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/cpu_pm.h>
20 #include <linux/errno.h>
21 #include <linux/err.h>
22 #include <linux/kvm_host.h>
23 #include <linux/list.h>
24 #include <linux/module.h>
25 #include <linux/vmalloc.h>
27 #include <linux/mman.h>
28 #include <linux/sched.h>
29 #include <linux/kvm.h>
30 #include <linux/kvm_irqfd.h>
31 #include <linux/irqbypass.h>
32 #include <trace/events/kvm.h>
33 #include <kvm/arm_pmu.h>
35 #define CREATE_TRACE_POINTS
38 #include <linux/uaccess.h>
39 #include <asm/ptrace.h>
41 #include <asm/tlbflush.h>
42 #include <asm/cacheflush.h>
44 #include <asm/kvm_arm.h>
45 #include <asm/kvm_asm.h>
46 #include <asm/kvm_mmu.h>
47 #include <asm/kvm_emulate.h>
48 #include <asm/kvm_coproc.h>
49 #include <asm/kvm_psci.h>
50 #include <asm/sections.h>
53 __asm__(".arch_extension virt");
56 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
57 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
59 /* Per-CPU variable containing the currently running vcpu. */
60 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
62 /* The VMID used in the VTTBR */
63 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
64 static u32 kvm_next_vmid;
65 static unsigned int kvm_vmid_bits __read_mostly;
66 static DEFINE_SPINLOCK(kvm_vmid_lock);
68 static bool vgic_present;
70 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
72 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
74 BUG_ON(preemptible());
75 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
79 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
80 * Must be called from non-preemptible context
82 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
84 BUG_ON(preemptible());
85 return __this_cpu_read(kvm_arm_running_vcpu);
89 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
91 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
93 return &kvm_arm_running_vcpu;
96 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
98 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
101 int kvm_arch_hardware_setup(void)
106 void kvm_arch_check_processor_compat(void *rtn)
113 * kvm_arch_init_vm - initializes a VM data structure
114 * @kvm: pointer to the KVM struct
116 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
123 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
124 if (!kvm->arch.last_vcpu_ran)
127 for_each_possible_cpu(cpu)
128 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
130 ret = kvm_alloc_stage2_pgd(kvm);
134 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
136 goto out_free_stage2_pgd;
138 kvm_vgic_early_init(kvm);
140 /* Mark the initial VMID generation invalid */
141 kvm->arch.vmid_gen = 0;
143 /* The maximum number of VCPUs is limited by the host's GIC model */
144 kvm->arch.max_vcpus = vgic_present ?
145 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
149 kvm_free_stage2_pgd(kvm);
151 free_percpu(kvm->arch.last_vcpu_ran);
152 kvm->arch.last_vcpu_ran = NULL;
156 bool kvm_arch_has_vcpu_debugfs(void)
161 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
166 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
168 return VM_FAULT_SIGBUS;
173 * kvm_arch_destroy_vm - destroy the VM data structure
174 * @kvm: pointer to the KVM struct
176 void kvm_arch_destroy_vm(struct kvm *kvm)
180 kvm_vgic_destroy(kvm);
182 free_percpu(kvm->arch.last_vcpu_ran);
183 kvm->arch.last_vcpu_ran = NULL;
185 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
187 kvm_arch_vcpu_free(kvm->vcpus[i]);
188 kvm->vcpus[i] = NULL;
191 atomic_set(&kvm->online_vcpus, 0);
194 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
198 case KVM_CAP_IRQCHIP:
201 case KVM_CAP_IOEVENTFD:
202 case KVM_CAP_DEVICE_CTRL:
203 case KVM_CAP_USER_MEMORY:
