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 <trace/events/kvm.h>
31 #include <kvm/arm_pmu.h>
32 #include <kvm/arm_psci.h>
34 #define CREATE_TRACE_POINTS
37 #include <linux/uaccess.h>
38 #include <asm/ptrace.h>
40 #include <asm/tlbflush.h>
41 #include <asm/cacheflush.h>
43 #include <asm/kvm_arm.h>
44 #include <asm/kvm_asm.h>
45 #include <asm/kvm_mmu.h>
46 #include <asm/kvm_emulate.h>
47 #include <asm/kvm_coproc.h>
48 #include <asm/sections.h>
51 __asm__(".arch_extension virt");
54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
55 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
57 /* Per-CPU variable containing the currently running vcpu. */
58 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
60 /* The VMID used in the VTTBR */
61 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
62 static u32 kvm_next_vmid;
63 static unsigned int kvm_vmid_bits __read_mostly;
64 static DEFINE_SPINLOCK(kvm_vmid_lock);
66 static bool vgic_present;
68 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
70 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
72 BUG_ON(preemptible());
73 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
77 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
78 * Must be called from non-preemptible context
80 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
82 BUG_ON(preemptible());
83 return __this_cpu_read(kvm_arm_running_vcpu);
87 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
89 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
91 return &kvm_arm_running_vcpu;
94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
96 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
99 int kvm_arch_hardware_setup(void)
104 void kvm_arch_check_processor_compat(void *rtn)
111 * kvm_arch_init_vm - initializes a VM data structure
112 * @kvm: pointer to the KVM struct
114 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
121 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
122 if (!kvm->arch.last_vcpu_ran)
125 for_each_possible_cpu(cpu)
126 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
128 ret = kvm_alloc_stage2_pgd(kvm);
132 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
134 goto out_free_stage2_pgd;
136 kvm_vgic_early_init(kvm);
138 /* Mark the initial VMID generation invalid */
139 kvm->arch.vmid_gen = 0;
141 /* The maximum number of VCPUs is limited by the host's GIC model */
142 kvm->arch.max_vcpus = vgic_present ?
143 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
147 kvm_free_stage2_pgd(kvm);
149 free_percpu(kvm->arch.last_vcpu_ran);
150 kvm->arch.last_vcpu_ran = NULL;
154 bool kvm_arch_has_vcpu_debugfs(void)
159 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
164 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
166 return VM_FAULT_SIGBUS;
171 * kvm_arch_destroy_vm - destroy the VM data structure
172 * @kvm: pointer to the KVM struct
174 void kvm_arch_destroy_vm(struct kvm *kvm)
178 free_percpu(kvm->arch.last_vcpu_ran);
179 kvm->arch.last_vcpu_ran = NULL;
181 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
183 kvm_arch_vcpu_free(kvm->vcpus[i]);
184 kvm->vcpus[i] = NULL;
188 kvm_vgic_destroy(kvm);
191 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
195 case KVM_CAP_IRQCHIP:
198 case KVM_CAP_IOEVENTFD:
199 case KVM_CAP_DEVICE_CTRL:
200 case KVM_CAP_USER_MEMORY:
201 case KVM_CAP_SYNC_MMU:
202 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
203 case KVM_CAP_ONE_REG:
204 case KVM_CAP_ARM_PSCI:
205 case KVM_CAP_ARM_PSCI_0_2:
206 case KVM_CAP_READONLY_MEM:
