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 <linux/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 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
123 if (!kvm->arch.last_vcpu_ran)
126 for_each_possible_cpu(cpu)
127 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
129 ret = kvm_alloc_stage2_pgd(kvm);
133 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
135 goto out_free_stage2_pgd;
137 kvm_vgic_early_init(kvm);
140 /* Mark the initial VMID generation invalid */
141 kvm->arch.vmid_gen = 0;
143 /* The maximum number of VCPUs is limited by the host's GIC model */
144 kvm->arch.max_vcpus = vgic_present ?
145 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
149 kvm_free_stage2_pgd(kvm);
151 free_percpu(kvm->arch.last_vcpu_ran);
152 kvm->arch.last_vcpu_ran = NULL;
156 bool kvm_arch_has_vcpu_debugfs(void)
161 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
166 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
168 return VM_FAULT_SIGBUS;
173 * kvm_arch_destroy_vm - destroy the VM data structure
174 * @kvm: pointer to the KVM struct
176 void kvm_arch_destroy_vm(struct kvm *kvm)
180 free_percpu(kvm->arch.last_vcpu_ran);
181 kvm->arch.last_vcpu_ran = NULL;
183 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
185 kvm_arch_vcpu_free(kvm->vcpus[i]);
186 kvm->vcpus[i] = NULL;
190 kvm_vgic_destroy(kvm);
193 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
197 case KVM_CAP_IRQCHIP:
200 case KVM_CAP_IOEVENTFD:
201 case KVM_CAP_DEVICE_CTRL:
202 case KVM_CAP_USER_MEMORY:
203 case KVM_CAP_SYNC_MMU:
204 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
205 case KVM_CAP_ONE_REG:
206 case KVM_CAP_ARM_PSCI:
207 case KVM_CAP_ARM_PSCI_0_2:
208 case KVM_CAP_READONLY_MEM:
209 case KVM_CAP_MP_STATE:
212 case KVM_CAP_COALESCED_MMIO:
213 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
215 case KVM_CAP_ARM_SET_DEVICE_ADDR:
218 case KVM_CAP_NR_VCPUS:
219 r = num_online_cpus();
221 case KVM_CAP_MAX_VCPUS:
224 case KVM_CAP_MSI_DEVID:
228 r = kvm->arch.vgic.msis_require_devid;
231 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
237 long kvm_arch_dev_ioctl(struct file *filp,
238 unsigned int ioctl, unsigned long arg)
244 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
247 struct kvm_vcpu *vcpu;
249 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
254 if (id >= kvm->arch.max_vcpus) {
259 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
265 err = kvm_vcpu_init(vcpu, kvm, id);
269 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
275 kvm_vcpu_uninit(vcpu);
277 kmem_cache_free(kvm_vcpu_cache, vcpu);
282 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
284 kvm_vgic_vcpu_early_init(vcpu);
287 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
289 kvm_mmu_free_memory_caches(vcpu);
290 kvm_timer_vcpu_terminate(vcpu);
291 kvm_vgic_vcpu_destroy(vcpu);
292 kvm_pmu_vcpu_destroy(vcpu);
293 kvm_vcpu_uninit(vcpu);
294 kmem_cache_free(kvm_vcpu_cache, vcpu);
297 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
299 kvm_arch_vcpu_free(vcpu);
302 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
304 return kvm_timer_should_fire(vcpu);
307 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
309 kvm_timer_schedule(vcpu);
312 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
314 kvm_timer_unschedule(vcpu);
317 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
319 /* Force users to call KVM_ARM_VCPU_INIT */
320 vcpu->arch.target = -1;
321 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
323 /* Set up the timer */
324 kvm_timer_vcpu_init(vcpu);
326 kvm_arm_reset_debug_ptr(vcpu);
331 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
335 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
338 * We might get preempted before the vCPU actually runs, but
339 * over-invalidation doesn't affect correctness.
341 if (*last_ran != vcpu->vcpu_id) {
342 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
343 *last_ran = vcpu->vcpu_id;
347 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
349 kvm_arm_set_running_vcpu(vcpu);
352 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
355 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
356 * if the vcpu is no longer assigned to a cpu. This is used for the
357 * optimized make_all_cpus_request path.