204 case KVM_CAP_SYNC_MMU:
205 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
206 case KVM_CAP_ONE_REG:
207 case KVM_CAP_ARM_PSCI:
208 case KVM_CAP_ARM_PSCI_0_2:
209 case KVM_CAP_READONLY_MEM:
210 case KVM_CAP_MP_STATE:
211 case KVM_CAP_IMMEDIATE_EXIT:
214 case KVM_CAP_ARM_SET_DEVICE_ADDR:
217 case KVM_CAP_NR_VCPUS:
218 r = num_online_cpus();
220 case KVM_CAP_MAX_VCPUS:
223 case KVM_CAP_NR_MEMSLOTS:
224 r = KVM_USER_MEM_SLOTS;
226 case KVM_CAP_MSI_DEVID:
230 r = kvm->arch.vgic.msis_require_devid;
232 case KVM_CAP_ARM_USER_IRQ:
234 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
235 * (bump this number if adding more devices)
240 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
246 long kvm_arch_dev_ioctl(struct file *filp,
247 unsigned int ioctl, unsigned long arg)
253 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
256 struct kvm_vcpu *vcpu;
258 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
263 if (id >= kvm->arch.max_vcpus) {
268 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
274 err = kvm_vcpu_init(vcpu, kvm, id);
278 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
284 kvm_vcpu_uninit(vcpu);
286 kmem_cache_free(kvm_vcpu_cache, vcpu);
291 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
293 kvm_vgic_vcpu_early_init(vcpu);
296 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
298 kvm_mmu_free_memory_caches(vcpu);
299 kvm_timer_vcpu_terminate(vcpu);
300 kvm_pmu_vcpu_destroy(vcpu);
301 kvm_vcpu_uninit(vcpu);
302 kmem_cache_free(kvm_vcpu_cache, vcpu);
305 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
307 kvm_arch_vcpu_free(vcpu);
310 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
312 return kvm_timer_is_pending(vcpu);
315 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
317 kvm_timer_schedule(vcpu);
318 kvm_vgic_v4_enable_doorbell(vcpu);
321 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
323 kvm_timer_unschedule(vcpu);
324 kvm_vgic_v4_disable_doorbell(vcpu);
327 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
329 /* Force users to call KVM_ARM_VCPU_INIT */
330 vcpu->arch.target = -1;
331 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
333 /* Set up the timer */
334 kvm_timer_vcpu_init(vcpu);
336 kvm_arm_reset_debug_ptr(vcpu);
338 return kvm_vgic_vcpu_init(vcpu);
341 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
345 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
348 * We might get preempted before the vCPU actually runs, but
349 * over-invalidation doesn't affect correctness.
351 if (*last_ran != vcpu->vcpu_id) {
352 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
353 *last_ran = vcpu->vcpu_id;
357 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
359 kvm_arm_set_running_vcpu(vcpu);
361 kvm_timer_vcpu_load(vcpu);
364 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
366 kvm_timer_vcpu_put(vcpu);
371 kvm_arm_set_running_vcpu(NULL);
374 static void vcpu_power_off(struct kvm_vcpu *vcpu)
376 vcpu->arch.power_off = true;
377 kvm_make_request(KVM_REQ_SLEEP, vcpu);
381 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
382 struct kvm_mp_state *mp_state)
386 if (vcpu->arch.power_off)
387 mp_state->mp_state = KVM_MP_STATE_STOPPED;
389 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
395 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
396 struct kvm_mp_state *mp_state)
402 switch (mp_state->mp_state) {
403 case KVM_MP_STATE_RUNNABLE:
404 vcpu->arch.power_off = false;
406 case KVM_MP_STATE_STOPPED:
407 vcpu_power_off(vcpu);
418 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
419 * @v: The VCPU pointer
421 * If the guest CPU is not waiting for interrupts or an interrupt line is
422 * asserted, the CPU is by definition runnable.
