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/module.h>
24 #include <linux/vmalloc.h>
26 #include <linux/mman.h>
27 #include <linux/sched.h>
28 #include <linux/kvm.h>
29 #include <trace/events/kvm.h>
30 #include <kvm/arm_pmu.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>
47 #include <asm/sections.h>
50 __asm__(".arch_extension virt");
53 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
54 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
55 static unsigned long hyp_default_vectors;
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 ret = kvm_alloc_stage2_pgd(kvm);
125 ret = create_hyp_mappings(kvm, kvm + 1);
127 goto out_free_stage2_pgd;
129 kvm_vgic_early_init(kvm);
132 /* Mark the initial VMID generation invalid */
133 kvm->arch.vmid_gen = 0;
135 /* The maximum number of VCPUs is limited by the host's GIC model */
136 kvm->arch.max_vcpus = vgic_present ?
137 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
141 kvm_free_stage2_pgd(kvm);
146 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
148 return VM_FAULT_SIGBUS;
153 * kvm_arch_destroy_vm - destroy the VM data structure
154 * @kvm: pointer to the KVM struct
156 void kvm_arch_destroy_vm(struct kvm *kvm)
160 kvm_free_stage2_pgd(kvm);
162 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
164 kvm_arch_vcpu_free(kvm->vcpus[i]);
165 kvm->vcpus[i] = NULL;
169 kvm_vgic_destroy(kvm);
172 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
176 case KVM_CAP_IRQCHIP:
179 case KVM_CAP_IOEVENTFD:
180 case KVM_CAP_DEVICE_CTRL:
181 case KVM_CAP_USER_MEMORY:
182 case KVM_CAP_SYNC_MMU:
183 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
184 case KVM_CAP_ONE_REG:
185 case KVM_CAP_ARM_PSCI:
186 case KVM_CAP_ARM_PSCI_0_2:
187 case KVM_CAP_READONLY_MEM:
188 case KVM_CAP_MP_STATE:
191 case KVM_CAP_COALESCED_MMIO:
192 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
194 case KVM_CAP_ARM_SET_DEVICE_ADDR:
197 case KVM_CAP_NR_VCPUS:
198 r = num_online_cpus();
200 case KVM_CAP_MAX_VCPUS:
204 r = kvm_arch_dev_ioctl_check_extension(ext);
210 long kvm_arch_dev_ioctl(struct file *filp,
211 unsigned int ioctl, unsigned long arg)
217 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
220 struct kvm_vcpu *vcpu;
222 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
227 if (id >= kvm->arch.max_vcpus) {
232 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
238 err = kvm_vcpu_init(vcpu, kvm, id);
242 err = create_hyp_mappings(vcpu, vcpu + 1);
248 kvm_vcpu_uninit(vcpu);
250 kmem_cache_free(kvm_vcpu_cache, vcpu);
255 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
257 kvm_vgic_vcpu_early_init(vcpu);
260 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
262 kvm_mmu_free_memory_caches(vcpu);
263 kvm_timer_vcpu_terminate(vcpu);
264 kvm_vgic_vcpu_destroy(vcpu);
265 kvm_pmu_vcpu_destroy(vcpu);
266 kvm_vcpu_uninit(vcpu);
267 kmem_cache_free(kvm_vcpu_cache, vcpu);
270 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
272 kvm_arch_vcpu_free(vcpu);
275 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
277 return kvm_timer_should_fire(vcpu);
280 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
282 kvm_timer_schedule(vcpu);
285 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
287 kvm_timer_unschedule(vcpu);
290 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
292 /* Force users to call KVM_ARM_VCPU_INIT */
293 vcpu->arch.target = -1;
294 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
296 /* Set up the timer */
297 kvm_timer_vcpu_init(vcpu);
299 kvm_arm_reset_debug_ptr(vcpu);
304 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
307 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
309 kvm_arm_set_running_vcpu(vcpu);
312 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
315 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
316 * if the vcpu is no longer assigned to a cpu. This is used for the
317 * optimized make_all_cpus_request path.