361 kvm_arm_set_running_vcpu(NULL);
362 kvm_timer_vcpu_put(vcpu);
365 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
366 struct kvm_mp_state *mp_state)
368 if (vcpu->arch.power_off)
369 mp_state->mp_state = KVM_MP_STATE_STOPPED;
371 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
376 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
377 struct kvm_mp_state *mp_state)
379 switch (mp_state->mp_state) {
380 case KVM_MP_STATE_RUNNABLE:
381 vcpu->arch.power_off = false;
383 case KVM_MP_STATE_STOPPED:
384 vcpu->arch.power_off = true;
394 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
395 * @v: The VCPU pointer
397 * If the guest CPU is not waiting for interrupts or an interrupt line is
398 * asserted, the CPU is by definition runnable.
400 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
402 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
403 && !v->arch.power_off && !v->arch.pause);
406 /* Just ensure a guest exit from a particular CPU */
407 static void exit_vm_noop(void *info)
411 void force_vm_exit(const cpumask_t *mask)
414 smp_call_function_many(mask, exit_vm_noop, NULL, true);
419 * need_new_vmid_gen - check that the VMID is still valid
420 * @kvm: The VM's VMID to check
422 * return true if there is a new generation of VMIDs being used
424 * The hardware supports only 256 values with the value zero reserved for the
425 * host, so we check if an assigned value belongs to a previous generation,
426 * which which requires us to assign a new value. If we're the first to use a
427 * VMID for the new generation, we must flush necessary caches and TLBs on all
430 static bool need_new_vmid_gen(struct kvm *kvm)
432 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
436 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
437 * @kvm The guest that we are about to run
439 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
440 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
443 static void update_vttbr(struct kvm *kvm)
445 phys_addr_t pgd_phys;
448 if (!need_new_vmid_gen(kvm))
451 spin_lock(&kvm_vmid_lock);
454 * We need to re-check the vmid_gen here to ensure that if another vcpu
455 * already allocated a valid vmid for this vm, then this vcpu should
458 if (!need_new_vmid_gen(kvm)) {
459 spin_unlock(&kvm_vmid_lock);
463 /* First user of a new VMID generation? */
464 if (unlikely(kvm_next_vmid == 0)) {
465 atomic64_inc(&kvm_vmid_gen);
469 * On SMP we know no other CPUs can use this CPU's or each
470 * other's VMID after force_vm_exit returns since the
471 * kvm_vmid_lock blocks them from reentry to the guest.
473 force_vm_exit(cpu_all_mask);
475 * Now broadcast TLB + ICACHE invalidation over the inner
476 * shareable domain to make sure all data structures are
479 kvm_call_hyp(__kvm_flush_vm_context);
482 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
483 kvm->arch.vmid = kvm_next_vmid;
485 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
487 /* update vttbr to be used with the new vmid */
488 pgd_phys = virt_to_phys(kvm->arch.pgd);
489 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
490 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
491 kvm->arch.vttbr = pgd_phys | vmid;
493 spin_unlock(&kvm_vmid_lock);
496 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
498 struct kvm *kvm = vcpu->kvm;
501 if (likely(vcpu->arch.has_run_once))
504 vcpu->arch.has_run_once = true;
507 * Map the VGIC hardware resources before running a vcpu the first
510 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
511 ret = kvm_vgic_map_resources(kvm);
517 * Enable the arch timers only if we have an in-kernel VGIC
518 * and it has been properly initialized, since we cannot handle
519 * interrupts from the virtual timer with a userspace gic.
