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 <linux/kvm_irqfd.h>
31 #include <linux/irqbypass.h>
32 #include <trace/events/kvm.h>
33 #include <kvm/arm_pmu.h>
35 #define CREATE_TRACE_POINTS
38 #include <linux/uaccess.h>
39 #include <asm/ptrace.h>
41 #include <asm/tlbflush.h>
42 #include <asm/cacheflush.h>
44 #include <asm/kvm_arm.h>
45 #include <asm/kvm_asm.h>
46 #include <asm/kvm_mmu.h>
47 #include <asm/kvm_emulate.h>
48 #include <asm/kvm_coproc.h>
49 #include <asm/kvm_psci.h>
50 #include <asm/sections.h>
53 __asm__(".arch_extension virt");
56 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
57 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
59 /* Per-CPU variable containing the currently running vcpu. */
60 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
62 /* The VMID used in the VTTBR */
63 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
64 static u32 kvm_next_vmid;
65 static unsigned int kvm_vmid_bits __read_mostly;
66 static DEFINE_SPINLOCK(kvm_vmid_lock);
68 static bool vgic_present;
70 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
72 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
74 BUG_ON(preemptible());
75 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
79 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
80 * Must be called from non-preemptible context
82 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
84 BUG_ON(preemptible());
85 return __this_cpu_read(kvm_arm_running_vcpu);
89 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
91 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
93 return &kvm_arm_running_vcpu;
96 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
98 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
101 int kvm_arch_hardware_setup(void)
106 void kvm_arch_check_processor_compat(void *rtn)
113 * kvm_arch_init_vm - initializes a VM data structure
114 * @kvm: pointer to the KVM struct
116 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
123 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
124 if (!kvm->arch.last_vcpu_ran)
127 for_each_possible_cpu(cpu)
128 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
130 ret = kvm_alloc_stage2_pgd(kvm);
134 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
136 goto out_free_stage2_pgd;
138 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 kvm_vgic_destroy(kvm);
182 free_percpu(kvm->arch.last_vcpu_ran);
183 kvm->arch.last_vcpu_ran = NULL;
185 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
187 kvm_arch_vcpu_free(kvm->vcpus[i]);
188 kvm->vcpus[i] = NULL;
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:
210 case KVM_CAP_IMMEDIATE_EXIT:
213 case KVM_CAP_ARM_SET_DEVICE_ADDR:
216 case KVM_CAP_NR_VCPUS:
217 r = num_online_cpus();
219 case KVM_CAP_MAX_VCPUS:
222 case KVM_CAP_NR_MEMSLOTS:
223 r = KVM_USER_MEM_SLOTS;
225 case KVM_CAP_MSI_DEVID:
229 r = kvm->arch.vgic.msis_require_devid;
231 case KVM_CAP_ARM_USER_IRQ:
233 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
234 * (bump this number if adding more devices)
239 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
245 long kvm_arch_dev_ioctl(struct file *filp,
246 unsigned int ioctl, unsigned long arg)
252 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
255 struct kvm_vcpu *vcpu;
257 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
262 if (id >= kvm->arch.max_vcpus) {
267 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
273 err = kvm_vcpu_init(vcpu, kvm, id);
277 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
283 kvm_vcpu_uninit(vcpu);
285 kmem_cache_free(kvm_vcpu_cache, vcpu);
290 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
292 kvm_vgic_vcpu_early_init(vcpu);
295 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
297 kvm_mmu_free_memory_caches(vcpu);
298 kvm_timer_vcpu_terminate(vcpu);
299 kvm_vgic_vcpu_destroy(vcpu);
300 kvm_pmu_vcpu_destroy(vcpu);
301 kvm_vcpu_uninit(vcpu);
302 kmem_cache_free(kvm_vcpu_cache, vcpu);
305 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
307 kvm_arch_vcpu_free(vcpu);
310 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
312 return kvm_timer_is_pending(vcpu);
315 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
317 kvm_timer_schedule(vcpu);
318 kvm_vgic_v4_enable_doorbell(vcpu);
321 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
323 kvm_timer_unschedule(vcpu);
324 kvm_vgic_v4_disable_doorbell(vcpu);
327 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
329 /* Force users to call KVM_ARM_VCPU_INIT */
330 vcpu->arch.target = -1;
331 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
333 /* Set up the timer */
334 kvm_timer_vcpu_init(vcpu);
336 kvm_arm_reset_debug_ptr(vcpu);
338 return kvm_vgic_vcpu_init(vcpu);
341 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
345 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
348 * We might get preempted before the vCPU actually runs, but
349 * over-invalidation doesn't affect correctness.
