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>
33 #define CREATE_TRACE_POINTS
36 #include <asm/uaccess.h>
37 #include <asm/ptrace.h>
39 #include <asm/tlbflush.h>
40 #include <asm/cacheflush.h>
42 #include <asm/kvm_arm.h>
43 #include <asm/kvm_asm.h>
44 #include <asm/kvm_mmu.h>
45 #include <asm/kvm_emulate.h>
46 #include <asm/kvm_coproc.h>
47 #include <asm/kvm_psci.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;
56 static unsigned long hyp_default_vectors;
58 /* Per-CPU variable containing the currently running vcpu. */
59 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
61 /* The VMID used in the VTTBR */
62 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
63 static u32 kvm_next_vmid;
64 static unsigned int kvm_vmid_bits __read_mostly;
65 static DEFINE_SPINLOCK(kvm_vmid_lock);
67 static bool vgic_present;
69 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
71 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
73 BUG_ON(preemptible());
74 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
78 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
79 * Must be called from non-preemptible context
81 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
83 BUG_ON(preemptible());
84 return __this_cpu_read(kvm_arm_running_vcpu);
88 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
90 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
92 return &kvm_arm_running_vcpu;
95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
97 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
100 int kvm_arch_hardware_setup(void)
105 void kvm_arch_check_processor_compat(void *rtn)
112 * kvm_arch_init_vm - initializes a VM data structure
113 * @kvm: pointer to the KVM struct
115 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
122 ret = kvm_alloc_stage2_pgd(kvm);
126 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
128 goto out_free_stage2_pgd;
130 kvm_vgic_early_init(kvm);
133 /* Mark the initial VMID generation invalid */
134 kvm->arch.vmid_gen = 0;
136 /* The maximum number of VCPUs is limited by the host's GIC model */
137 kvm->arch.max_vcpus = vgic_present ?
138 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
142 kvm_free_stage2_pgd(kvm);
147 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
149 return VM_FAULT_SIGBUS;
154 * kvm_arch_destroy_vm - destroy the VM data structure
155 * @kvm: pointer to the KVM struct
157 void kvm_arch_destroy_vm(struct kvm *kvm)
161 kvm_free_stage2_pgd(kvm);
163 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
165 kvm_arch_vcpu_free(kvm->vcpus[i]);
166 kvm->vcpus[i] = NULL;
170 kvm_vgic_destroy(kvm);
173 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
177 case KVM_CAP_IRQCHIP:
180 case KVM_CAP_IOEVENTFD:
181 case KVM_CAP_DEVICE_CTRL:
182 case KVM_CAP_USER_MEMORY:
183 case KVM_CAP_SYNC_MMU:
184 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
185 case KVM_CAP_ONE_REG:
186 case KVM_CAP_ARM_PSCI:
187 case KVM_CAP_ARM_PSCI_0_2:
188 case KVM_CAP_READONLY_MEM:
189 case KVM_CAP_MP_STATE:
192 case KVM_CAP_COALESCED_MMIO:
193 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
195 case KVM_CAP_ARM_SET_DEVICE_ADDR:
198 case KVM_CAP_NR_VCPUS:
199 r = num_online_cpus();
201 case KVM_CAP_MAX_VCPUS:
205 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
211 long kvm_arch_dev_ioctl(struct file *filp,
212 unsigned int ioctl, unsigned long arg)
218 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
221 struct kvm_vcpu *vcpu;
223 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
228 if (id >= kvm->arch.max_vcpus) {
233 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
239 err = kvm_vcpu_init(vcpu, kvm, id);
243 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
249 kvm_vcpu_uninit(vcpu);
251 kmem_cache_free(kvm_vcpu_cache, vcpu);
256 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
258 kvm_vgic_vcpu_early_init(vcpu);
261 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
263 kvm_mmu_free_memory_caches(vcpu);
264 kvm_timer_vcpu_terminate(vcpu);
265 kvm_vgic_vcpu_destroy(vcpu);
266 kvm_pmu_vcpu_destroy(vcpu);
267 kvm_vcpu_uninit(vcpu);
268 kmem_cache_free(kvm_vcpu_cache, vcpu);
271 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
273 kvm_arch_vcpu_free(vcpu);
276 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
278 return kvm_timer_should_fire(vcpu);
281 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
283 kvm_timer_schedule(vcpu);
286 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
288 kvm_timer_unschedule(vcpu);
291 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
293 /* Force users to call KVM_ARM_VCPU_INIT */
294 vcpu->arch.target = -1;
295 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
297 /* Set up the timer */
298 kvm_timer_vcpu_init(vcpu);
300 kvm_arm_reset_debug_ptr(vcpu);
305 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
308 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
310 kvm_arm_set_running_vcpu(vcpu);
313 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
316 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
317 * if the vcpu is no longer assigned to a cpu. This is used for the
318 * optimized make_all_cpus_request path.