207 case KVM_CAP_MP_STATE:
208 case KVM_CAP_IMMEDIATE_EXIT:
211 case KVM_CAP_ARM_SET_DEVICE_ADDR:
214 case KVM_CAP_NR_VCPUS:
215 r = num_online_cpus();
217 case KVM_CAP_MAX_VCPUS:
220 case KVM_CAP_NR_MEMSLOTS:
221 r = KVM_USER_MEM_SLOTS;
223 case KVM_CAP_MSI_DEVID:
227 r = kvm->arch.vgic.msis_require_devid;
229 case KVM_CAP_ARM_USER_IRQ:
231 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
232 * (bump this number if adding more devices)
237 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
243 long kvm_arch_dev_ioctl(struct file *filp,
244 unsigned int ioctl, unsigned long arg)
250 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
253 struct kvm_vcpu *vcpu;
255 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
260 if (id >= kvm->arch.max_vcpus) {
265 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
271 err = kvm_vcpu_init(vcpu, kvm, id);
275 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
281 kvm_vcpu_uninit(vcpu);
283 kmem_cache_free(kvm_vcpu_cache, vcpu);
288 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
290 kvm_vgic_vcpu_early_init(vcpu);
293 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
295 kvm_mmu_free_memory_caches(vcpu);
296 kvm_timer_vcpu_terminate(vcpu);
297 kvm_vgic_vcpu_destroy(vcpu);
298 kvm_pmu_vcpu_destroy(vcpu);
299 kvm_vcpu_uninit(vcpu);
300 kmem_cache_free(kvm_vcpu_cache, vcpu);
303 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
305 kvm_arch_vcpu_free(vcpu);
308 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
310 return kvm_timer_should_fire(vcpu_vtimer(vcpu)) ||
311 kvm_timer_should_fire(vcpu_ptimer(vcpu));
314 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
316 kvm_timer_schedule(vcpu);
319 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
321 kvm_timer_unschedule(vcpu);
324 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
326 /* Force users to call KVM_ARM_VCPU_INIT */
327 vcpu->arch.target = -1;
328 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
330 /* Set up the timer */
331 kvm_timer_vcpu_init(vcpu);
333 kvm_arm_reset_debug_ptr(vcpu);
335 return kvm_vgic_vcpu_init(vcpu);
338 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
342 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
345 * We might get preempted before the vCPU actually runs, but
346 * over-invalidation doesn't affect correctness.
348 if (*last_ran != vcpu->vcpu_id) {
349 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
350 *last_ran = vcpu->vcpu_id;
354 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
356 kvm_arm_set_running_vcpu(vcpu);
361 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
367 kvm_arm_set_running_vcpu(NULL);
368 kvm_timer_vcpu_put(vcpu);
371 static void vcpu_power_off(struct kvm_vcpu *vcpu)
373 vcpu->arch.power_off = true;
374 kvm_make_request(KVM_REQ_SLEEP, vcpu);
378 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
379 struct kvm_mp_state *mp_state)
381 if (vcpu->arch.power_off)
382 mp_state->mp_state = KVM_MP_STATE_STOPPED;
384 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
389 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
390 struct kvm_mp_state *mp_state)
392 switch (mp_state->mp_state) {
393 case KVM_MP_STATE_RUNNABLE:
394 vcpu->arch.power_off = false;
396 case KVM_MP_STATE_STOPPED:
397 vcpu_power_off(vcpu);
407 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
408 * @v: The VCPU pointer
410 * If the guest CPU is not waiting for interrupts or an interrupt line is
411 * asserted, the CPU is by definition runnable.