424 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
426 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
427 && !v->arch.power_off && !v->arch.pause);
430 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
432 return vcpu_mode_priv(vcpu);
435 /* Just ensure a guest exit from a particular CPU */
436 static void exit_vm_noop(void *info)
440 void force_vm_exit(const cpumask_t *mask)
443 smp_call_function_many(mask, exit_vm_noop, NULL, true);
448 * need_new_vmid_gen - check that the VMID is still valid
449 * @kvm: The VM's VMID to check
451 * return true if there is a new generation of VMIDs being used
453 * The hardware supports only 256 values with the value zero reserved for the
454 * host, so we check if an assigned value belongs to a previous generation,
455 * which which requires us to assign a new value. If we're the first to use a
456 * VMID for the new generation, we must flush necessary caches and TLBs on all
459 static bool need_new_vmid_gen(struct kvm *kvm)
461 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
465 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
466 * @kvm The guest that we are about to run
468 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
469 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
472 static void update_vttbr(struct kvm *kvm)
474 phys_addr_t pgd_phys;
477 if (!need_new_vmid_gen(kvm))
480 spin_lock(&kvm_vmid_lock);
483 * We need to re-check the vmid_gen here to ensure that if another vcpu
484 * already allocated a valid vmid for this vm, then this vcpu should
487 if (!need_new_vmid_gen(kvm)) {
488 spin_unlock(&kvm_vmid_lock);
492 /* First user of a new VMID generation? */
493 if (unlikely(kvm_next_vmid == 0)) {
494 atomic64_inc(&kvm_vmid_gen);
498 * On SMP we know no other CPUs can use this CPU's or each
499 * other's VMID after force_vm_exit returns since the
500 * kvm_vmid_lock blocks them from reentry to the guest.
502 force_vm_exit(cpu_all_mask);
504 * Now broadcast TLB + ICACHE invalidation over the inner
505 * shareable domain to make sure all data structures are
508 kvm_call_hyp(__kvm_flush_vm_context);
511 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
512 kvm->arch.vmid = kvm_next_vmid;
514 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
516 /* update vttbr to be used with the new vmid */
517 pgd_phys = virt_to_phys(kvm->arch.pgd);
518 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
519 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
520 kvm->arch.vttbr = pgd_phys | vmid;
522 spin_unlock(&kvm_vmid_lock);
525 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
527 struct kvm *kvm = vcpu->kvm;
530 if (likely(vcpu->arch.has_run_once))
533 vcpu->arch.has_run_once = true;
536 * Map the VGIC hardware resources before running a vcpu the first
539 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
540 ret = kvm_vgic_map_resources(kvm);
545 ret = kvm_timer_enable(vcpu);
549 ret = kvm_arm_pmu_v3_enable(vcpu);
554 bool kvm_arch_intc_initialized(struct kvm *kvm)
556 return vgic_initialized(kvm);
559 void kvm_arm_halt_guest(struct kvm *kvm)
562 struct kvm_vcpu *vcpu;
564 kvm_for_each_vcpu(i, vcpu, kvm)
565 vcpu->arch.pause = true;
566 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
569 void kvm_arm_resume_guest(struct kvm *kvm)
572 struct kvm_vcpu *vcpu;
574 kvm_for_each_vcpu(i, vcpu, kvm) {
575 vcpu->arch.pause = false;
576 swake_up(kvm_arch_vcpu_wq(vcpu));
580 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
582 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
584 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
585 (!vcpu->arch.pause)));
587 if (vcpu->arch.power_off || vcpu->arch.pause) {
588 /* Awaken to handle a signal, request we sleep again later. */
589 kvm_make_request(KVM_REQ_SLEEP, vcpu);
593 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
595 return vcpu->arch.target >= 0;
598 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
600 if (kvm_request_pending(vcpu)) {
601 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
602 vcpu_req_sleep(vcpu);
605 * Clear IRQ_PENDING requests that were made to guarantee
606 * that a VCPU sees new virtual interrupts.
608 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
613 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
614 * @vcpu: The VCPU pointer
615 * @run: The kvm_run structure pointer used for userspace state exchange
617 * This function is called through the VCPU_RUN ioctl called from user space. It
618 * will execute VM code in a loop until the time slice for the process is used
619 * or some emulation is needed from user space in which case the function will
620 * return with return value 0 and with the kvm_run structure filled in with the
621 * required data for the requested emulation.
623 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
627 if (unlikely(!kvm_vcpu_initialized(vcpu)))
632 ret = kvm_vcpu_first_run_init(vcpu);
636 if (run->exit_reason == KVM_EXIT_MMIO) {
637 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
640 if (kvm_arm_handle_step_debug(vcpu, vcpu->run)) {
647 if (run->immediate_exit) {
652 kvm_sigset_activate(vcpu);
655 run->exit_reason = KVM_EXIT_UNKNOWN;
658 * Check conditions before entering the guest
662 update_vttbr(vcpu->kvm);
664 check_vcpu_requests(vcpu);
667 * Preparing the interrupts to be injected also
668 * involves poking the GIC, which must be done in a
669 * non-preemptible context.