322 kvm_arm_set_running_vcpu(NULL);
323 kvm_timer_vcpu_put(vcpu);
326 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
327 struct kvm_mp_state *mp_state)
329 if (vcpu->arch.power_off)
330 mp_state->mp_state = KVM_MP_STATE_STOPPED;
332 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
337 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
338 struct kvm_mp_state *mp_state)
340 switch (mp_state->mp_state) {
341 case KVM_MP_STATE_RUNNABLE:
342 vcpu->arch.power_off = false;
344 case KVM_MP_STATE_STOPPED:
345 vcpu->arch.power_off = true;
355 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
356 * @v: The VCPU pointer
358 * If the guest CPU is not waiting for interrupts or an interrupt line is
359 * asserted, the CPU is by definition runnable.
361 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
363 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
364 && !v->arch.power_off && !v->arch.pause);
367 /* Just ensure a guest exit from a particular CPU */
368 static void exit_vm_noop(void *info)
372 void force_vm_exit(const cpumask_t *mask)
375 smp_call_function_many(mask, exit_vm_noop, NULL, true);
380 * need_new_vmid_gen - check that the VMID is still valid
381 * @kvm: The VM's VMID to check
383 * return true if there is a new generation of VMIDs being used
385 * The hardware supports only 256 values with the value zero reserved for the
386 * host, so we check if an assigned value belongs to a previous generation,
387 * which which requires us to assign a new value. If we're the first to use a
388 * VMID for the new generation, we must flush necessary caches and TLBs on all
391 static bool need_new_vmid_gen(struct kvm *kvm)
393 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
397 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
398 * @kvm The guest that we are about to run
400 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
401 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
404 static void update_vttbr(struct kvm *kvm)
406 phys_addr_t pgd_phys;
409 if (!need_new_vmid_gen(kvm))
412 spin_lock(&kvm_vmid_lock);
415 * We need to re-check the vmid_gen here to ensure that if another vcpu
416 * already allocated a valid vmid for this vm, then this vcpu should
419 if (!need_new_vmid_gen(kvm)) {
420 spin_unlock(&kvm_vmid_lock);
424 /* First user of a new VMID generation? */
425 if (unlikely(kvm_next_vmid == 0)) {
426 atomic64_inc(&kvm_vmid_gen);
430 * On SMP we know no other CPUs can use this CPU's or each
431 * other's VMID after force_vm_exit returns since the
432 * kvm_vmid_lock blocks them from reentry to the guest.
434 force_vm_exit(cpu_all_mask);
436 * Now broadcast TLB + ICACHE invalidation over the inner
437 * shareable domain to make sure all data structures are
440 kvm_call_hyp(__kvm_flush_vm_context);
443 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
444 kvm->arch.vmid = kvm_next_vmid;
446 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
448 /* update vttbr to be used with the new vmid */
449 pgd_phys = virt_to_phys(kvm->arch.pgd);
450 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
451 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
452 kvm->arch.vttbr = pgd_phys | vmid;
454 spin_unlock(&kvm_vmid_lock);
457 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
459 struct kvm *kvm = vcpu->kvm;
462 if (likely(vcpu->arch.has_run_once))
465 vcpu->arch.has_run_once = true;
468 * Map the VGIC hardware resources before running a vcpu the first
471 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
472 ret = kvm_vgic_map_resources(kvm);
478 * Enable the arch timers only if we have an in-kernel VGIC
479 * and it has been properly initialized, since we cannot handle
480 * interrupts from the virtual timer with a userspace gic.