413 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
415 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
416 && !v->arch.power_off && !v->arch.pause);
419 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
421 return vcpu_mode_priv(vcpu);
424 /* Just ensure a guest exit from a particular CPU */
425 static void exit_vm_noop(void *info)
429 void force_vm_exit(const cpumask_t *mask)
432 smp_call_function_many(mask, exit_vm_noop, NULL, true);
437 * need_new_vmid_gen - check that the VMID is still valid
438 * @kvm: The VM's VMID to check
440 * return true if there is a new generation of VMIDs being used
442 * The hardware supports only 256 values with the value zero reserved for the
443 * host, so we check if an assigned value belongs to a previous generation,
444 * which which requires us to assign a new value. If we're the first to use a
445 * VMID for the new generation, we must flush necessary caches and TLBs on all
448 static bool need_new_vmid_gen(struct kvm *kvm)
450 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
454 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
455 * @kvm The guest that we are about to run
457 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
458 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
461 static void update_vttbr(struct kvm *kvm)
463 phys_addr_t pgd_phys;
466 if (!need_new_vmid_gen(kvm))
469 spin_lock(&kvm_vmid_lock);
472 * We need to re-check the vmid_gen here to ensure that if another vcpu
473 * already allocated a valid vmid for this vm, then this vcpu should
476 if (!need_new_vmid_gen(kvm)) {
477 spin_unlock(&kvm_vmid_lock);
481 /* First user of a new VMID generation? */
482 if (unlikely(kvm_next_vmid == 0)) {
483 atomic64_inc(&kvm_vmid_gen);
487 * On SMP we know no other CPUs can use this CPU's or each
488 * other's VMID after force_vm_exit returns since the
489 * kvm_vmid_lock blocks them from reentry to the guest.
491 force_vm_exit(cpu_all_mask);
493 * Now broadcast TLB + ICACHE invalidation over the inner
494 * shareable domain to make sure all data structures are
497 kvm_call_hyp(__kvm_flush_vm_context);
500 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
501 kvm->arch.vmid = kvm_next_vmid;
503 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
505 /* update vttbr to be used with the new vmid */
506 pgd_phys = virt_to_phys(kvm->arch.pgd);
507 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
508 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
509 kvm->arch.vttbr = pgd_phys | vmid;
511 spin_unlock(&kvm_vmid_lock);
514 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
516 struct kvm *kvm = vcpu->kvm;
519 if (likely(vcpu->arch.has_run_once))
522 vcpu->arch.has_run_once = true;
525 * Map the VGIC hardware resources before running a vcpu the first
528 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
529 ret = kvm_vgic_map_resources(kvm);
534 ret = kvm_timer_enable(vcpu);
538 ret = kvm_arm_pmu_v3_enable(vcpu);
543 bool kvm_arch_intc_initialized(struct kvm *kvm)
545 return vgic_initialized(kvm);
548 void kvm_arm_halt_guest(struct kvm *kvm)
551 struct kvm_vcpu *vcpu;
553 kvm_for_each_vcpu(i, vcpu, kvm)
554 vcpu->arch.pause = true;
555 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
558 void kvm_arm_resume_guest(struct kvm *kvm)
561 struct kvm_vcpu *vcpu;
563 kvm_for_each_vcpu(i, vcpu, kvm) {
564 vcpu->arch.pause = false;
565 swake_up(kvm_arch_vcpu_wq(vcpu));
569 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
571 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
573 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
574 (!vcpu->arch.pause)));
576 if (vcpu->arch.power_off || vcpu->arch.pause) {
577 /* Awaken to handle a signal, request we sleep again later. */
578 kvm_make_request(KVM_REQ_SLEEP, vcpu);
582 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
584 return vcpu->arch.target >= 0;
587 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
589 if (kvm_request_pending(vcpu)) {
590 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
591 vcpu_req_sleep(vcpu);
594 * Clear IRQ_PENDING requests that were made to guarantee
595 * that a VCPU sees new virtual interrupts.
597 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
602 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
603 * @vcpu: The VCPU pointer
604 * @run: The kvm_run structure pointer used for userspace state exchange
606 * This function is called through the VCPU_RUN ioctl called from user space. It
607 * will execute VM code in a loop until the time slice for the process is used
608 * or some emulation is needed from user space in which case the function will
609 * return with return value 0 and with the kvm_run structure filled in with the
610 * required data for the requested emulation.
612 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
616 if (unlikely(!kvm_vcpu_initialized(vcpu)))
619 ret = kvm_vcpu_first_run_init(vcpu);
623 if (run->exit_reason == KVM_EXIT_MMIO) {
624 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
629 if (run->immediate_exit)
632 kvm_sigset_activate(vcpu);
635 run->exit_reason = KVM_EXIT_UNKNOWN;
638 * Check conditions before entering the guest
642 update_vttbr(vcpu->kvm);
644 check_vcpu_requests(vcpu);
647 * Preparing the interrupts to be injected also
648 * involves poking the GIC, which must be done in a
649 * non-preemptible context.