321 kvm_arm_set_running_vcpu(NULL);
322 kvm_timer_vcpu_put(vcpu);
325 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
326 struct kvm_mp_state *mp_state)
328 if (vcpu->arch.power_off)
329 mp_state->mp_state = KVM_MP_STATE_STOPPED;
331 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
336 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
337 struct kvm_mp_state *mp_state)
339 switch (mp_state->mp_state) {
340 case KVM_MP_STATE_RUNNABLE:
341 vcpu->arch.power_off = false;
343 case KVM_MP_STATE_STOPPED:
344 vcpu->arch.power_off = true;
354 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
355 * @v: The VCPU pointer
357 * If the guest CPU is not waiting for interrupts or an interrupt line is
358 * asserted, the CPU is by definition runnable.
360 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
362 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
363 && !v->arch.power_off && !v->arch.pause);
366 /* Just ensure a guest exit from a particular CPU */
367 static void exit_vm_noop(void *info)
371 void force_vm_exit(const cpumask_t *mask)
374 smp_call_function_many(mask, exit_vm_noop, NULL, true);
379 * need_new_vmid_gen - check that the VMID is still valid
380 * @kvm: The VM's VMID to checkt
382 * return true if there is a new generation of VMIDs being used
384 * The hardware supports only 256 values with the value zero reserved for the
385 * host, so we check if an assigned value belongs to a previous generation,
386 * which which requires us to assign a new value. If we're the first to use a
387 * VMID for the new generation, we must flush necessary caches and TLBs on all
390 static bool need_new_vmid_gen(struct kvm *kvm)
392 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
396 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
397 * @kvm The guest that we are about to run
399 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
400 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
403 static void update_vttbr(struct kvm *kvm)
405 phys_addr_t pgd_phys;
408 if (!need_new_vmid_gen(kvm))
411 spin_lock(&kvm_vmid_lock);
414 * We need to re-check the vmid_gen here to ensure that if another vcpu
415 * already allocated a valid vmid for this vm, then this vcpu should
418 if (!need_new_vmid_gen(kvm)) {
419 spin_unlock(&kvm_vmid_lock);
423 /* First user of a new VMID generation? */
424 if (unlikely(kvm_next_vmid == 0)) {
425 atomic64_inc(&kvm_vmid_gen);
429 * On SMP we know no other CPUs can use this CPU's or each
430 * other's VMID after force_vm_exit returns since the
431 * kvm_vmid_lock blocks them from reentry to the guest.
433 force_vm_exit(cpu_all_mask);
435 * Now broadcast TLB + ICACHE invalidation over the inner
436 * shareable domain to make sure all data structures are
439 kvm_call_hyp(__kvm_flush_vm_context);
442 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
443 kvm->arch.vmid = kvm_next_vmid;
445 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
447 /* update vttbr to be used with the new vmid */
448 pgd_phys = virt_to_phys(kvm->arch.pgd);
449 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
450 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
451 kvm->arch.vttbr = pgd_phys | vmid;
453 spin_unlock(&kvm_vmid_lock);
456 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
458 struct kvm *kvm = vcpu->kvm;
461 if (likely(vcpu->arch.has_run_once))
464 vcpu->arch.has_run_once = true;
467 * Map the VGIC hardware resources before running a vcpu the first
470 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
471 ret = kvm_vgic_map_resources(kvm);
477 * Enable the arch timers only if we have an in-kernel VGIC
478 * and it has been properly initialized, since we cannot handle
479 * interrupts from the virtual timer with a userspace gic.
481 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
482 ret = kvm_timer_enable(vcpu);
487 bool kvm_arch_intc_initialized(struct kvm *kvm)
489 return vgic_initialized(kvm);
492 void kvm_arm_halt_guest(struct kvm *kvm)
495 struct kvm_vcpu *vcpu;
497 kvm_for_each_vcpu(i, vcpu, kvm)
498 vcpu->arch.pause = true;
499 kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
502 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
504 vcpu->arch.pause = true;
508 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
510 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
512 vcpu->arch.pause = false;
516 void kvm_arm_resume_guest(struct kvm *kvm)
519 struct kvm_vcpu *vcpu;
521 kvm_for_each_vcpu(i, vcpu, kvm)
522 kvm_arm_resume_vcpu(vcpu);
525 static void vcpu_sleep(struct kvm_vcpu *vcpu)
527 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
529 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
530 (!vcpu->arch.pause)));
533 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
535 return vcpu->arch.target >= 0;
539 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
540 * @vcpu: The VCPU pointer
541 * @run: The kvm_run structure pointer used for userspace state exchange
543 * This function is called through the VCPU_RUN ioctl called from user space. It
544 * will execute VM code in a loop until the time slice for the process is used
545 * or some emulation is needed from user space in which case the function will
546 * return with return value 0 and with the kvm_run structure filled in with the
547 * required data for the requested emulation.