482 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
483 ret = kvm_timer_enable(vcpu);
488 bool kvm_arch_intc_initialized(struct kvm *kvm)
490 return vgic_initialized(kvm);
493 void kvm_arm_halt_guest(struct kvm *kvm)
496 struct kvm_vcpu *vcpu;
498 kvm_for_each_vcpu(i, vcpu, kvm)
499 vcpu->arch.pause = true;
500 kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
503 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
505 vcpu->arch.pause = true;
509 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
511 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
513 vcpu->arch.pause = false;
517 void kvm_arm_resume_guest(struct kvm *kvm)
520 struct kvm_vcpu *vcpu;
522 kvm_for_each_vcpu(i, vcpu, kvm)
523 kvm_arm_resume_vcpu(vcpu);
526 static void vcpu_sleep(struct kvm_vcpu *vcpu)
528 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
530 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
531 (!vcpu->arch.pause)));
534 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
536 return vcpu->arch.target >= 0;
540 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
541 * @vcpu: The VCPU pointer
542 * @run: The kvm_run structure pointer used for userspace state exchange
544 * This function is called through the VCPU_RUN ioctl called from user space. It
545 * will execute VM code in a loop until the time slice for the process is used
546 * or some emulation is needed from user space in which case the function will
547 * return with return value 0 and with the kvm_run structure filled in with the
548 * required data for the requested emulation.
550 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
555 if (unlikely(!kvm_vcpu_initialized(vcpu)))
558 ret = kvm_vcpu_first_run_init(vcpu);
562 if (run->exit_reason == KVM_EXIT_MMIO) {
563 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
568 if (vcpu->sigset_active)
569 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
572 run->exit_reason = KVM_EXIT_UNKNOWN;
575 * Check conditions before entering the guest
579 update_vttbr(vcpu->kvm);
581 if (vcpu->arch.power_off || vcpu->arch.pause)
585 * Preparing the interrupts to be injected also
586 * involves poking the GIC, which must be done in a
587 * non-preemptible context.
590 kvm_pmu_flush_hwstate(vcpu);
591 kvm_timer_flush_hwstate(vcpu);
592 kvm_vgic_flush_hwstate(vcpu);
597 * Re-check atomic conditions
599 if (signal_pending(current)) {
601 run->exit_reason = KVM_EXIT_INTR;
604 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
605 vcpu->arch.power_off || vcpu->arch.pause) {
607 kvm_pmu_sync_hwstate(vcpu);
608 kvm_timer_sync_hwstate(vcpu);
609 kvm_vgic_sync_hwstate(vcpu);
614 kvm_arm_setup_debug(vcpu);
616 /**************************************************************
619 trace_kvm_entry(*vcpu_pc(vcpu));
620 guest_enter_irqoff();
621 vcpu->mode = IN_GUEST_MODE;
623 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
625 vcpu->mode = OUTSIDE_GUEST_MODE;
629 *************************************************************/
631 kvm_arm_clear_debug(vcpu);
634 * We may have taken a host interrupt in HYP mode (ie
635 * while executing the guest). This interrupt is still
636 * pending, as we haven't serviced it yet!
638 * We're now back in SVC mode, with interrupts
639 * disabled. Enabling the interrupts now will have
640 * the effect of taking the interrupt again, in SVC
646 * We do local_irq_enable() before calling guest_exit() so
647 * that if a timer interrupt hits while running the guest we
648 * account that tick as being spent in the guest. We enable
649 * preemption after calling guest_exit() so that if we get
650 * preempted we make sure ticks after that is not counted as
654 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
657 * We must sync the PMU and timer state before the vgic state so
658 * that the vgic can properly sample the updated state of the
661 kvm_pmu_sync_hwstate(vcpu);
662 kvm_timer_sync_hwstate(vcpu);
664 kvm_vgic_sync_hwstate(vcpu);
668 ret = handle_exit(vcpu, run, ret);
671 if (vcpu->sigset_active)
672 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
676 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
682 if (number == KVM_ARM_IRQ_CPU_IRQ)
683 bit_index = __ffs(HCR_VI);
684 else /* KVM_ARM_IRQ_CPU_FIQ */
685 bit_index = __ffs(HCR_VF);
687 ptr = (unsigned long *)&vcpu->arch.irq_lines;
689 set = test_and_set_bit(bit_index, ptr);
691 set = test_and_clear_bit(bit_index, ptr);
694 * If we didn't change anything, no need to wake up or kick other CPUs
700 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
701 * trigger a world-switch round on the running physical CPU to set the
702 * virtual IRQ/FIQ fields in the HCR appropriately.
