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.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/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>
32 #define CREATE_TRACE_POINTS
35 #include <asm/uaccess.h>
36 #include <asm/ptrace.h>
38 #include <asm/tlbflush.h>
39 #include <asm/cacheflush.h>
41 #include <asm/kvm_arm.h>
42 #include <asm/kvm_asm.h>
43 #include <asm/kvm_mmu.h>
44 #include <asm/kvm_emulate.h>
45 #include <asm/kvm_coproc.h>
46 #include <asm/kvm_psci.h>
49 __asm__(".arch_extension virt");
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
54 static unsigned long hyp_default_vectors;
56 /* Per-CPU variable containing the currently running vcpu. */
57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
59 /* The VMID used in the VTTBR */
60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
61 static u8 kvm_next_vmid;
62 static DEFINE_SPINLOCK(kvm_vmid_lock);
64 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
66 BUG_ON(preemptible());
67 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
71 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
72 * Must be called from non-preemptible context
74 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
76 BUG_ON(preemptible());
77 return __this_cpu_read(kvm_arm_running_vcpu);
81 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
83 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
85 return &kvm_arm_running_vcpu;
88 int kvm_arch_hardware_enable(void)
93 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
95 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
98 int kvm_arch_hardware_setup(void)
103 void kvm_arch_check_processor_compat(void *rtn)
110 * kvm_arch_init_vm - initializes a VM data structure
111 * @kvm: pointer to the KVM struct
113 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
120 ret = kvm_alloc_stage2_pgd(kvm);
124 ret = create_hyp_mappings(kvm, kvm + 1);
126 goto out_free_stage2_pgd;
128 kvm_vgic_early_init(kvm);
131 /* Mark the initial VMID generation invalid */
132 kvm->arch.vmid_gen = 0;
134 /* The maximum number of VCPUs is limited by the host's GIC model */
135 kvm->arch.max_vcpus = kvm_vgic_get_max_vcpus();
139 kvm_free_stage2_pgd(kvm);
144 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
146 return VM_FAULT_SIGBUS;
151 * kvm_arch_destroy_vm - destroy the VM data structure
152 * @kvm: pointer to the KVM struct
154 void kvm_arch_destroy_vm(struct kvm *kvm)
158 kvm_free_stage2_pgd(kvm);
160 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
162 kvm_arch_vcpu_free(kvm->vcpus[i]);
163 kvm->vcpus[i] = NULL;
167 kvm_vgic_destroy(kvm);
170 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
174 case KVM_CAP_IRQCHIP:
175 case KVM_CAP_IOEVENTFD:
176 case KVM_CAP_DEVICE_CTRL:
177 case KVM_CAP_USER_MEMORY:
178 case KVM_CAP_SYNC_MMU:
179 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
180 case KVM_CAP_ONE_REG:
181 case KVM_CAP_ARM_PSCI:
182 case KVM_CAP_ARM_PSCI_0_2:
183 case KVM_CAP_READONLY_MEM:
184 case KVM_CAP_MP_STATE:
187 case KVM_CAP_COALESCED_MMIO:
188 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
190 case KVM_CAP_ARM_SET_DEVICE_ADDR:
193 case KVM_CAP_NR_VCPUS:
194 r = num_online_cpus();
196 case KVM_CAP_MAX_VCPUS:
200 r = kvm_arch_dev_ioctl_check_extension(ext);
206 long kvm_arch_dev_ioctl(struct file *filp,
207 unsigned int ioctl, unsigned long arg)
213 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
216 struct kvm_vcpu *vcpu;
218 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
223 if (id >= kvm->arch.max_vcpus) {
228 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
234 err = kvm_vcpu_init(vcpu, kvm, id);
238 err = create_hyp_mappings(vcpu, vcpu + 1);
244 kvm_vcpu_uninit(vcpu);
246 kmem_cache_free(kvm_vcpu_cache, vcpu);
251 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
253 kvm_vgic_vcpu_early_init(vcpu);
256 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
258 kvm_mmu_free_memory_caches(vcpu);
259 kvm_timer_vcpu_terminate(vcpu);
260 kvm_vgic_vcpu_destroy(vcpu);
261 kmem_cache_free(kvm_vcpu_cache, vcpu);
264 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
266 kvm_arch_vcpu_free(vcpu);
269 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
271 return kvm_timer_should_fire(vcpu);
274 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
276 /* Force users to call KVM_ARM_VCPU_INIT */
277 vcpu->arch.target = -1;
278 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
280 /* Set up the timer */
281 kvm_timer_vcpu_init(vcpu);
283 kvm_arm_reset_debug_ptr(vcpu);
288 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
291 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
293 kvm_arm_set_running_vcpu(vcpu);
296 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
299 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
300 * if the vcpu is no longer assigned to a cpu. This is used for the
301 * optimized make_all_cpus_request path.