673 /* Flush FP/SIMD state that can't survive guest entry/exit */
674 kvm_fpsimd_flush_cpu_state();
676 kvm_pmu_flush_hwstate(vcpu);
680 kvm_vgic_flush_hwstate(vcpu);
683 * If we have a singal pending, or need to notify a userspace
684 * irqchip about timer or PMU level changes, then we exit (and
685 * update the timer level state in kvm_timer_update_run
688 if (signal_pending(current) ||
689 kvm_timer_should_notify_user(vcpu) ||
690 kvm_pmu_should_notify_user(vcpu)) {
692 run->exit_reason = KVM_EXIT_INTR;
696 * Ensure we set mode to IN_GUEST_MODE after we disable
697 * interrupts and before the final VCPU requests check.
698 * See the comment in kvm_vcpu_exiting_guest_mode() and
699 * Documentation/virtual/kvm/vcpu-requests.rst
701 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
703 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
704 kvm_request_pending(vcpu)) {
705 vcpu->mode = OUTSIDE_GUEST_MODE;
706 kvm_pmu_sync_hwstate(vcpu);
707 kvm_timer_sync_hwstate(vcpu);
708 kvm_vgic_sync_hwstate(vcpu);
714 kvm_arm_setup_debug(vcpu);
716 /**************************************************************
719 trace_kvm_entry(*vcpu_pc(vcpu));
720 guest_enter_irqoff();
722 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
724 vcpu->mode = OUTSIDE_GUEST_MODE;
728 *************************************************************/
730 kvm_arm_clear_debug(vcpu);
733 * We must sync the PMU state before the vgic state so
734 * that the vgic can properly sample the updated state of the
737 kvm_pmu_sync_hwstate(vcpu);
740 * Sync the vgic state before syncing the timer state because
741 * the timer code needs to know if the virtual timer
742 * interrupts are active.
744 kvm_vgic_sync_hwstate(vcpu);
747 * Sync the timer hardware state before enabling interrupts as
748 * we don't want vtimer interrupts to race with syncing the
749 * timer virtual interrupt state.
751 kvm_timer_sync_hwstate(vcpu);
754 * We may have taken a host interrupt in HYP mode (ie
755 * while executing the guest). This interrupt is still
756 * pending, as we haven't serviced it yet!
758 * We're now back in SVC mode, with interrupts
759 * disabled. Enabling the interrupts now will have
760 * the effect of taking the interrupt again, in SVC
766 * We do local_irq_enable() before calling guest_exit() so
767 * that if a timer interrupt hits while running the guest we
768 * account that tick as being spent in the guest. We enable
769 * preemption after calling guest_exit() so that if we get
770 * preempted we make sure ticks after that is not counted as
774 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
778 ret = handle_exit(vcpu, run, ret);
781 /* Tell userspace about in-kernel device output levels */
782 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
783 kvm_timer_update_run(vcpu);
784 kvm_pmu_update_run(vcpu);
787 kvm_sigset_deactivate(vcpu);
794 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
800 if (number == KVM_ARM_IRQ_CPU_IRQ)
801 bit_index = __ffs(HCR_VI);
802 else /* KVM_ARM_IRQ_CPU_FIQ */
803 bit_index = __ffs(HCR_VF);
805 ptr = (unsigned long *)&vcpu->arch.irq_lines;
807 set = test_and_set_bit(bit_index, ptr);
809 set = test_and_clear_bit(bit_index, ptr);
812 * If we didn't change anything, no need to wake up or kick other CPUs
818 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
819 * trigger a world-switch round on the running physical CPU to set the
820 * virtual IRQ/FIQ fields in the HCR appropriately.