709 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
712 u32 irq = irq_level->irq;
713 unsigned int irq_type, vcpu_idx, irq_num;
714 int nrcpus = atomic_read(&kvm->online_vcpus);
715 struct kvm_vcpu *vcpu = NULL;
716 bool level = irq_level->level;
718 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
719 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
720 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
722 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
725 case KVM_ARM_IRQ_TYPE_CPU:
726 if (irqchip_in_kernel(kvm))
729 if (vcpu_idx >= nrcpus)
732 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
736 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
739 return vcpu_interrupt_line(vcpu, irq_num, level);
740 case KVM_ARM_IRQ_TYPE_PPI:
741 if (!irqchip_in_kernel(kvm))
744 if (vcpu_idx >= nrcpus)
747 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
751 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
754 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
755 case KVM_ARM_IRQ_TYPE_SPI:
756 if (!irqchip_in_kernel(kvm))
759 if (irq_num < VGIC_NR_PRIVATE_IRQS)
762 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
768 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
769 const struct kvm_vcpu_init *init)
772 int phys_target = kvm_target_cpu();
774 if (init->target != phys_target)
778 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
779 * use the same target.
781 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
784 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
785 for (i = 0; i < sizeof(init->features) * 8; i++) {
786 bool set = (init->features[i / 32] & (1 << (i % 32)));
788 if (set && i >= KVM_VCPU_MAX_FEATURES)
792 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
793 * use the same feature set.
795 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
796 test_bit(i, vcpu->arch.features) != set)
800 set_bit(i, vcpu->arch.features);
803 vcpu->arch.target = phys_target;
805 /* Now we know what it is, we can reset it. */
806 return kvm_reset_vcpu(vcpu);
810 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
811 struct kvm_vcpu_init *init)
815 ret = kvm_vcpu_set_target(vcpu, init);
820 * Ensure a rebooted VM will fault in RAM pages and detect if the
821 * guest MMU is turned off and flush the caches as needed.
823 if (vcpu->arch.has_run_once)
824 stage2_unmap_vm(vcpu->kvm);
826 vcpu_reset_hcr(vcpu);
829 * Handle the "start in power-off" case.
831 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
832 vcpu->arch.power_off = true;
834 vcpu->arch.power_off = false;
839 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
840 struct kvm_device_attr *attr)
844 switch (attr->group) {
846 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
853 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
854 struct kvm_device_attr *attr)
858 switch (attr->group) {
860 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
867 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
868 struct kvm_device_attr *attr)
872 switch (attr->group) {
874 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
881 long kvm_arch_vcpu_ioctl(struct file *filp,
882 unsigned int ioctl, unsigned long arg)
884 struct kvm_vcpu *vcpu = filp->private_data;
885 void __user *argp = (void __user *)arg;
886 struct kvm_device_attr attr;
889 case KVM_ARM_VCPU_INIT: {
890 struct kvm_vcpu_init init;
892 if (copy_from_user(&init, argp, sizeof(init)))
895 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
897 case KVM_SET_ONE_REG:
898 case KVM_GET_ONE_REG: {
899 struct kvm_one_reg reg;
901 if (unlikely(!kvm_vcpu_initialized(vcpu)))
904 if (copy_from_user(®, argp, sizeof(reg)))
906 if (ioctl == KVM_SET_ONE_REG)
907 return kvm_arm_set_reg(vcpu, ®);
909 return kvm_arm_get_reg(vcpu, ®);
911 case KVM_GET_REG_LIST: {
912 struct kvm_reg_list __user *user_list = argp;
913 struct kvm_reg_list reg_list;
916 if (unlikely(!kvm_vcpu_initialized(vcpu)))
919 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
922 reg_list.n = kvm_arm_num_regs(vcpu);
923 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
927 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
929 case KVM_SET_DEVICE_ATTR: {
930 if (copy_from_user(&attr, argp, sizeof(attr)))
932 return kvm_arm_vcpu_set_attr(vcpu, &attr);
934 case KVM_GET_DEVICE_ATTR: {
935 if (copy_from_user(&attr, argp, sizeof(attr)))
937 return kvm_arm_vcpu_get_attr(vcpu, &attr);
939 case KVM_HAS_DEVICE_ATTR: {
940 if (copy_from_user(&attr, argp, sizeof(attr)))
942 return kvm_arm_vcpu_has_attr(vcpu, &attr);
950 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
952 * @log: slot id and address to which we copy the log
954 * Steps 1-4 below provide general overview of dirty page logging. See
955 * kvm_get_dirty_log_protect() function description for additional details.