521 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
522 ret = kvm_timer_enable(vcpu);
527 bool kvm_arch_intc_initialized(struct kvm *kvm)
529 return vgic_initialized(kvm);
532 void kvm_arm_halt_guest(struct kvm *kvm)
535 struct kvm_vcpu *vcpu;
537 kvm_for_each_vcpu(i, vcpu, kvm)
538 vcpu->arch.pause = true;
539 kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
542 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
544 vcpu->arch.pause = true;
548 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
550 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
552 vcpu->arch.pause = false;
556 void kvm_arm_resume_guest(struct kvm *kvm)
559 struct kvm_vcpu *vcpu;
561 kvm_for_each_vcpu(i, vcpu, kvm)
562 kvm_arm_resume_vcpu(vcpu);
565 static void vcpu_sleep(struct kvm_vcpu *vcpu)
567 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
569 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
570 (!vcpu->arch.pause)));
573 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
575 return vcpu->arch.target >= 0;
579 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
580 * @vcpu: The VCPU pointer
581 * @run: The kvm_run structure pointer used for userspace state exchange
583 * This function is called through the VCPU_RUN ioctl called from user space. It
584 * will execute VM code in a loop until the time slice for the process is used
585 * or some emulation is needed from user space in which case the function will
586 * return with return value 0 and with the kvm_run structure filled in with the
587 * required data for the requested emulation.
589 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
594 if (unlikely(!kvm_vcpu_initialized(vcpu)))
597 ret = kvm_vcpu_first_run_init(vcpu);
601 if (run->exit_reason == KVM_EXIT_MMIO) {
602 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
607 if (vcpu->sigset_active)
608 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
611 run->exit_reason = KVM_EXIT_UNKNOWN;
614 * Check conditions before entering the guest
618 update_vttbr(vcpu->kvm);
620 if (vcpu->arch.power_off || vcpu->arch.pause)
624 * Preparing the interrupts to be injected also
625 * involves poking the GIC, which must be done in a
626 * non-preemptible context.
629 kvm_pmu_flush_hwstate(vcpu);
630 kvm_timer_flush_hwstate(vcpu);
631 kvm_vgic_flush_hwstate(vcpu);
636 * Re-check atomic conditions
638 if (signal_pending(current)) {
640 run->exit_reason = KVM_EXIT_INTR;
643 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
644 vcpu->arch.power_off || vcpu->arch.pause) {
646 kvm_pmu_sync_hwstate(vcpu);
647 kvm_timer_sync_hwstate(vcpu);
648 kvm_vgic_sync_hwstate(vcpu);
653 kvm_arm_setup_debug(vcpu);
655 /**************************************************************
658 trace_kvm_entry(*vcpu_pc(vcpu));
659 guest_enter_irqoff();
660 vcpu->mode = IN_GUEST_MODE;
662 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
664 vcpu->mode = OUTSIDE_GUEST_MODE;
668 *************************************************************/
670 kvm_arm_clear_debug(vcpu);
673 * We may have taken a host interrupt in HYP mode (ie
674 * while executing the guest). This interrupt is still
675 * pending, as we haven't serviced it yet!
677 * We're now back in SVC mode, with interrupts
678 * disabled. Enabling the interrupts now will have
679 * the effect of taking the interrupt again, in SVC
685 * We do local_irq_enable() before calling guest_exit() so
686 * that if a timer interrupt hits while running the guest we
687 * account that tick as being spent in the guest. We enable
688 * preemption after calling guest_exit() so that if we get
689 * preempted we make sure ticks after that is not counted as
693 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
696 * We must sync the PMU and timer state before the vgic state so
697 * that the vgic can properly sample the updated state of the
700 kvm_pmu_sync_hwstate(vcpu);
701 kvm_timer_sync_hwstate(vcpu);
703 kvm_vgic_sync_hwstate(vcpu);
707 ret = handle_exit(vcpu, run, ret);
710 if (vcpu->sigset_active)
711 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
715 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
721 if (number == KVM_ARM_IRQ_CPU_IRQ)
722 bit_index = __ffs(HCR_VI);
723 else /* KVM_ARM_IRQ_CPU_FIQ */
724 bit_index = __ffs(HCR_VF);
726 ptr = (unsigned long *)&vcpu->arch.irq_lines;
728 set = test_and_set_bit(bit_index, ptr);
730 set = test_and_clear_bit(bit_index, ptr);
733 * If we didn't change anything, no need to wake up or kick other CPUs
739 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
740 * trigger a world-switch round on the running physical CPU to set the
741 * virtual IRQ/FIQ fields in the HCR appropriately.
