activate(vcpu);
687 run->exit_reason = KVM_EXIT_UNKNOWN;
690 * Check conditions before entering the guest
694 update_vttbr(vcpu->kvm);
696 check_vcpu_requests(vcpu);
699 * Preparing the interrupts to be injected also
700 * involves poking the GIC, which must be done in a
701 * non-preemptible context.
705 kvm_pmu_flush_hwstate(vcpu);
709 kvm_vgic_flush_hwstate(vcpu);
712 * Exit if we have a signal pending so that we can deliver the
713 * signal to user space.
715 if (signal_pending(current)) {
717 run->exit_reason = KVM_EXIT_INTR;
721 * If we're using a userspace irqchip, then check if we need
722 * to tell a userspace irqchip about timer or PMU level
723 * changes and if so, exit to userspace (the actual level
724 * state gets updated in kvm_timer_update_run and
725 * kvm_pmu_update_run below).
727 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
728 if (kvm_timer_should_notify_user(vcpu) ||
729 kvm_pmu_should_notify_user(vcpu)) {
731 run->exit_reason = KVM_EXIT_INTR;
736 * Ensure we set mode to IN_GUEST_MODE after we disable
737 * interrupts and before the final VCPU requests check.
738 * See the comment in kvm_vcpu_exiting_guest_mode() and
739 * Documentation/virtual/kvm/vcpu-requests.rst
741 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
743 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
744 kvm_request_pending(vcpu)) {
745 vcpu->mode = OUTSIDE_GUEST_MODE;
746 isb(); /* Ensure work in x_flush_hwstate is committed */
747 kvm_pmu_sync_hwstate(vcpu);
748 if (static_branch_unlikely(&userspace_irqchip_in_use))
749 kvm_timer_sync_hwstate(vcpu);
750 kvm_vgic_sync_hwstate(vcpu);
756 kvm_arm_setup_debug(vcpu);
758 /**************************************************************
761 trace_kvm_entry(*vcpu_pc(vcpu));
762 guest_enter_irqoff();
765 kvm_arm_vhe_guest_enter();
766 ret = kvm_vcpu_run_vhe(vcpu);
767 kvm_arm_vhe_guest_exit();
769 ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
772 vcpu->mode = OUTSIDE_GUEST_MODE;
776 *************************************************************/
778 kvm_arm_clear_debug(vcpu);
781 * We must sync the PMU state before the vgic state so
782 * that the vgic can properly sample the updated state of the
785 kvm_pmu_sync_hwstate(vcpu);
788 * Sync the vgic state before syncing the timer state because
789 * the timer code needs to know if the virtual timer
790 * interrupts are active.
792 kvm_vgic_sync_hwstate(vcpu);
795 * Sync the timer hardware state before enabling interrupts as
796 * we don't want vtimer interrupts to race with syncing the
797 * timer virtual interrupt state.
799 if (static_branch_unlikely(&userspace_irqchip_in_use))
800 kvm_timer_sync_hwstate(vcpu);
802 kvm_arch_vcpu_ctxsync_fp(vcpu);
805 * We may have taken a host interrupt in HYP mode (ie
806 * while executing the guest). This interrupt is still
807 * pending, as we haven't serviced it yet!
809 * We're now back in SVC mode, with interrupts
810 * disabled. Enabling the interrupts now will have
811 * the effect of taking the interrupt again, in SVC
817 * We do local_irq_enable() before calling guest_exit() so
818 * that if a timer interrupt hits while running the guest we
819 * account that tick as being spent in the guest. We enable
820 * preemption after calling guest_exit() so that if we get
821 * preempted we make sure ticks after that is not counted as
825 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
827 /* Exit types that need handling before we can be preempted */
828 handle_exit_early(vcpu, run, ret);
832 ret = handle_exit(vcpu, run, ret);
835 /* Tell userspace about in-kernel device output levels */
836 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
837 kvm_timer_update_run(vcpu);
838 kvm_pmu_update_run(vcpu);
841 kvm_sigset_deactivate(vcpu);
847 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
853 if (number == KVM_ARM_IRQ_CPU_IRQ)
854 bit_index = __ffs(HCR_VI);
855 else /* KVM_ARM_IRQ_CPU_FIQ */
856 bit_index = __ffs(HCR_VF);
858 hcr = vcpu_hcr(vcpu);
860 set = test_and_set_bit(bit_index, hcr);
862 set = test_and_clear_bit(bit_index, hcr);
865 * If we didn't change anything, no need to wake up or kick other CPUs
871 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
872 * trigger a world-switch round on the running physical CPU to set the
873 * virtual IRQ/FIQ fields in the HCR appropriately.