653 kvm_pmu_flush_hwstate(vcpu);
655 kvm_timer_flush_hwstate(vcpu);
656 kvm_vgic_flush_hwstate(vcpu);
661 * If we have a singal pending, or need to notify a userspace
662 * irqchip about timer or PMU level changes, then we exit (and
663 * update the timer level state in kvm_timer_update_run
666 if (signal_pending(current) ||
667 kvm_timer_should_notify_user(vcpu) ||
668 kvm_pmu_should_notify_user(vcpu)) {
670 run->exit_reason = KVM_EXIT_INTR;
674 * Ensure we set mode to IN_GUEST_MODE after we disable
675 * interrupts and before the final VCPU requests check.
676 * See the comment in kvm_vcpu_exiting_guest_mode() and
677 * Documentation/virtual/kvm/vcpu-requests.rst
679 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
681 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
682 kvm_request_pending(vcpu)) {
683 vcpu->mode = OUTSIDE_GUEST_MODE;
685 kvm_pmu_sync_hwstate(vcpu);
686 kvm_timer_sync_hwstate(vcpu);
687 kvm_vgic_sync_hwstate(vcpu);
692 kvm_arm_setup_debug(vcpu);
694 /**************************************************************
697 trace_kvm_entry(*vcpu_pc(vcpu));
698 guest_enter_irqoff();
700 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
702 vcpu->mode = OUTSIDE_GUEST_MODE;
706 *************************************************************/
708 kvm_arm_clear_debug(vcpu);
711 * We may have taken a host interrupt in HYP mode (ie
712 * while executing the guest). This interrupt is still
713 * pending, as we haven't serviced it yet!
715 * We're now back in SVC mode, with interrupts
716 * disabled. Enabling the interrupts now will have
717 * the effect of taking the interrupt again, in SVC
723 * We do local_irq_enable() before calling guest_exit() so
724 * that if a timer interrupt hits while running the guest we
725 * account that tick as being spent in the guest. We enable
726 * preemption after calling guest_exit() so that if we get
727 * preempted we make sure ticks after that is not counted as
731 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
734 * We must sync the PMU and timer state before the vgic state so
735 * that the vgic can properly sample the updated state of the
738 kvm_pmu_sync_hwstate(vcpu);
739 kvm_timer_sync_hwstate(vcpu);
741 kvm_vgic_sync_hwstate(vcpu);
745 ret = handle_exit(vcpu, run, ret);
748 /* Tell userspace about in-kernel device output levels */
749 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
750 kvm_timer_update_run(vcpu);
751 kvm_pmu_update_run(vcpu);
754 kvm_sigset_deactivate(vcpu);
759 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
765 if (number == KVM_ARM_IRQ_CPU_IRQ)
766 bit_index = __ffs(HCR_VI);
767 else /* KVM_ARM_IRQ_CPU_FIQ */
768 bit_index = __ffs(HCR_VF);
770 ptr = (unsigned long *)&vcpu->arch.irq_lines;
772 set = test_and_set_bit(bit_index, ptr);
774 set = test_and_clear_bit(bit_index, ptr);
777 * If we didn't change anything, no need to wake up or kick other CPUs
783 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
784 * trigger a world-switch round on the running physical CPU to set the
785 * virtual IRQ/FIQ fields in the HCR appropriately.