549 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
554 if (unlikely(!kvm_vcpu_initialized(vcpu)))
557 ret = kvm_vcpu_first_run_init(vcpu);
561 if (run->exit_reason == KVM_EXIT_MMIO) {
562 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
567 if (vcpu->sigset_active)
568 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
571 run->exit_reason = KVM_EXIT_UNKNOWN;
574 * Check conditions before entering the guest
578 update_vttbr(vcpu->kvm);
580 if (vcpu->arch.power_off || vcpu->arch.pause)
584 * Preparing the interrupts to be injected also
585 * involves poking the GIC, which must be done in a
586 * non-preemptible context.
589 kvm_pmu_flush_hwstate(vcpu);
590 kvm_timer_flush_hwstate(vcpu);
591 kvm_vgic_flush_hwstate(vcpu);
596 * Re-check atomic conditions
598 if (signal_pending(current)) {
600 run->exit_reason = KVM_EXIT_INTR;
603 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
604 vcpu->arch.power_off || vcpu->arch.pause) {
606 kvm_pmu_sync_hwstate(vcpu);
607 kvm_timer_sync_hwstate(vcpu);
608 kvm_vgic_sync_hwstate(vcpu);
613 kvm_arm_setup_debug(vcpu);
615 /**************************************************************
618 trace_kvm_entry(*vcpu_pc(vcpu));
620 vcpu->mode = IN_GUEST_MODE;
622 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
624 vcpu->mode = OUTSIDE_GUEST_MODE;
628 *************************************************************/
630 kvm_arm_clear_debug(vcpu);
633 * We may have taken a host interrupt in HYP mode (ie
634 * while executing the guest). This interrupt is still
635 * pending, as we haven't serviced it yet!
637 * We're now back in SVC mode, with interrupts
638 * disabled. Enabling the interrupts now will have
639 * the effect of taking the interrupt again, in SVC
645 * We do local_irq_enable() before calling kvm_guest_exit() so
646 * that if a timer interrupt hits while running the guest we
647 * account that tick as being spent in the guest. We enable
648 * preemption after calling kvm_guest_exit() so that if we get
649 * preempted we make sure ticks after that is not counted as
653 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
656 * We must sync the PMU and timer state before the vgic state so
657 * that the vgic can properly sample the updated state of the
660 kvm_pmu_sync_hwstate(vcpu);
661 kvm_timer_sync_hwstate(vcpu);
663 kvm_vgic_sync_hwstate(vcpu);
667 ret = handle_exit(vcpu, run, ret);
670 if (vcpu->sigset_active)
671 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
675 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
681 if (number == KVM_ARM_IRQ_CPU_IRQ)
682 bit_index = __ffs(HCR_VI);
683 else /* KVM_ARM_IRQ_CPU_FIQ */
684 bit_index = __ffs(HCR_VF);
686 ptr = (unsigned long *)&vcpu->arch.irq_lines;
688 set = test_and_set_bit(bit_index, ptr);
690 set = test_and_clear_bit(bit_index, ptr);
693 * If we didn't change anything, no need to wake up or kick other CPUs
699 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
700 * trigger a world-switch round on the running physical CPU to set the
701 * virtual IRQ/FIQ fields in the HCR appropriately.