305 kvm_arm_set_running_vcpu(NULL);
308 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
309 struct kvm_mp_state *mp_state)
311 if (vcpu->arch.pause)
312 mp_state->mp_state = KVM_MP_STATE_STOPPED;
314 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
319 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
320 struct kvm_mp_state *mp_state)
322 switch (mp_state->mp_state) {
323 case KVM_MP_STATE_RUNNABLE:
324 vcpu->arch.pause = false;
326 case KVM_MP_STATE_STOPPED:
327 vcpu->arch.pause = true;
337 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
338 * @v: The VCPU pointer
340 * If the guest CPU is not waiting for interrupts or an interrupt line is
341 * asserted, the CPU is by definition runnable.
343 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
345 return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
348 /* Just ensure a guest exit from a particular CPU */
349 static void exit_vm_noop(void *info)
353 void force_vm_exit(const cpumask_t *mask)
355 smp_call_function_many(mask, exit_vm_noop, NULL, true);
359 * need_new_vmid_gen - check that the VMID is still valid
360 * @kvm: The VM's VMID to checkt
362 * return true if there is a new generation of VMIDs being used
364 * The hardware supports only 256 values with the value zero reserved for the
365 * host, so we check if an assigned value belongs to a previous generation,
366 * which which requires us to assign a new value. If we're the first to use a
367 * VMID for the new generation, we must flush necessary caches and TLBs on all
370 static bool need_new_vmid_gen(struct kvm *kvm)
372 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
376 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
377 * @kvm The guest that we are about to run
379 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
380 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
383 static void update_vttbr(struct kvm *kvm)
385 phys_addr_t pgd_phys;
388 if (!need_new_vmid_gen(kvm))
391 spin_lock(&kvm_vmid_lock);
394 * We need to re-check the vmid_gen here to ensure that if another vcpu
395 * already allocated a valid vmid for this vm, then this vcpu should
398 if (!need_new_vmid_gen(kvm)) {
399 spin_unlock(&kvm_vmid_lock);
403 /* First user of a new VMID generation? */
404 if (unlikely(kvm_next_vmid == 0)) {
405 atomic64_inc(&kvm_vmid_gen);
409 * On SMP we know no other CPUs can use this CPU's or each
410 * other's VMID after force_vm_exit returns since the
411 * kvm_vmid_lock blocks them from reentry to the guest.
413 force_vm_exit(cpu_all_mask);
415 * Now broadcast TLB + ICACHE invalidation over the inner
416 * shareable domain to make sure all data structures are
419 kvm_call_hyp(__kvm_flush_vm_context);
422 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
423 kvm->arch.vmid = kvm_next_vmid;
426 /* update vttbr to be used with the new vmid */
427 pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
428 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
429 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
430 kvm->arch.vttbr = pgd_phys | vmid;
432 spin_unlock(&kvm_vmid_lock);
435 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
437 struct kvm *kvm = vcpu->kvm;
440 if (likely(vcpu->arch.has_run_once))
443 vcpu->arch.has_run_once = true;
446 * Map the VGIC hardware resources before running a vcpu the first
449 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
450 ret = kvm_vgic_map_resources(kvm);
456 * Enable the arch timers only if we have an in-kernel VGIC
457 * and it has been properly initialized, since we cannot handle
458 * interrupts from the virtual timer with a userspace gic.
460 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
461 kvm_timer_enable(kvm);
466 bool kvm_arch_intc_initialized(struct kvm *kvm)
468 return vgic_initialized(kvm);
471 static void vcpu_pause(struct kvm_vcpu *vcpu)
473 wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
475 wait_event_interruptible(*wq, !vcpu->arch.pause);
478 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
480 return vcpu->arch.target >= 0;
484 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
485 * @vcpu: The VCPU pointer
486 * @run: The kvm_run structure pointer used for userspace state exchange
488 * This function is called through the VCPU_RUN ioctl called from user space. It
489 * will execute VM code in a loop until the time slice for the process is used
490 * or some emulation is needed from user space in which case the function will
491 * return with return value 0 and with the kvm_run structure filled in with the
492 * required data for the requested emulation.