787 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
793 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
796 u32 irq = irq_level->irq;
797 unsigned int irq_type, vcpu_idx, irq_num;
798 int nrcpus = atomic_read(&kvm->online_vcpus);
799 struct kvm_vcpu *vcpu = NULL;
800 bool level = irq_level->level;
802 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
803 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
804 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
806 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
809 case KVM_ARM_IRQ_TYPE_CPU:
810 if (irqchip_in_kernel(kvm))
813 if (vcpu_idx >= nrcpus)
816 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
820 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
823 return vcpu_interrupt_line(vcpu, irq_num, level);
824 case KVM_ARM_IRQ_TYPE_PPI:
825 if (!irqchip_in_kernel(kvm))
828 if (vcpu_idx >= nrcpus)
831 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
835 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
838 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
839 case KVM_ARM_IRQ_TYPE_SPI:
840 if (!irqchip_in_kernel(kvm))
843 if (irq_num < VGIC_NR_PRIVATE_IRQS)
846 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
852 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
853 const struct kvm_vcpu_init *init)
856 int phys_target = kvm_target_cpu();
858 if (init->target != phys_target)
862 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
863 * use the same target.
865 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
868 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
869 for (i = 0; i < sizeof(init->features) * 8; i++) {
870 bool set = (init->features[i / 32] & (1 << (i % 32)));
872 if (set && i >= KVM_VCPU_MAX_FEATURES)
876 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
877 * use the same feature set.
879 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
880 test_bit(i, vcpu->arch.features) != set)
884 set_bit(i, vcpu->arch.features);
887 vcpu->arch.target = phys_target;
889 /* Now we know what it is, we can reset it. */
890 return kvm_reset_vcpu(vcpu);
894 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
895 struct kvm_vcpu_init *init)
899 ret = kvm_vcpu_set_target(vcpu, init);
904 * Ensure a rebooted VM will fault in RAM pages and detect if the
905 * guest MMU is turned off and flush the caches as needed.
907 if (vcpu->arch.has_run_once)
908 stage2_unmap_vm(vcpu->kvm);
910 vcpu_reset_hcr(vcpu);
913 * Handle the "start in power-off" case.
915 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
916 vcpu_power_off(vcpu);
918 vcpu->arch.power_off = false;
923 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
924 struct kvm_device_attr *attr)
928 switch (attr->group) {
930 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
937 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
938 struct kvm_device_attr *attr)
942 switch (attr->group) {
944 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
951 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
952 struct kvm_device_attr *attr)
956 switch (attr->group) {
958 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
965 long kvm_arch_vcpu_ioctl(struct file *filp,
966 unsigned int ioctl, unsigned long arg)
968 struct kvm_vcpu *vcpu = filp->private_data;
969 void __user *argp = (void __user *)arg;
970 struct kvm_device_attr attr;
973 case KVM_ARM_VCPU_INIT: {
974 struct kvm_vcpu_init init;
976 if (copy_from_user(&init, argp, sizeof(init)))
979 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
981 case KVM_SET_ONE_REG:
982 case KVM_GET_ONE_REG: {
983 struct kvm_one_reg reg;
985 if (unlikely(!kvm_vcpu_initialized(vcpu)))
988 if (copy_from_user(®, argp, sizeof(reg)))
990 if (ioctl == KVM_SET_ONE_REG)
991 return kvm_arm_set_reg(vcpu, ®);
993 return kvm_arm_get_reg(vcpu, ®);
995 case KVM_GET_REG_LIST: {
996 struct kvm_reg_list __user *user_list = argp;
997 struct kvm_reg_list reg_list;
1000 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1003 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1006 reg_list.n = kvm_arm_num_regs(vcpu);
1007 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1011 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1013 case KVM_SET_DEVICE_ATTR: {
1014 if (copy_from_user(&attr, argp, sizeof(attr)))
1016 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1018 case KVM_GET_DEVICE_ATTR: {
1019 if (copy_from_user(&attr, argp, sizeof(attr)))
1021 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1023 case KVM_HAS_DEVICE_ATTR: {
1024 if (copy_from_user(&attr, argp, sizeof(attr)))
1026 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1034 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1035 * @kvm: kvm instance
1036 * @log: slot id and address to which we copy the log
1038 * Steps 1-4 below provide general overview of dirty page logging. See
1039 * kvm_get_dirty_log_protect() function description for additional details.