708 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
711 u32 irq = irq_level->irq;
712 unsigned int irq_type, vcpu_idx, irq_num;
713 int nrcpus = atomic_read(&kvm->online_vcpus);
714 struct kvm_vcpu *vcpu = NULL;
715 bool level = irq_level->level;
717 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
718 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
719 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
721 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
724 case KVM_ARM_IRQ_TYPE_CPU:
725 if (irqchip_in_kernel(kvm))
728 if (vcpu_idx >= nrcpus)
731 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
735 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
738 return vcpu_interrupt_line(vcpu, irq_num, level);
739 case KVM_ARM_IRQ_TYPE_PPI:
740 if (!irqchip_in_kernel(kvm))
743 if (vcpu_idx >= nrcpus)
746 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
750 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
753 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
754 case KVM_ARM_IRQ_TYPE_SPI:
755 if (!irqchip_in_kernel(kvm))
758 if (irq_num < VGIC_NR_PRIVATE_IRQS)
761 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
767 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
768 const struct kvm_vcpu_init *init)
771 int phys_target = kvm_target_cpu();
773 if (init->target != phys_target)
777 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
778 * use the same target.
780 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
783 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
784 for (i = 0; i < sizeof(init->features) * 8; i++) {
785 bool set = (init->features[i / 32] & (1 << (i % 32)));
787 if (set && i >= KVM_VCPU_MAX_FEATURES)
791 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
792 * use the same feature set.
794 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
795 test_bit(i, vcpu->arch.features) != set)
799 set_bit(i, vcpu->arch.features);
802 vcpu->arch.target = phys_target;
804 /* Now we know what it is, we can reset it. */
805 return kvm_reset_vcpu(vcpu);
809 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
810 struct kvm_vcpu_init *init)
814 ret = kvm_vcpu_set_target(vcpu, init);
819 * Ensure a rebooted VM will fault in RAM pages and detect if the
820 * guest MMU is turned off and flush the caches as needed.
822 if (vcpu->arch.has_run_once)
823 stage2_unmap_vm(vcpu->kvm);
825 vcpu_reset_hcr(vcpu);
828 * Handle the "start in power-off" case.
830 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
831 vcpu->arch.power_off = true;
833 vcpu->arch.power_off = false;
838 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
839 struct kvm_device_attr *attr)
843 switch (attr->group) {
845 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
852 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
853 struct kvm_device_attr *attr)
857 switch (attr->group) {
859 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
866 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
867 struct kvm_device_attr *attr)
871 switch (attr->group) {
873 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
880 long kvm_arch_vcpu_ioctl(struct file *filp,
881 unsigned int ioctl, unsigned long arg)
883 struct kvm_vcpu *vcpu = filp->private_data;
884 void __user *argp = (void __user *)arg;
885 struct kvm_device_attr attr;
888 case KVM_ARM_VCPU_INIT: {
889 struct kvm_vcpu_init init;
891 if (copy_from_user(&init, argp, sizeof(init)))
894 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
896 case KVM_SET_ONE_REG:
897 case KVM_GET_ONE_REG: {
898 struct kvm_one_reg reg;
900 if (unlikely(!kvm_vcpu_initialized(vcpu)))
903 if (copy_from_user(®, argp, sizeof(reg)))
905 if (ioctl == KVM_SET_ONE_REG)
906 return kvm_arm_set_reg(vcpu, ®);
908 return kvm_arm_get_reg(vcpu, ®);
910 case KVM_GET_REG_LIST: {
911 struct kvm_reg_list __user *user_list = argp;
912 struct kvm_reg_list reg_list;
915 if (unlikely(!kvm_vcpu_initialized(vcpu)))
918 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
921 reg_list.n = kvm_arm_num_regs(vcpu);
922 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
926 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
928 case KVM_SET_DEVICE_ATTR: {
929 if (copy_from_user(&attr, argp, sizeof(attr)))
931 return kvm_arm_vcpu_set_attr(vcpu, &attr);
933 case KVM_GET_DEVICE_ATTR: {
934 if (copy_from_user(&attr, argp, sizeof(attr)))
936 return kvm_arm_vcpu_get_attr(vcpu, &attr);
938 case KVM_HAS_DEVICE_ATTR: {
939 if (copy_from_user(&attr, argp, sizeof(attr)))
941 return kvm_arm_vcpu_has_attr(vcpu, &attr);
949 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
951 * @log: slot id and address to which we copy the log
953 * Steps 1-4 below provide general overview of dirty page logging. See
954 * kvm_get_dirty_log_protect() function description for additional details.