957 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
958 * always flush the TLB (step 4) even if previous step failed and the dirty
959 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
960 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
961 * writes will be marked dirty for next log read.
963 * 1. Take a snapshot of the bit and clear it if needed.
964 * 2. Write protect the corresponding page.
965 * 3. Copy the snapshot to the userspace.
966 * 4. Flush TLB's if needed.
968 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
970 bool is_dirty = false;
973 mutex_lock(&kvm->slots_lock);
975 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
978 kvm_flush_remote_tlbs(kvm);
980 mutex_unlock(&kvm->slots_lock);
984 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
985 struct kvm_arm_device_addr *dev_addr)
987 unsigned long dev_id, type;
989 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
990 KVM_ARM_DEVICE_ID_SHIFT;
991 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
992 KVM_ARM_DEVICE_TYPE_SHIFT;
995 case KVM_ARM_DEVICE_VGIC_V2:
998 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1004 long kvm_arch_vm_ioctl(struct file *filp,
1005 unsigned int ioctl, unsigned long arg)
1007 struct kvm *kvm = filp->private_data;
1008 void __user *argp = (void __user *)arg;
1011 case KVM_CREATE_IRQCHIP: {
1015 mutex_lock(&kvm->lock);
1016 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1017 mutex_unlock(&kvm->lock);
1020 case KVM_ARM_SET_DEVICE_ADDR: {
1021 struct kvm_arm_device_addr dev_addr;
1023 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1025 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1027 case KVM_ARM_PREFERRED_TARGET: {
1029 struct kvm_vcpu_init init;
1031 err = kvm_vcpu_preferred_target(&init);
1035 if (copy_to_user(argp, &init, sizeof(init)))
1045 static void cpu_init_hyp_mode(void *dummy)
1047 phys_addr_t pgd_ptr;
1048 unsigned long hyp_stack_ptr;
1049 unsigned long stack_page;
1050 unsigned long vector_ptr;
1052 /* Switch from the HYP stub to our own HYP init vector */
1053 __hyp_set_vectors(kvm_get_idmap_vector());
1055 pgd_ptr = kvm_mmu_get_httbr();
1056 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1057 hyp_stack_ptr = stack_page + PAGE_SIZE;
1058 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1060 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1061 __cpu_init_stage2();
1063 kvm_arm_init_debug();
1066 static void cpu_hyp_reinit(void)
1068 if (is_kernel_in_hyp_mode()) {
1070 * __cpu_init_stage2() is safe to call even if the PM
1071 * event was cancelled before the CPU was reset.
1073 __cpu_init_stage2();
1075 if (__hyp_get_vectors() == hyp_default_vectors)
1076 cpu_init_hyp_mode(NULL);
1080 static void cpu_hyp_reset(void)
1082 if (!is_kernel_in_hyp_mode())
1083 __cpu_reset_hyp_mode(hyp_default_vectors,
1084 kvm_get_idmap_start());
1087 static void _kvm_arch_hardware_enable(void *discard)
1089 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1091 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1095 int kvm_arch_hardware_enable(void)
1097 _kvm_arch_hardware_enable(NULL);
1101 static void _kvm_arch_hardware_disable(void *discard)
1103 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1105 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1109 void kvm_arch_hardware_disable(void)
1111 _kvm_arch_hardware_disable(NULL);
1114 #ifdef CONFIG_CPU_PM
1115 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1120 * kvm_arm_hardware_enabled is left with its old value over
1121 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1126 if (__this_cpu_read(kvm_arm_hardware_enabled))
1128 * don't update kvm_arm_hardware_enabled here
1129 * so that the hardware will be re-enabled
1130 * when we resume. See below.