1041 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1042 * always flush the TLB (step 4) even if previous step failed and the dirty
1043 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1044 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1045 * writes will be marked dirty for next log read.
1047 * 1. Take a snapshot of the bit and clear it if needed.
1048 * 2. Write protect the corresponding page.
1049 * 3. Copy the snapshot to the userspace.
1050 * 4. Flush TLB's if needed.
1052 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1054 bool is_dirty = false;
1057 mutex_lock(&kvm->slots_lock);
1059 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1062 kvm_flush_remote_tlbs(kvm);
1064 mutex_unlock(&kvm->slots_lock);
1068 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1069 struct kvm_arm_device_addr *dev_addr)
1071 unsigned long dev_id, type;
1073 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1074 KVM_ARM_DEVICE_ID_SHIFT;
1075 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1076 KVM_ARM_DEVICE_TYPE_SHIFT;
1079 case KVM_ARM_DEVICE_VGIC_V2:
1082 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1088 long kvm_arch_vm_ioctl(struct file *filp,
1089 unsigned int ioctl, unsigned long arg)
1091 struct kvm *kvm = filp->private_data;
1092 void __user *argp = (void __user *)arg;
1095 case KVM_CREATE_IRQCHIP: {
1099 mutex_lock(&kvm->lock);
1100 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1101 mutex_unlock(&kvm->lock);
1104 case KVM_ARM_SET_DEVICE_ADDR: {
1105 struct kvm_arm_device_addr dev_addr;
1107 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1109 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1111 case KVM_ARM_PREFERRED_TARGET: {
1113 struct kvm_vcpu_init init;
1115 err = kvm_vcpu_preferred_target(&init);
1119 if (copy_to_user(argp, &init, sizeof(init)))
1129 static void cpu_init_hyp_mode(void *dummy)
1131 phys_addr_t pgd_ptr;
1132 unsigned long hyp_stack_ptr;
1133 unsigned long stack_page;
1134 unsigned long vector_ptr;
1136 /* Switch from the HYP stub to our own HYP init vector */
1137 __hyp_set_vectors(kvm_get_idmap_vector());
1139 pgd_ptr = kvm_mmu_get_httbr();
1140 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1141 hyp_stack_ptr = stack_page + PAGE_SIZE;
1142 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1144 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1145 __cpu_init_stage2();
1147 kvm_arm_init_debug();
1150 static void cpu_hyp_reset(void)
1152 if (!is_kernel_in_hyp_mode())
1153 __hyp_reset_vectors();
1156 static void cpu_hyp_reinit(void)
1160 if (is_kernel_in_hyp_mode()) {
1162 * __cpu_init_stage2() is safe to call even if the PM
1163 * event was cancelled before the CPU was reset.
1165 __cpu_init_stage2();
1166 kvm_timer_init_vhe();
1168 cpu_init_hyp_mode(NULL);
1172 kvm_vgic_init_cpu_hardware();
1175 static void _kvm_arch_hardware_enable(void *discard)
1177 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1179 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1183 int kvm_arch_hardware_enable(void)
1185 _kvm_arch_hardware_enable(NULL);
1189 static void _kvm_arch_hardware_disable(void *discard)
1191 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1193 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1197 void kvm_arch_hardware_disable(void)
1199 _kvm_arch_hardware_disable(NULL);
1202 #ifdef CONFIG_CPU_PM
1203 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1208 * kvm_arm_hardware_enabled is left with its old value over
1209 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1214 if (__this_cpu_read(kvm_arm_hardware_enabled))
1216 * don't update kvm_arm_hardware_enabled here
1217 * so that the hardware will be re-enabled
1218 * when we resume. See below.