956 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
957 * always flush the TLB (step 4) even if previous step failed and the dirty
958 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
959 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
960 * writes will be marked dirty for next log read.
962 * 1. Take a snapshot of the bit and clear it if needed.
963 * 2. Write protect the corresponding page.
964 * 3. Copy the snapshot to the userspace.
965 * 4. Flush TLB's if needed.
967 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
969 bool is_dirty = false;
972 mutex_lock(&kvm->slots_lock);
974 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
977 kvm_flush_remote_tlbs(kvm);
979 mutex_unlock(&kvm->slots_lock);
983 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
984 struct kvm_arm_device_addr *dev_addr)
986 unsigned long dev_id, type;
988 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
989 KVM_ARM_DEVICE_ID_SHIFT;
990 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
991 KVM_ARM_DEVICE_TYPE_SHIFT;
994 case KVM_ARM_DEVICE_VGIC_V2:
997 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1003 long kvm_arch_vm_ioctl(struct file *filp,
1004 unsigned int ioctl, unsigned long arg)
1006 struct kvm *kvm = filp->private_data;
1007 void __user *argp = (void __user *)arg;
1010 case KVM_CREATE_IRQCHIP: {
1013 return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1015 case KVM_ARM_SET_DEVICE_ADDR: {
1016 struct kvm_arm_device_addr dev_addr;
1018 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1020 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1022 case KVM_ARM_PREFERRED_TARGET: {
1024 struct kvm_vcpu_init init;
1026 err = kvm_vcpu_preferred_target(&init);
1030 if (copy_to_user(argp, &init, sizeof(init)))
1040 static void cpu_init_hyp_mode(void *dummy)
1042 phys_addr_t boot_pgd_ptr;
1043 phys_addr_t pgd_ptr;
1044 unsigned long hyp_stack_ptr;
1045 unsigned long stack_page;
1046 unsigned long vector_ptr;
1048 /* Switch from the HYP stub to our own HYP init vector */
1049 __hyp_set_vectors(kvm_get_idmap_vector());
1051 boot_pgd_ptr = kvm_mmu_get_boot_httbr();
1052 pgd_ptr = kvm_mmu_get_httbr();
1053 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1054 hyp_stack_ptr = stack_page + PAGE_SIZE;
1055 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1057 __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
1058 __cpu_init_stage2();
1060 kvm_arm_init_debug();
1063 static void cpu_hyp_reinit(void)
1065 if (is_kernel_in_hyp_mode()) {
1067 * __cpu_init_stage2() is safe to call even if the PM
1068 * event was cancelled before the CPU was reset.
1070 __cpu_init_stage2();
1072 if (__hyp_get_vectors() == hyp_default_vectors)
1073 cpu_init_hyp_mode(NULL);
1077 static void cpu_hyp_reset(void)
1079 phys_addr_t boot_pgd_ptr;
1080 phys_addr_t phys_idmap_start;
1082 if (!is_kernel_in_hyp_mode()) {
1083 boot_pgd_ptr = kvm_mmu_get_boot_httbr();
1084 phys_idmap_start = kvm_get_idmap_start();
1086 __cpu_reset_hyp_mode(boot_pgd_ptr, phys_idmap_start);
1090 static void _kvm_arch_hardware_enable(void *discard)
1092 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1094 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1098 int kvm_arch_hardware_enable(void)
1100 _kvm_arch_hardware_enable(NULL);
1104 static void _kvm_arch_hardware_disable(void *discard)
1106 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1108 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1112 void kvm_arch_hardware_disable(void)
1114 _kvm_arch_hardware_disable(NULL);
1117 #ifdef CONFIG_CPU_PM
1118 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1123 * kvm_arm_hardware_enabled is left with its old value over
1124 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1129 if (__this_cpu_read(kvm_arm_hardware_enabled))
1131 * don't update kvm_arm_hardware_enabled here
1132 * so that the hardware will be re-enabled
1133 * when we resume. See below.