1136 if (__this_cpu_read(kvm_arm_hardware_enabled))
1137 /* The hardware was enabled before suspend. */
1147 static struct notifier_block hyp_init_cpu_pm_nb = {
1148 .notifier_call = hyp_init_cpu_pm_notifier,
1151 static void __init hyp_cpu_pm_init(void)
1153 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1155 static void __init hyp_cpu_pm_exit(void)
1157 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1160 static inline void hyp_cpu_pm_init(void)
1163 static inline void hyp_cpu_pm_exit(void)
1168 static void teardown_common_resources(void)
1170 free_percpu(kvm_host_cpu_state);
1173 static int init_common_resources(void)
1175 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1176 if (!kvm_host_cpu_state) {
1177 kvm_err("Cannot allocate host CPU state\n");
1184 static int init_subsystems(void)
1189 * Enable hardware so that subsystem initialisation can access EL2.
1191 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1194 * Register CPU lower-power notifier
1199 * Init HYP view of VGIC
1201 err = kvm_vgic_hyp_init();
1204 vgic_present = true;
1208 vgic_present = false;
1216 * Init HYP architected timer support
1218 err = kvm_timer_hyp_init();
1223 kvm_coproc_table_init();
1226 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1231 static void teardown_hyp_mode(void)
1235 if (is_kernel_in_hyp_mode())
1239 for_each_possible_cpu(cpu)
1240 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1244 static int init_vhe_mode(void)
1246 /* set size of VMID supported by CPU */
1247 kvm_vmid_bits = kvm_get_vmid_bits();
1248 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1250 kvm_info("VHE mode initialized successfully\n");
1255 * Inits Hyp-mode on all online CPUs
1257 static int init_hyp_mode(void)
1263 * Allocate Hyp PGD and setup Hyp identity mapping
1265 err = kvm_mmu_init();
1270 * It is probably enough to obtain the default on one
1271 * CPU. It's unlikely to be different on the others.
1273 hyp_default_vectors = __hyp_get_vectors();
1276 * Allocate stack pages for Hypervisor-mode
1278 for_each_possible_cpu(cpu) {
1279 unsigned long stack_page;
1281 stack_page = __get_free_page(GFP_KERNEL);
1287 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1291 * Map the Hyp-code called directly from the host
1293 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1294 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1296 kvm_err("Cannot map world-switch code\n");
1300 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1301 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1303 kvm_err("Cannot map rodata section\n");
1308 * Map the Hyp stack pages
1310 for_each_possible_cpu(cpu) {
1311 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1312 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1316 kvm_err("Cannot map hyp stack\n");
1321 for_each_possible_cpu(cpu) {
1322 kvm_cpu_context_t *cpu_ctxt;
1324 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1325 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1328 kvm_err("Cannot map host CPU state: %d\n", err);
1333 /* set size of VMID supported by CPU */
1334 kvm_vmid_bits = kvm_get_vmid_bits();
1335 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1337 kvm_info("Hyp mode initialized successfully\n");
1342 teardown_hyp_mode();
1343 kvm_err("error initializing Hyp mode: %d\n", err);
1347 static void check_kvm_target_cpu(void *ret)
1349 *(int *)ret = kvm_target_cpu();
1352 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1354 struct kvm_vcpu *vcpu;
1357 mpidr &= MPIDR_HWID_BITMASK;
1358 kvm_for_each_vcpu(i, vcpu, kvm) {
1359 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1366 * Initialize Hyp-mode and memory mappings on all CPUs.
1368 int kvm_arch_init(void *opaque)
1373 if (!is_hyp_mode_available()) {
1374 kvm_err("HYP mode not available\n");
1378 for_each_online_cpu(cpu) {
1379 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1381 kvm_err("Error, CPU %d not supported!\n", cpu);
1386 err = init_common_resources();
1390 if (is_kernel_in_hyp_mode())
1391 err = init_vhe_mode();
1393 err = init_hyp_mode();
1397 err = init_subsystems();
1404 teardown_hyp_mode();
1406 teardown_common_resources();
1410 /* NOP: Compiling as a module not supported */
1411 void kvm_arch_exit(void)
1413 kvm_perf_teardown();
1416 static int arm_init(void)
1418 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1422 module_init(arm_init);