822 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
828 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
831 u32 irq = irq_level->irq;
832 unsigned int irq_type, vcpu_idx, irq_num;
833 int nrcpus = atomic_read(&kvm->online_vcpus);
834 struct kvm_vcpu *vcpu = NULL;
835 bool level = irq_level->level;
837 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
838 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
839 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
841 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
844 case KVM_ARM_IRQ_TYPE_CPU:
845 if (irqchip_in_kernel(kvm))
848 if (vcpu_idx >= nrcpus)
851 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
855 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
858 return vcpu_interrupt_line(vcpu, irq_num, level);
859 case KVM_ARM_IRQ_TYPE_PPI:
860 if (!irqchip_in_kernel(kvm))
863 if (vcpu_idx >= nrcpus)
866 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
870 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
873 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
874 case KVM_ARM_IRQ_TYPE_SPI:
875 if (!irqchip_in_kernel(kvm))
878 if (irq_num < VGIC_NR_PRIVATE_IRQS)
881 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
887 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
888 const struct kvm_vcpu_init *init)
891 int phys_target = kvm_target_cpu();
893 if (init->target != phys_target)
897 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
898 * use the same target.
900 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
903 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
904 for (i = 0; i < sizeof(init->features) * 8; i++) {
905 bool set = (init->features[i / 32] & (1 << (i % 32)));
907 if (set && i >= KVM_VCPU_MAX_FEATURES)
911 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
912 * use the same feature set.
914 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
915 test_bit(i, vcpu->arch.features) != set)
919 set_bit(i, vcpu->arch.features);
922 vcpu->arch.target = phys_target;
924 /* Now we know what it is, we can reset it. */
925 return kvm_reset_vcpu(vcpu);
929 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
930 struct kvm_vcpu_init *init)
934 ret = kvm_vcpu_set_target(vcpu, init);
939 * Ensure a rebooted VM will fault in RAM pages and detect if the
940 * guest MMU is turned off and flush the caches as needed.
942 if (vcpu->arch.has_run_once)
943 stage2_unmap_vm(vcpu->kvm);
945 vcpu_reset_hcr(vcpu);
948 * Handle the "start in power-off" case.
950 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
951 vcpu_power_off(vcpu);
953 vcpu->arch.power_off = false;
958 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
959 struct kvm_device_attr *attr)
963 switch (attr->group) {
965 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
972 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
973 struct kvm_device_attr *attr)
977 switch (attr->group) {
979 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
986 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
987 struct kvm_device_attr *attr)
991 switch (attr->group) {
993 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1000 long kvm_arch_vcpu_ioctl(struct file *filp,
1001 unsigned int ioctl, unsigned long arg)
1003 struct kvm_vcpu *vcpu = filp->private_data;
1004 void __user *argp = (void __user *)arg;
1005 struct kvm_device_attr attr;
1011 case KVM_ARM_VCPU_INIT: {
1012 struct kvm_vcpu_init init;
1015 if (copy_from_user(&init, argp, sizeof(init)))
1018 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1021 case KVM_SET_ONE_REG:
1022 case KVM_GET_ONE_REG: {
1023 struct kvm_one_reg reg;
1026 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1030 if (copy_from_user(®, argp, sizeof(reg)))
1033 if (ioctl == KVM_SET_ONE_REG)
1034 r = kvm_arm_set_reg(vcpu, ®);
1036 r = kvm_arm_get_reg(vcpu, ®);
1039 case KVM_GET_REG_LIST: {
1040 struct kvm_reg_list __user *user_list = argp;
1041 struct kvm_reg_list reg_list;
1045 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1049 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1052 reg_list.n = kvm_arm_num_regs(vcpu);
1053 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1058 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1061 case KVM_SET_DEVICE_ATTR: {
1063 if (copy_from_user(&attr, argp, sizeof(attr)))
1065 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1068 case KVM_GET_DEVICE_ATTR: {
1070 if (copy_from_user(&attr, argp, sizeof(attr)))
1072 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1075 case KVM_HAS_DEVICE_ATTR: {
1077 if (copy_from_user(&attr, argp, sizeof(attr)))
1079 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1091 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1092 * @kvm: kvm instance
1093 * @log: slot id and address to which we copy the log
1095 * Steps 1-4 below provide general overview of dirty page logging. See
1096 * kvm_get_dirty_log_protect() function description for additional details.
1098 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1099 * always flush the TLB (step 4) even if previous step failed and the dirty
1100 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1101 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1102 * writes will be marked dirty for next log read.