748 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
751 u32 irq = irq_level->irq;
752 unsigned int irq_type, vcpu_idx, irq_num;
753 int nrcpus = atomic_read(&kvm->online_vcpus);
754 struct kvm_vcpu *vcpu = NULL;
755 bool level = irq_level->level;
757 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
758 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
759 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
761 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
764 case KVM_ARM_IRQ_TYPE_CPU:
765 if (irqchip_in_kernel(kvm))
768 if (vcpu_idx >= nrcpus)
771 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
775 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
778 return vcpu_interrupt_line(vcpu, irq_num, level);
779 case KVM_ARM_IRQ_TYPE_PPI:
780 if (!irqchip_in_kernel(kvm))
783 if (vcpu_idx >= nrcpus)
786 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
790 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
793 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
794 case KVM_ARM_IRQ_TYPE_SPI:
795 if (!irqchip_in_kernel(kvm))
798 if (irq_num < VGIC_NR_PRIVATE_IRQS)
801 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
807 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
808 const struct kvm_vcpu_init *init)
811 int phys_target = kvm_target_cpu();
813 if (init->target != phys_target)
817 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
818 * use the same target.
820 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
823 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
824 for (i = 0; i < sizeof(init->features) * 8; i++) {
825 bool set = (init->features[i / 32] & (1 << (i % 32)));
827 if (set && i >= KVM_VCPU_MAX_FEATURES)
831 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
832 * use the same feature set.
834 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
835 test_bit(i, vcpu->arch.features) != set)
839 set_bit(i, vcpu->arch.features);
842 vcpu->arch.target = phys_target;
844 /* Now we know what it is, we can reset it. */
845 return kvm_reset_vcpu(vcpu);
849 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
850 struct kvm_vcpu_init *init)
854 ret = kvm_vcpu_set_target(vcpu, init);
859 * Ensure a rebooted VM will fault in RAM pages and detect if the
860 * guest MMU is turned off and flush the caches as needed.
862 if (vcpu->arch.has_run_once)
863 stage2_unmap_vm(vcpu->kvm);
865 vcpu_reset_hcr(vcpu);
868 * Handle the "start in power-off" case.
870 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
871 vcpu->arch.power_off = true;
873 vcpu->arch.power_off = false;
878 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
879 struct kvm_device_attr *attr)
883 switch (attr->group) {
885 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
892 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
893 struct kvm_device_attr *attr)
897 switch (attr->group) {
899 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
906 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
907 struct kvm_device_attr *attr)
911 switch (attr->group) {
913 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
920 long kvm_arch_vcpu_ioctl(struct file *filp,
921 unsigned int ioctl, unsigned long arg)
923 struct kvm_vcpu *vcpu = filp->private_data;
924 void __user *argp = (void __user *)arg;
925 struct kvm_device_attr attr;
928 case KVM_ARM_VCPU_INIT: {
929 struct kvm_vcpu_init init;
931 if (copy_from_user(&init, argp, sizeof(init)))
934 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
936 case KVM_SET_ONE_REG:
937 case KVM_GET_ONE_REG: {
938 struct kvm_one_reg reg;
940 if (unlikely(!kvm_vcpu_initialized(vcpu)))
943 if (copy_from_user(®, argp, sizeof(reg)))
945 if (ioctl == KVM_SET_ONE_REG)
946 return kvm_arm_set_reg(vcpu, ®);
948 return kvm_arm_get_reg(vcpu, ®);
950 case KVM_GET_REG_LIST: {
951 struct kvm_reg_list __user *user_list = argp;
952 struct kvm_reg_list reg_list;
955 if (unlikely(!kvm_vcpu_initialized(vcpu)))
958 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
961 reg_list.n = kvm_arm_num_regs(vcpu);
962 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
966 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
968 case KVM_SET_DEVICE_ATTR: {
969 if (copy_from_user(&attr, argp, sizeof(attr)))
971 return kvm_arm_vcpu_set_attr(vcpu, &attr);
973 case KVM_GET_DEVICE_ATTR: {
974 if (copy_from_user(&attr, argp, sizeof(attr)))
976 return kvm_arm_vcpu_get_attr(vcpu, &attr);
978 case KVM_HAS_DEVICE_ATTR: {
979 if (copy_from_user(&attr, argp, sizeof(attr)))
981 return kvm_arm_vcpu_has_attr(vcpu, &attr);
989 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
991 * @log: slot id and address to which we copy the log
993 * Steps 1-4 below provide general overview of dirty page logging. See
994 * kvm_get_dirty_log_protect() function description for additional details.