494 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
499 if (unlikely(!kvm_vcpu_initialized(vcpu)))
502 ret = kvm_vcpu_first_run_init(vcpu);
506 if (run->exit_reason == KVM_EXIT_MMIO) {
507 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
512 if (vcpu->sigset_active)
513 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
516 run->exit_reason = KVM_EXIT_UNKNOWN;
519 * Check conditions before entering the guest
523 update_vttbr(vcpu->kvm);
525 if (vcpu->arch.pause)
529 * Disarming the background timer must be done in a
530 * preemptible context, as this call may sleep.
532 kvm_timer_flush_hwstate(vcpu);
535 * Preparing the interrupts to be injected also
536 * involves poking the GIC, which must be done in a
537 * non-preemptible context.
540 kvm_vgic_flush_hwstate(vcpu);
545 * Re-check atomic conditions
547 if (signal_pending(current)) {
549 run->exit_reason = KVM_EXIT_INTR;
552 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
554 kvm_vgic_sync_hwstate(vcpu);
556 kvm_timer_sync_hwstate(vcpu);
560 kvm_arm_setup_debug(vcpu);
562 /**************************************************************
565 trace_kvm_entry(*vcpu_pc(vcpu));
567 vcpu->mode = IN_GUEST_MODE;
569 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
571 vcpu->mode = OUTSIDE_GUEST_MODE;
574 *************************************************************/
576 kvm_arm_clear_debug(vcpu);
579 * We may have taken a host interrupt in HYP mode (ie
580 * while executing the guest). This interrupt is still
581 * pending, as we haven't serviced it yet!
583 * We're now back in SVC mode, with interrupts
584 * disabled. Enabling the interrupts now will have
585 * the effect of taking the interrupt again, in SVC
591 * We do local_irq_enable() before calling kvm_guest_exit() so
592 * that if a timer interrupt hits while running the guest we
593 * account that tick as being spent in the guest. We enable
594 * preemption after calling kvm_guest_exit() so that if we get
595 * preempted we make sure ticks after that is not counted as
599 trace_kvm_exit(kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
601 kvm_vgic_sync_hwstate(vcpu);
605 kvm_timer_sync_hwstate(vcpu);
607 ret = handle_exit(vcpu, run, ret);
610 if (vcpu->sigset_active)
611 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
615 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
621 if (number == KVM_ARM_IRQ_CPU_IRQ)
622 bit_index = __ffs(HCR_VI);
623 else /* KVM_ARM_IRQ_CPU_FIQ */
624 bit_index = __ffs(HCR_VF);
626 ptr = (unsigned long *)&vcpu->arch.irq_lines;
628 set = test_and_set_bit(bit_index, ptr);
630 set = test_and_clear_bit(bit_index, ptr);
633 * If we didn't change anything, no need to wake up or kick other CPUs
639 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
640 * trigger a world-switch round on the running physical CPU to set the
641 * virtual IRQ/FIQ fields in the HCR appropriately.
648 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
651 u32 irq = irq_level->irq;
652 unsigned int irq_type, vcpu_idx, irq_num;
653 int nrcpus = atomic_read(&kvm->online_vcpus);
654 struct kvm_vcpu *vcpu = NULL;
655 bool level = irq_level->level;
657 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
658 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
659 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
661 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
664 case KVM_ARM_IRQ_TYPE_CPU:
665 if (irqchip_in_kernel(kvm))
668 if (vcpu_idx >= nrcpus)
671 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
675 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
678 return vcpu_interrupt_line(vcpu, irq_num, level);
679 case KVM_ARM_IRQ_TYPE_PPI:
680 if (!irqchip_in_kernel(kvm))
683 if (vcpu_idx >= nrcpus)
686 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
690 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
693 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
694 case KVM_ARM_IRQ_TYPE_SPI:
695 if (!irqchip_in_kernel(kvm))
698 if (irq_num < VGIC_NR_PRIVATE_IRQS)
701 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
707 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
708 const struct kvm_vcpu_init *init)
711 int phys_target = kvm_target_cpu();
713 if (init->target != phys_target)
717 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
718 * use the same target.
720 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
723 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
724 for (i = 0; i < sizeof(init->features) * 8; i++) {
725 bool set = (init->features[i / 32] & (1 << (i % 32)));
727 if (set && i >= KVM_VCPU_MAX_FEATURES)
731 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
732 * use the same feature set.