875 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
881 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
884 u32 irq = irq_level->irq;
885 unsigned int irq_type, vcpu_idx, irq_num;
886 int nrcpus = atomic_read(&kvm->online_vcpus);
887 struct kvm_vcpu *vcpu = NULL;
888 bool level = irq_level->level;
890 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
891 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
892 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
894 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
897 case KVM_ARM_IRQ_TYPE_CPU:
898 if (irqchip_in_kernel(kvm))
901 if (vcpu_idx >= nrcpus)
904 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
908 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
911 return vcpu_interrupt_line(vcpu, irq_num, level);
912 case KVM_ARM_IRQ_TYPE_PPI:
913 if (!irqchip_in_kernel(kvm))
916 if (vcpu_idx >= nrcpus)
919 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
923 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
926 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
927 case KVM_ARM_IRQ_TYPE_SPI:
928 if (!irqchip_in_kernel(kvm))
931 if (irq_num < VGIC_NR_PRIVATE_IRQS)
934 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
940 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
941 const struct kvm_vcpu_init *init)
944 int phys_target = kvm_target_cpu();
946 if (init->target != phys_target)
950 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
951 * use the same target.
953 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
956 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
957 for (i = 0; i < sizeof(init->features) * 8; i++) {
958 bool set = (init->features[i / 32] & (1 << (i % 32)));
960 if (set && i >= KVM_VCPU_MAX_FEATURES)
964 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
965 * use the same feature set.
967 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
968 test_bit(i, vcpu->arch.features) != set)
972 set_bit(i, vcpu->arch.features);
975 vcpu->arch.target = phys_target;
977 /* Now we know what it is, we can reset it. */
978 return kvm_reset_vcpu(vcpu);
982 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
983 struct kvm_vcpu_init *init)
987 ret = kvm_vcpu_set_target(vcpu, init);
992 * Ensure a rebooted VM will fault in RAM pages and detect if the
993 * guest MMU is turned off and flush the caches as needed.
995 if (vcpu->arch.has_run_once)
996 stage2_unmap_vm(vcpu->kvm);
998 vcpu_reset_hcr(vcpu);
1001 * Handle the "start in power-off" case.
1003 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1004 vcpu_power_off(vcpu);
1006 vcpu->arch.power_off = false;
1011 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1012 struct kvm_device_attr *attr)
1016 switch (attr->group) {
1018 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1025 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1026 struct kvm_device_attr *attr)
1030 switch (attr->group) {
1032 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1039 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1040 struct kvm_device_attr *attr)
1044 switch (attr->group) {
1046 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1053 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1054 struct kvm_vcpu_events *events)
1056 memset(events, 0, sizeof(*events));
1058 return __kvm_arm_vcpu_get_events(vcpu, events);
1061 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1062 struct kvm_vcpu_events *events)
1066 /* check whether the reserved field is zero */
1067 for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1068 if (events->reserved[i])
1071 /* check whether the pad field is zero */
1072 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1073 if (events->exception.pad[i])
1076 return __kvm_arm_vcpu_set_events(vcpu, events);
1079 long kvm_arch_vcpu_ioctl(struct file *filp,
1080 unsigned int ioctl, unsigned long arg)
1082 struct kvm_vcpu *vcpu = filp->private_data;
1083 void __user *argp = (void __user *)arg;
1084 struct kvm_device_attr attr;
1088 case KVM_ARM_VCPU_INIT: {
1089 struct kvm_vcpu_init init;
1092 if (copy_from_user(&init, argp, sizeof(init)))
1095 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1098 case KVM_SET_ONE_REG:
1099 case KVM_GET_ONE_REG: {
1100 struct kvm_one_reg reg;
1103 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1107 if (copy_from_user(®, argp, sizeof(reg)))
1110 if (ioctl == KVM_SET_ONE_REG)
1111 r = kvm_arm_set_reg(vcpu, ®);
1113 r = kvm_arm_get_reg(vcpu, ®);
1116 case KVM_GET_REG_LIST: {
1117 struct kvm_reg_list __user *user_list = argp;
1118 struct kvm_reg_list reg_list;
1122 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1126 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1129 reg_list.n = kvm_arm_num_regs(vcpu);
1130 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1135 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1138 case KVM_SET_DEVICE_ATTR: {
1140 if (copy_from_user(&attr, argp, sizeof(attr)))
1142 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1145 case KVM_GET_DEVICE_ATTR: {
1147 if (copy_from_user(&attr, argp, sizeof(attr)))
1149 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1152 case KVM_HAS_DEVICE_ATTR: {
1154 if (copy_from_user(&attr, argp, sizeof(attr)))
1156 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1159 case KVM_GET_VCPU_EVENTS: {
1160 struct kvm_vcpu_events events;
1162 if (kvm_arm_vcpu_get_events(vcpu, &events))
1165 if (copy_to_user(argp, &events, sizeof(events)))
1170 case KVM_SET_VCPU_EVENTS: {
1171 struct kvm_vcpu_events events;
1173 if (copy_from_user(&events, argp, sizeof(events)))
1176 return kvm_arm_vcpu_set_events(vcpu, &events);
1186 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1187 * @kvm: kvm instance
1188 * @log: slot id and address to which we copy the log
1190 * Steps 1-4 below provide general overview of dirty page logging. See
1191 * kvm_get_dirty_log_protect() function description for additional details.