1139 if (__this_cpu_read(kvm_arm_hardware_enabled))
1140 /* The hardware was enabled before suspend. */
1150 static struct notifier_block hyp_init_cpu_pm_nb = {
1151 .notifier_call = hyp_init_cpu_pm_notifier,
1154 static void __init hyp_cpu_pm_init(void)
1156 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1158 static void __init hyp_cpu_pm_exit(void)
1160 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1163 static inline void hyp_cpu_pm_init(void)
1166 static inline void hyp_cpu_pm_exit(void)
1171 static void teardown_common_resources(void)
1173 free_percpu(kvm_host_cpu_state);
1176 static int init_common_resources(void)
1178 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1179 if (!kvm_host_cpu_state) {
1180 kvm_err("Cannot allocate host CPU state\n");
1187 static int init_subsystems(void)
1192 * Enable hardware so that subsystem initialisation can access EL2.
1194 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1197 * Register CPU lower-power notifier
1202 * Init HYP view of VGIC
1204 err = kvm_vgic_hyp_init();
1207 vgic_present = true;
1211 vgic_present = false;
1219 * Init HYP architected timer support
1221 err = kvm_timer_hyp_init();
1226 kvm_coproc_table_init();
1229 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1234 static void teardown_hyp_mode(void)
1238 if (is_kernel_in_hyp_mode())
1242 for_each_possible_cpu(cpu)
1243 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1247 static int init_vhe_mode(void)
1249 /* set size of VMID supported by CPU */
1250 kvm_vmid_bits = kvm_get_vmid_bits();
1251 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1253 kvm_info("VHE mode initialized successfully\n");
1258 * Inits Hyp-mode on all online CPUs
1260 static int init_hyp_mode(void)
1266 * Allocate Hyp PGD and setup Hyp identity mapping
1268 err = kvm_mmu_init();
1273 * It is probably enough to obtain the default on one
1274 * CPU. It's unlikely to be different on the others.
1276 hyp_default_vectors = __hyp_get_vectors();
1279 * Allocate stack pages for Hypervisor-mode
1281 for_each_possible_cpu(cpu) {
1282 unsigned long stack_page;
1284 stack_page = __get_free_page(GFP_KERNEL);
1290 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1294 * Map the Hyp-code called directly from the host
1296 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1297 kvm_ksym_ref(__hyp_text_end));
1299 kvm_err("Cannot map world-switch code\n");
1303 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1304 kvm_ksym_ref(__end_rodata));
1306 kvm_err("Cannot map rodata section\n");
1311 * Map the Hyp stack pages
1313 for_each_possible_cpu(cpu) {
1314 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1315 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
1318 kvm_err("Cannot map hyp stack\n");
1323 for_each_possible_cpu(cpu) {
1324 kvm_cpu_context_t *cpu_ctxt;
1326 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1327 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1330 kvm_err("Cannot map host CPU state: %d\n", err);
1335 #ifndef CONFIG_HOTPLUG_CPU
1336 free_boot_hyp_pgd();
1339 /* set size of VMID supported by CPU */
1340 kvm_vmid_bits = kvm_get_vmid_bits();
1341 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1343 kvm_info("Hyp mode initialized successfully\n");
1348 teardown_hyp_mode();
1349 kvm_err("error initializing Hyp mode: %d\n", err);
1353 static void check_kvm_target_cpu(void *ret)
1355 *(int *)ret = kvm_target_cpu();
1358 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1360 struct kvm_vcpu *vcpu;
1363 mpidr &= MPIDR_HWID_BITMASK;
1364 kvm_for_each_vcpu(i, vcpu, kvm) {
1365 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1372 * Initialize Hyp-mode and memory mappings on all CPUs.
1374 int kvm_arch_init(void *opaque)
1379 if (!is_hyp_mode_available()) {
1380 kvm_err("HYP mode not available\n");
1384 for_each_online_cpu(cpu) {
1385 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1387 kvm_err("Error, CPU %d not supported!\n", cpu);
1392 err = init_common_resources();
1396 if (is_kernel_in_hyp_mode())
1397 err = init_vhe_mode();
1399 err = init_hyp_mode();
1403 err = init_subsystems();
1410 teardown_hyp_mode();
1412 teardown_common_resources();
1416 /* NOP: Compiling as a module not supported */
1417 void kvm_arch_exit(void)
1419 kvm_perf_teardown();
1422 static int arm_init(void)
1424 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1428 module_init(arm_init);