1224 if (__this_cpu_read(kvm_arm_hardware_enabled))
1225 /* The hardware was enabled before suspend. */
1235 static struct notifier_block hyp_init_cpu_pm_nb = {
1236 .notifier_call = hyp_init_cpu_pm_notifier,
1239 static void __init hyp_cpu_pm_init(void)
1241 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1243 static void __init hyp_cpu_pm_exit(void)
1245 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1248 static inline void hyp_cpu_pm_init(void)
1251 static inline void hyp_cpu_pm_exit(void)
1256 static void teardown_common_resources(void)
1258 free_percpu(kvm_host_cpu_state);
1261 static int init_common_resources(void)
1263 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1264 if (!kvm_host_cpu_state) {
1265 kvm_err("Cannot allocate host CPU state\n");
1269 /* set size of VMID supported by CPU */
1270 kvm_vmid_bits = kvm_get_vmid_bits();
1271 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1276 static int init_subsystems(void)
1281 * Enable hardware so that subsystem initialisation can access EL2.
1283 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1286 * Register CPU lower-power notifier
1291 * Init HYP view of VGIC
1293 err = kvm_vgic_hyp_init();
1296 vgic_present = true;
1300 vgic_present = false;
1308 * Init HYP architected timer support
1310 err = kvm_timer_hyp_init();
1315 kvm_coproc_table_init();
1318 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1323 static void teardown_hyp_mode(void)
1328 for_each_possible_cpu(cpu)
1329 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1334 * Inits Hyp-mode on all online CPUs
1336 static int init_hyp_mode(void)
1342 * Allocate Hyp PGD and setup Hyp identity mapping
1344 err = kvm_mmu_init();
1349 * Allocate stack pages for Hypervisor-mode
1351 for_each_possible_cpu(cpu) {
1352 unsigned long stack_page;
1354 stack_page = __get_free_page(GFP_KERNEL);
1360 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1364 * Map the Hyp-code called directly from the host
1366 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1367 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1369 kvm_err("Cannot map world-switch code\n");
1373 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1374 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1376 kvm_err("Cannot map rodata section\n");
1380 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1381 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1383 kvm_err("Cannot map bss section\n");
1387 err = kvm_map_vectors();
1389 kvm_err("Cannot map vectors\n");
1394 * Map the Hyp stack pages
1396 for_each_possible_cpu(cpu) {
1397 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1398 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1402 kvm_err("Cannot map hyp stack\n");
1407 for_each_possible_cpu(cpu) {
1408 kvm_cpu_context_t *cpu_ctxt;
1410 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1411 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1414 kvm_err("Cannot map host CPU state: %d\n", err);
1422 teardown_hyp_mode();
1423 kvm_err("error initializing Hyp mode: %d\n", err);
1427 static void check_kvm_target_cpu(void *ret)
1429 *(int *)ret = kvm_target_cpu();
1432 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1434 struct kvm_vcpu *vcpu;
1437 mpidr &= MPIDR_HWID_BITMASK;
1438 kvm_for_each_vcpu(i, vcpu, kvm) {
1439 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1446 * Initialize Hyp-mode and memory mappings on all CPUs.
1448 int kvm_arch_init(void *opaque)
1454 if (!is_hyp_mode_available()) {
1455 kvm_err("HYP mode not available\n");
1459 for_each_online_cpu(cpu) {
1460 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1462 kvm_err("Error, CPU %d not supported!\n", cpu);
1467 err = init_common_resources();
1471 in_hyp_mode = is_kernel_in_hyp_mode();
1474 err = init_hyp_mode();
1479 err = init_subsystems();
1484 kvm_info("VHE mode initialized successfully\n");
1486 kvm_info("Hyp mode initialized successfully\n");
1492 teardown_hyp_mode();
1494 teardown_common_resources();
1498 /* NOP: Compiling as a module not supported */
1499 void kvm_arch_exit(void)
1501 kvm_perf_teardown();
1504 static int arm_init(void)
1506 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1510 module_init(arm_init);