1193 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1194 * always flush the TLB (step 4) even if previous step failed and the dirty
1195 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1196 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1197 * writes will be marked dirty for next log read.
1199 * 1. Take a snapshot of the bit and clear it if needed.
1200 * 2. Write protect the corresponding page.
1201 * 3. Copy the snapshot to the userspace.
1202 * 4. Flush TLB's if needed.
1204 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1206 bool is_dirty = false;
1209 mutex_lock(&kvm->slots_lock);
1211 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1214 kvm_flush_remote_tlbs(kvm);
1216 mutex_unlock(&kvm->slots_lock);
1220 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1221 struct kvm_arm_device_addr *dev_addr)
1223 unsigned long dev_id, type;
1225 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1226 KVM_ARM_DEVICE_ID_SHIFT;
1227 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1228 KVM_ARM_DEVICE_TYPE_SHIFT;
1231 case KVM_ARM_DEVICE_VGIC_V2:
1234 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1240 long kvm_arch_vm_ioctl(struct file *filp,
1241 unsigned int ioctl, unsigned long arg)
1243 struct kvm *kvm = filp->private_data;
1244 void __user *argp = (void __user *)arg;
1247 case KVM_CREATE_IRQCHIP: {
1251 mutex_lock(&kvm->lock);
1252 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1253 mutex_unlock(&kvm->lock);
1256 case KVM_ARM_SET_DEVICE_ADDR: {
1257 struct kvm_arm_device_addr dev_addr;
1259 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1261 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1263 case KVM_ARM_PREFERRED_TARGET: {
1265 struct kvm_vcpu_init init;
1267 err = kvm_vcpu_preferred_target(&init);
1271 if (copy_to_user(argp, &init, sizeof(init)))
1281 static void cpu_init_hyp_mode(void *dummy)
1283 phys_addr_t pgd_ptr;
1284 unsigned long hyp_stack_ptr;
1285 unsigned long stack_page;
1286 unsigned long vector_ptr;
1288 /* Switch from the HYP stub to our own HYP init vector */
1289 __hyp_set_vectors(kvm_get_idmap_vector());
1291 pgd_ptr = kvm_mmu_get_httbr();
1292 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1293 hyp_stack_ptr = stack_page + PAGE_SIZE;
1294 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1296 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1297 __cpu_init_stage2();
1299 kvm_arm_init_debug();
1302 static void cpu_hyp_reset(void)
1304 if (!is_kernel_in_hyp_mode())
1305 __hyp_reset_vectors();
1308 static void cpu_hyp_reinit(void)
1312 if (is_kernel_in_hyp_mode()) {
1314 * __cpu_init_stage2() is safe to call even if the PM
1315 * event was cancelled before the CPU was reset.
1317 __cpu_init_stage2();
1318 kvm_timer_init_vhe();
1320 cpu_init_hyp_mode(NULL);
1324 kvm_vgic_init_cpu_hardware();
1327 static void _kvm_arch_hardware_enable(void *discard)
1329 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1331 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1335 int kvm_arch_hardware_enable(void)
1337 _kvm_arch_hardware_enable(NULL);
1341 static void _kvm_arch_hardware_disable(void *discard)
1343 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1345 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1349 void kvm_arch_hardware_disable(void)
1351 _kvm_arch_hardware_disable(NULL);
1354 #ifdef CONFIG_CPU_PM
1355 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1360 * kvm_arm_hardware_enabled is left with its old value over
1361 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1366 if (__this_cpu_read(kvm_arm_hardware_enabled))
1368 * don't update kvm_arm_hardware_enabled here
1369 * so that the hardware will be re-enabled
1370 * when we resume. See below.