1104 * 1. Take a snapshot of the bit and clear it if needed.
1105 * 2. Write protect the corresponding page.
1106 * 3. Copy the snapshot to the userspace.
1107 * 4. Flush TLB's if needed.
1109 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1111 bool is_dirty = false;
1114 mutex_lock(&kvm->slots_lock);
1116 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1119 kvm_flush_remote_tlbs(kvm);
1121 mutex_unlock(&kvm->slots_lock);
1125 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1126 struct kvm_arm_device_addr *dev_addr)
1128 unsigned long dev_id, type;
1130 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1131 KVM_ARM_DEVICE_ID_SHIFT;
1132 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1133 KVM_ARM_DEVICE_TYPE_SHIFT;
1136 case KVM_ARM_DEVICE_VGIC_V2:
1139 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1145 long kvm_arch_vm_ioctl(struct file *filp,
1146 unsigned int ioctl, unsigned long arg)
1148 struct kvm *kvm = filp->private_data;
1149 void __user *argp = (void __user *)arg;
1152 case KVM_CREATE_IRQCHIP: {
1156 mutex_lock(&kvm->lock);
1157 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1158 mutex_unlock(&kvm->lock);
1161 case KVM_ARM_SET_DEVICE_ADDR: {
1162 struct kvm_arm_device_addr dev_addr;
1164 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1166 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1168 case KVM_ARM_PREFERRED_TARGET: {
1170 struct kvm_vcpu_init init;
1172 err = kvm_vcpu_preferred_target(&init);
1176 if (copy_to_user(argp, &init, sizeof(init)))
1186 static void cpu_init_hyp_mode(void *dummy)
1188 phys_addr_t pgd_ptr;
1189 unsigned long hyp_stack_ptr;
1190 unsigned long stack_page;
1191 unsigned long vector_ptr;
1193 /* Switch from the HYP stub to our own HYP init vector */
1194 __hyp_set_vectors(kvm_get_idmap_vector());
1196 pgd_ptr = kvm_mmu_get_httbr();
1197 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1198 hyp_stack_ptr = stack_page + PAGE_SIZE;
1199 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1201 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1202 __cpu_init_stage2();
1204 kvm_arm_init_debug();
1207 static void cpu_hyp_reset(void)
1209 if (!is_kernel_in_hyp_mode())
1210 __hyp_reset_vectors();
1213 static void cpu_hyp_reinit(void)
1217 if (is_kernel_in_hyp_mode()) {
1219 * __cpu_init_stage2() is safe to call even if the PM
1220 * event was cancelled before the CPU was reset.
1222 __cpu_init_stage2();
1223 kvm_timer_init_vhe();
1225 cpu_init_hyp_mode(NULL);
1229 kvm_vgic_init_cpu_hardware();
1232 static void _kvm_arch_hardware_enable(void *discard)
1234 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1236 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1240 int kvm_arch_hardware_enable(void)
1242 _kvm_arch_hardware_enable(NULL);
1246 static void _kvm_arch_hardware_disable(void *discard)
1248 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1250 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1254 void kvm_arch_hardware_disable(void)
1256 _kvm_arch_hardware_disable(NULL);
1259 #ifdef CONFIG_CPU_PM
1260 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1265 * kvm_arm_hardware_enabled is left with its old value over
1266 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1271 if (__this_cpu_read(kvm_arm_hardware_enabled))
1273 * don't update kvm_arm_hardware_enabled here
1274 * so that the hardware will be re-enabled
1275 * when we resume. See below.
1281 if (__this_cpu_read(kvm_arm_hardware_enabled))
1282 /* The hardware was enabled before suspend. */
1292 static struct notifier_block hyp_init_cpu_pm_nb = {
1293 .notifier_call = hyp_init_cpu_pm_notifier,
1296 static void __init hyp_cpu_pm_init(void)
1298 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1300 static void __init hyp_cpu_pm_exit(void)
1302 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1305 static inline void hyp_cpu_pm_init(void)
1308 static inline void hyp_cpu_pm_exit(void)
1313 static void teardown_common_resources(void)
1315 free_percpu(kvm_host_cpu_state);
1318 static int init_common_resources(void)
1320 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1321 if (!kvm_host_cpu_state) {
1322 kvm_err("Cannot allocate host CPU state\n");
1326 /* set size of VMID supported by CPU */
1327 kvm_vmid_bits = kvm_get_vmid_bits();
1328 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1333 static int init_subsystems(void)
1338 * Enable hardware so that subsystem initialisation can access EL2.