734 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
735 test_bit(i, vcpu->arch.features) != set)
739 set_bit(i, vcpu->arch.features);
742 vcpu->arch.target = phys_target;
744 /* Now we know what it is, we can reset it. */
745 return kvm_reset_vcpu(vcpu);
749 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
750 struct kvm_vcpu_init *init)
754 ret = kvm_vcpu_set_target(vcpu, init);
759 * Ensure a rebooted VM will fault in RAM pages and detect if the
760 * guest MMU is turned off and flush the caches as needed.
762 if (vcpu->arch.has_run_once)
763 stage2_unmap_vm(vcpu->kvm);
765 vcpu_reset_hcr(vcpu);
768 * Handle the "start in power-off" case by marking the VCPU as paused.
770 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
771 vcpu->arch.pause = true;
773 vcpu->arch.pause = false;
778 long kvm_arch_vcpu_ioctl(struct file *filp,
779 unsigned int ioctl, unsigned long arg)
781 struct kvm_vcpu *vcpu = filp->private_data;
782 void __user *argp = (void __user *)arg;
785 case KVM_ARM_VCPU_INIT: {
786 struct kvm_vcpu_init init;
788 if (copy_from_user(&init, argp, sizeof(init)))
791 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
793 case KVM_SET_ONE_REG:
794 case KVM_GET_ONE_REG: {
795 struct kvm_one_reg reg;
797 if (unlikely(!kvm_vcpu_initialized(vcpu)))
800 if (copy_from_user(®, argp, sizeof(reg)))
802 if (ioctl == KVM_SET_ONE_REG)
803 return kvm_arm_set_reg(vcpu, ®);
805 return kvm_arm_get_reg(vcpu, ®);
807 case KVM_GET_REG_LIST: {
808 struct kvm_reg_list __user *user_list = argp;
809 struct kvm_reg_list reg_list;
812 if (unlikely(!kvm_vcpu_initialized(vcpu)))
815 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
818 reg_list.n = kvm_arm_num_regs(vcpu);
819 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
823 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
831 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
833 * @log: slot id and address to which we copy the log
835 * Steps 1-4 below provide general overview of dirty page logging. See
836 * kvm_get_dirty_log_protect() function description for additional details.
838 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
839 * always flush the TLB (step 4) even if previous step failed and the dirty
840 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
841 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
842 * writes will be marked dirty for next log read.
844 * 1. Take a snapshot of the bit and clear it if needed.
845 * 2. Write protect the corresponding page.
846 * 3. Copy the snapshot to the userspace.
847 * 4. Flush TLB's if needed.
849 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
851 bool is_dirty = false;
854 mutex_lock(&kvm->slots_lock);
856 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
859 kvm_flush_remote_tlbs(kvm);
861 mutex_unlock(&kvm->slots_lock);
865 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
866 struct kvm_arm_device_addr *dev_addr)
868 unsigned long dev_id, type;
870 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
871 KVM_ARM_DEVICE_ID_SHIFT;
872 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
873 KVM_ARM_DEVICE_TYPE_SHIFT;
876 case KVM_ARM_DEVICE_VGIC_V2:
877 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
883 long kvm_arch_vm_ioctl(struct file *filp,
884 unsigned int ioctl, unsigned long arg)
886 struct kvm *kvm = filp->private_data;
887 void __user *argp = (void __user *)arg;
890 case KVM_CREATE_IRQCHIP: {
891 return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
893 case KVM_ARM_SET_DEVICE_ADDR: {
894 struct kvm_arm_device_addr dev_addr;
896 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
898 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
900 case KVM_ARM_PREFERRED_TARGET: {
902 struct kvm_vcpu_init init;
904 err = kvm_vcpu_preferred_target(&init);
908 if (copy_to_user(argp, &init, sizeof(init)))
918 static void cpu_init_hyp_mode(void *dummy)
920 phys_addr_t boot_pgd_ptr;
922 unsigned long hyp_stack_ptr;
923 unsigned long stack_page;
924 unsigned long vector_ptr;
926 /* Switch from the HYP stub to our own HYP init vector */
927 __hyp_set_vectors(kvm_get_idmap_vector());
929 boot_pgd_ptr = kvm_mmu_get_boot_httbr();
930 pgd_ptr = kvm_mmu_get_httbr();