1375 case CPU_PM_ENTER_FAILED:
1377 if (__this_cpu_read(kvm_arm_hardware_enabled))
1378 /* The hardware was enabled before suspend. */
1388 static struct notifier_block hyp_init_cpu_pm_nb = {
1389 .notifier_call = hyp_init_cpu_pm_notifier,
1392 static void __init hyp_cpu_pm_init(void)
1394 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1396 static void __init hyp_cpu_pm_exit(void)
1398 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1401 static inline void hyp_cpu_pm_init(void)
1404 static inline void hyp_cpu_pm_exit(void)
1409 static int init_common_resources(void)
1411 /* set size of VMID supported by CPU */
1412 kvm_vmid_bits = kvm_get_vmid_bits();
1413 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1418 static int init_subsystems(void)
1423 * Enable hardware so that subsystem initialisation can access EL2.
1425 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1428 * Register CPU lower-power notifier
1433 * Init HYP view of VGIC
1435 err = kvm_vgic_hyp_init();
1438 vgic_present = true;
1442 vgic_present = false;
1450 * Init HYP architected timer support
1452 err = kvm_timer_hyp_init(vgic_present);
1457 kvm_coproc_table_init();
1460 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1465 static void teardown_hyp_mode(void)
1470 for_each_possible_cpu(cpu)
1471 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1476 * Inits Hyp-mode on all online CPUs
1478 static int init_hyp_mode(void)
1484 * Allocate Hyp PGD and setup Hyp identity mapping
1486 err = kvm_mmu_init();
1491 * Allocate stack pages for Hypervisor-mode
1493 for_each_possible_cpu(cpu) {
1494 unsigned long stack_page;
1496 stack_page = __get_free_page(GFP_KERNEL);
1502 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1506 * Map the Hyp-code called directly from the host
1508 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1509 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1511 kvm_err("Cannot map world-switch code\n");
1515 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1516 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1518 kvm_err("Cannot map rodata section\n");
1522 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1523 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1525 kvm_err("Cannot map bss section\n");
1529 err = kvm_map_vectors();
1531 kvm_err("Cannot map vectors\n");
1536 * Map the Hyp stack pages
1538 for_each_possible_cpu(cpu) {
1539 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1540 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1544 kvm_err("Cannot map hyp stack\n");
1549 for_each_possible_cpu(cpu) {
1550 kvm_cpu_context_t *cpu_ctxt;
1552 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1553 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1556 kvm_err("Cannot map host CPU state: %d\n", err);
1561 err = hyp_map_aux_data();
1563 kvm_err("Cannot map host auxilary data: %d\n", err);
1568 teardown_hyp_mode();
1569 kvm_err("error initializing Hyp mode: %d\n", err);
1573 static void check_kvm_target_cpu(void *ret)
1575 *(int *)ret = kvm_target_cpu();
1578 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1580 struct kvm_vcpu *vcpu;
1583 mpidr &= MPIDR_HWID_BITMASK;
1584 kvm_for_each_vcpu(i, vcpu, kvm) {
1585 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1591 bool kvm_arch_has_irq_bypass(void)
1596 int kvm_arch_irq_bypass_add_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 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1605 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1606 struct irq_bypass_producer *prod)
1608 struct kvm_kernel_irqfd *irqfd =
1609 container_of(cons, struct kvm_kernel_irqfd, consumer);
1611 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1615 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1617 struct kvm_kernel_irqfd *irqfd =
1618 container_of(cons, struct kvm_kernel_irqfd, consumer);
1620 kvm_arm_halt_guest(irqfd->kvm);
1623 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1625 struct kvm_kernel_irqfd *irqfd =
1626 container_of(cons, struct kvm_kernel_irqfd, consumer);
1628 kvm_arm_resume_guest(irqfd->kvm);
1632 * Initialize Hyp-mode and memory mappings on all CPUs.
1634 int kvm_arch_init(void *opaque)
1640 if (!is_hyp_mode_available()) {
1641 kvm_info("HYP mode not available\n");
1645 if (!kvm_arch_check_sve_has_vhe()) {
1646 kvm_pr_unimpl("SVE system without VHE unsupported. Broken cpu?");
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 in_hyp_mode = is_kernel_in_hyp_mode();
1665 err = init_hyp_mode();
1670 err = init_subsystems();
1675 kvm_info("VHE mode initialized successfully\n");
1677 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);