1340 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1343 * Register CPU lower-power notifier
1348 * Init HYP view of VGIC
1350 err = kvm_vgic_hyp_init();
1353 vgic_present = true;
1357 vgic_present = false;
1365 * Init HYP architected timer support
1367 err = kvm_timer_hyp_init();
1372 kvm_coproc_table_init();
1375 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1380 static void teardown_hyp_mode(void)
1385 for_each_possible_cpu(cpu)
1386 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1391 * Inits Hyp-mode on all online CPUs
1393 static int init_hyp_mode(void)
1399 * Allocate Hyp PGD and setup Hyp identity mapping
1401 err = kvm_mmu_init();
1406 * Allocate stack pages for Hypervisor-mode
1408 for_each_possible_cpu(cpu) {
1409 unsigned long stack_page;
1411 stack_page = __get_free_page(GFP_KERNEL);
1417 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1421 * Map the Hyp-code called directly from the host
1423 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1424 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1426 kvm_err("Cannot map world-switch code\n");
1430 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1431 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1433 kvm_err("Cannot map rodata section\n");
1437 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1438 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1440 kvm_err("Cannot map bss section\n");
1445 * Map the Hyp stack pages
1447 for_each_possible_cpu(cpu) {
1448 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1449 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1453 kvm_err("Cannot map hyp stack\n");
1458 for_each_possible_cpu(cpu) {
1459 kvm_cpu_context_t *cpu_ctxt;
1461 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1462 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1465 kvm_err("Cannot map host CPU state: %d\n", err);
1473 teardown_hyp_mode();
1474 kvm_err("error initializing Hyp mode: %d\n", err);
1478 static void check_kvm_target_cpu(void *ret)
1480 *(int *)ret = kvm_target_cpu();
1483 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1485 struct kvm_vcpu *vcpu;
1488 mpidr &= MPIDR_HWID_BITMASK;
1489 kvm_for_each_vcpu(i, vcpu, kvm) {
1490 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1496 bool kvm_arch_has_irq_bypass(void)
1501 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1502 struct irq_bypass_producer *prod)
1504 struct kvm_kernel_irqfd *irqfd =
1505 container_of(cons, struct kvm_kernel_irqfd, consumer);
1507 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1510 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1511 struct irq_bypass_producer *prod)
1513 struct kvm_kernel_irqfd *irqfd =
1514 container_of(cons, struct kvm_kernel_irqfd, consumer);
1516 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1520 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1522 struct kvm_kernel_irqfd *irqfd =
1523 container_of(cons, struct kvm_kernel_irqfd, consumer);
1525 kvm_arm_halt_guest(irqfd->kvm);
1528 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1530 struct kvm_kernel_irqfd *irqfd =
1531 container_of(cons, struct kvm_kernel_irqfd, consumer);
1533 kvm_arm_resume_guest(irqfd->kvm);
1537 * Initialize Hyp-mode and memory mappings on all CPUs.
1539 int kvm_arch_init(void *opaque)
1545 if (!is_hyp_mode_available()) {
1546 kvm_info("HYP mode not available\n");
1550 for_each_online_cpu(cpu) {
1551 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1553 kvm_err("Error, CPU %d not supported!\n", cpu);
1558 err = init_common_resources();
1562 in_hyp_mode = is_kernel_in_hyp_mode();
1565 err = init_hyp_mode();
1570 err = init_subsystems();
1575 kvm_info("VHE mode initialized successfully\n");
1577 kvm_info("Hyp mode initialized successfully\n");
1583 teardown_hyp_mode();
1585 teardown_common_resources();
1589 /* NOP: Compiling as a module not supported */
1590 void kvm_arch_exit(void)
1592 kvm_perf_teardown();
1595 static int arm_init(void)
1597 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1601 module_init(arm_init);