996 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
997 * always flush the TLB (step 4) even if previous step failed and the dirty
998 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
999 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1000 * writes will be marked dirty for next log read.
1002 * 1. Take a snapshot of the bit and clear it if needed.
1003 * 2. Write protect the corresponding page.
1004 * 3. Copy the snapshot to the userspace.
1005 * 4. Flush TLB's if needed.
1007 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1009 bool is_dirty = false;
1012 mutex_lock(&kvm->slots_lock);
1014 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1017 kvm_flush_remote_tlbs(kvm);
1019 mutex_unlock(&kvm->slots_lock);
1023 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1024 struct kvm_arm_device_addr *dev_addr)
1026 unsigned long dev_id, type;
1028 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1029 KVM_ARM_DEVICE_ID_SHIFT;
1030 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1031 KVM_ARM_DEVICE_TYPE_SHIFT;
1034 case KVM_ARM_DEVICE_VGIC_V2:
1037 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1043 long kvm_arch_vm_ioctl(struct file *filp,
1044 unsigned int ioctl, unsigned long arg)
1046 struct kvm *kvm = filp->private_data;
1047 void __user *argp = (void __user *)arg;
1050 case KVM_CREATE_IRQCHIP: {
1054 mutex_lock(&kvm->lock);
1055 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1056 mutex_unlock(&kvm->lock);
1059 case KVM_ARM_SET_DEVICE_ADDR: {
1060 struct kvm_arm_device_addr dev_addr;
1062 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1064 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1066 case KVM_ARM_PREFERRED_TARGET: {
1068 struct kvm_vcpu_init init;
1070 err = kvm_vcpu_preferred_target(&init);
1074 if (copy_to_user(argp, &init, sizeof(init)))
1084 static void cpu_init_hyp_mode(void *dummy)
1086 phys_addr_t pgd_ptr;
1087 unsigned long hyp_stack_ptr;
1088 unsigned long stack_page;
1089 unsigned long vector_ptr;
1091 /* Switch from the HYP stub to our own HYP init vector */
1092 __hyp_set_vectors(kvm_get_idmap_vector());
1094 pgd_ptr = kvm_mmu_get_httbr();
1095 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1096 hyp_stack_ptr = stack_page + PAGE_SIZE;
1097 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1099 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1100 __cpu_init_stage2();
1102 kvm_arm_init_debug();
1105 static void cpu_hyp_reinit(void)
1107 if (is_kernel_in_hyp_mode()) {
1109 * __cpu_init_stage2() is safe to call even if the PM
1110 * event was cancelled before the CPU was reset.
1112 __cpu_init_stage2();
1114 if (__hyp_get_vectors() == hyp_default_vectors)
1115 cpu_init_hyp_mode(NULL);
1119 static void cpu_hyp_reset(void)
1121 if (!is_kernel_in_hyp_mode())
1122 __cpu_reset_hyp_mode(hyp_default_vectors,
1123 kvm_get_idmap_start());
1126 static void _kvm_arch_hardware_enable(void *discard)
1128 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1130 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1134 int kvm_arch_hardware_enable(void)
1136 _kvm_arch_hardware_enable(NULL);
1140 static void _kvm_arch_hardware_disable(void *discard)
1142 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1144 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1148 void kvm_arch_hardware_disable(void)
1150 _kvm_arch_hardware_disable(NULL);
1153 #ifdef CONFIG_CPU_PM
1154 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1159 * kvm_arm_hardware_enabled is left with its old value over
1160 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1165 if (__this_cpu_read(kvm_arm_hardware_enabled))
1167 * don't update kvm_arm_hardware_enabled here
1168 * so that the hardware will be re-enabled
1169 * when we resume. See below.