931 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
932 hyp_stack_ptr = stack_page + PAGE_SIZE;
933 vector_ptr = (unsigned long)__kvm_hyp_vector;
935 __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
937 kvm_arm_init_debug();
940 static int hyp_init_cpu_notify(struct notifier_block *self,
941 unsigned long action, void *cpu)
945 case CPU_STARTING_FROZEN:
946 if (__hyp_get_vectors() == hyp_default_vectors)
947 cpu_init_hyp_mode(NULL);
954 static struct notifier_block hyp_init_cpu_nb = {
955 .notifier_call = hyp_init_cpu_notify,
959 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
963 if (cmd == CPU_PM_EXIT &&
964 __hyp_get_vectors() == hyp_default_vectors) {
965 cpu_init_hyp_mode(NULL);
972 static struct notifier_block hyp_init_cpu_pm_nb = {
973 .notifier_call = hyp_init_cpu_pm_notifier,
976 static void __init hyp_cpu_pm_init(void)
978 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
981 static inline void hyp_cpu_pm_init(void)
987 * Inits Hyp-mode on all online CPUs
989 static int init_hyp_mode(void)
995 * Allocate Hyp PGD and setup Hyp identity mapping
997 err = kvm_mmu_init();
1002 * It is probably enough to obtain the default on one
1003 * CPU. It's unlikely to be different on the others.
1005 hyp_default_vectors = __hyp_get_vectors();
1008 * Allocate stack pages for Hypervisor-mode
1010 for_each_possible_cpu(cpu) {
1011 unsigned long stack_page;
1013 stack_page = __get_free_page(GFP_KERNEL);
1016 goto out_free_stack_pages;
1019 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1023 * Map the Hyp-code called directly from the host
1025 err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
1027 kvm_err("Cannot map world-switch code\n");
1028 goto out_free_mappings;
1032 * Map the Hyp stack pages
1034 for_each_possible_cpu(cpu) {
1035 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1036 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
1039 kvm_err("Cannot map hyp stack\n");
1040 goto out_free_mappings;
1045 * Map the host CPU structures
1047 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1048 if (!kvm_host_cpu_state) {
1050 kvm_err("Cannot allocate host CPU state\n");
1051 goto out_free_mappings;
1054 for_each_possible_cpu(cpu) {
1055 kvm_cpu_context_t *cpu_ctxt;
1057 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1058 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1061 kvm_err("Cannot map host CPU state: %d\n", err);
1062 goto out_free_context;
1067 * Execute the init code on each CPU.
1069 on_each_cpu(cpu_init_hyp_mode, NULL, 1);
1072 * Init HYP view of VGIC
1074 err = kvm_vgic_hyp_init();
1076 goto out_free_context;
1079 * Init HYP architected timer support
1081 err = kvm_timer_hyp_init();
1083 goto out_free_mappings;
1085 #ifndef CONFIG_HOTPLUG_CPU
1086 free_boot_hyp_pgd();
1091 kvm_info("Hyp mode initialized successfully\n");
1095 free_percpu(kvm_host_cpu_state);
1098 out_free_stack_pages:
1099 for_each_possible_cpu(cpu)
1100 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1102 kvm_err("error initializing Hyp mode: %d\n", err);
1106 static void check_kvm_target_cpu(void *ret)
1108 *(int *)ret = kvm_target_cpu();
1111 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1113 struct kvm_vcpu *vcpu;
1116 mpidr &= MPIDR_HWID_BITMASK;
1117 kvm_for_each_vcpu(i, vcpu, kvm) {
1118 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1125 * Initialize Hyp-mode and memory mappings on all CPUs.
1127 int kvm_arch_init(void *opaque)
1132 if (!is_hyp_mode_available()) {
1133 kvm_err("HYP mode not available\n");
1137 for_each_online_cpu(cpu) {
1138 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1140 kvm_err("Error, CPU %d not supported!\n", cpu);
1145 cpu_notifier_register_begin();
1147 err = init_hyp_mode();
1151 err = __register_cpu_notifier(&hyp_init_cpu_nb);
1153 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1157 cpu_notifier_register_done();
1161 kvm_coproc_table_init();
1164 cpu_notifier_register_done();
1168 /* NOP: Compiling as a module not supported */
1169 void kvm_arch_exit(void)
1171 kvm_perf_teardown();
1174 static int arm_init(void)
1176 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1180 module_init(arm_init);