1175 if (__this_cpu_read(kvm_arm_hardware_enabled))
1176 /* The hardware was enabled before suspend. */
1186 static struct notifier_block hyp_init_cpu_pm_nb = {
1187 .notifier_call = hyp_init_cpu_pm_notifier,
1190 static void __init hyp_cpu_pm_init(void)
1192 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1194 static void __init hyp_cpu_pm_exit(void)
1196 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1199 static inline void hyp_cpu_pm_init(void)
1202 static inline void hyp_cpu_pm_exit(void)
1207 static void teardown_common_resources(void)
1209 free_percpu(kvm_host_cpu_state);
1212 static int init_common_resources(void)
1214 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1215 if (!kvm_host_cpu_state) {
1216 kvm_err("Cannot allocate host CPU state\n");
1220 /* set size of VMID supported by CPU */
1221 kvm_vmid_bits = kvm_get_vmid_bits();
1222 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1227 static int init_subsystems(void)
1232 * Enable hardware so that subsystem initialisation can access EL2.
1234 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1237 * Register CPU lower-power notifier
1242 * Init HYP view of VGIC
1244 err = kvm_vgic_hyp_init();
1247 vgic_present = true;
1251 vgic_present = false;
1259 * Init HYP architected timer support
1261 err = kvm_timer_hyp_init();
1266 kvm_coproc_table_init();
1269 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1274 static void teardown_hyp_mode(void)
1278 if (is_kernel_in_hyp_mode())
1282 for_each_possible_cpu(cpu)
1283 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1287 static int init_vhe_mode(void)
1289 kvm_info("VHE mode initialized successfully\n");
1294 * Inits Hyp-mode on all online CPUs
1296 static int init_hyp_mode(void)
1302 * Allocate Hyp PGD and setup Hyp identity mapping
1304 err = kvm_mmu_init();
1309 * It is probably enough to obtain the default on one
1310 * CPU. It's unlikely to be different on the others.
1312 hyp_default_vectors = __hyp_get_vectors();
1315 * Allocate stack pages for Hypervisor-mode
1317 for_each_possible_cpu(cpu) {
1318 unsigned long stack_page;
1320 stack_page = __get_free_page(GFP_KERNEL);
1326 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1330 * Map the Hyp-code called directly from the host
1332 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1333 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1335 kvm_err("Cannot map world-switch code\n");
1339 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1340 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1342 kvm_err("Cannot map rodata section\n");
1346 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1347 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1349 kvm_err("Cannot map bss section\n");
1354 * Map the Hyp stack pages
1356 for_each_possible_cpu(cpu) {
1357 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1358 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1362 kvm_err("Cannot map hyp stack\n");
1367 for_each_possible_cpu(cpu) {
1368 kvm_cpu_context_t *cpu_ctxt;
1370 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1371 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1374 kvm_err("Cannot map host CPU state: %d\n", err);
1379 kvm_info("Hyp mode initialized successfully\n");
1384 teardown_hyp_mode();
1385 kvm_err("error initializing Hyp mode: %d\n", err);
1389 static void check_kvm_target_cpu(void *ret)
1391 *(int *)ret = kvm_target_cpu();
1394 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1396 struct kvm_vcpu *vcpu;
1399 mpidr &= MPIDR_HWID_BITMASK;
1400 kvm_for_each_vcpu(i, vcpu, kvm) {
1401 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1408 * Initialize Hyp-mode and memory mappings on all CPUs.
1410 int kvm_arch_init(void *opaque)
1415 if (!is_hyp_mode_available()) {
1416 kvm_err("HYP mode not available\n");
1420 for_each_online_cpu(cpu) {
1421 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1423 kvm_err("Error, CPU %d not supported!\n", cpu);
1428 err = init_common_resources();
1432 if (is_kernel_in_hyp_mode())
1433 err = init_vhe_mode();
1435 err = init_hyp_mode();
1439 err = init_subsystems();
1446 teardown_hyp_mode();
1448 teardown_common_resources();
1452 /* NOP: Compiling as a module not supported */
1453 void kvm_arch_exit(void)
1455 kvm_perf_teardown();
1458 static int arm_init(void)
1460 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1464 module_init(arm_init);