351 if (*last_ran != vcpu->vcpu_id) {
352 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
353 *last_ran = vcpu->vcpu_id;
357 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
359 kvm_arm_set_running_vcpu(vcpu);
361 kvm_timer_vcpu_load(vcpu);
364 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
366 kvm_timer_vcpu_put(vcpu);
371 kvm_arm_set_running_vcpu(NULL);
374 static void vcpu_power_off(struct kvm_vcpu *vcpu)
376 vcpu->arch.power_off = true;
377 kvm_make_request(KVM_REQ_SLEEP, vcpu);
381 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
382 struct kvm_mp_state *mp_state)
384 if (vcpu->arch.power_off)
385 mp_state->mp_state = KVM_MP_STATE_STOPPED;
387 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
392 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
393 struct kvm_mp_state *mp_state)
395 switch (mp_state->mp_state) {
396 case KVM_MP_STATE_RUNNABLE:
397 vcpu->arch.power_off = false;
399 case KVM_MP_STATE_STOPPED:
400 vcpu_power_off(vcpu);
410 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
411 * @v: The VCPU pointer
413 * If the guest CPU is not waiting for interrupts or an interrupt line is
414 * asserted, the CPU is by definition runnable.
416 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
418 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
419 && !v->arch.power_off && !v->arch.pause);
422 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
424 return vcpu_mode_priv(vcpu);
427 /* Just ensure a guest exit from a particular CPU */
428 static void exit_vm_noop(void *info)
432 void force_vm_exit(const cpumask_t *mask)
435 smp_call_function_many(mask, exit_vm_noop, NULL, true);
440 * need_new_vmid_gen - check that the VMID is still valid
441 * @kvm: The VM's VMID to check
443 * return true if there is a new generation of VMIDs being used
445 * The hardware supports only 256 values with the value zero reserved for the
446 * host, so we check if an assigned value belongs to a previous generation,
447 * which which requires us to assign a new value. If we're the first to use a
448 * VMID for the new generation, we must flush necessary caches and TLBs on all
451 static bool need_new_vmid_gen(struct kvm *kvm)
453 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
457 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
458 * @kvm The guest that we are about to run
460 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
461 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
464 static void update_vttbr(struct kvm *kvm)
466 phys_addr_t pgd_phys;
469 if (!need_new_vmid_gen(kvm))
472 spin_lock(&kvm_vmid_lock);
475 * We need to re-check the vmid_gen here to ensure that if another vcpu
476 * already allocated a valid vmid for this vm, then this vcpu should
479 if (!need_new_vmid_gen(kvm)) {
480 spin_unlock(&kvm_vmid_lock);
484 /* First user of a new VMID generation? */
485 if (unlikely(kvm_next_vmid == 0)) {
486 atomic64_inc(&kvm_vmid_gen);
490 * On SMP we know no other CPUs can use this CPU's or each
491 * other's VMID after force_vm_exit returns since the
492 * kvm_vmid_lock blocks them from reentry to the guest.
494 force_vm_exit(cpu_all_mask);
496 * Now broadcast TLB + ICACHE invalidation over the inner
497 * shareable domain to make sure all data structures are
500 kvm_call_hyp(__kvm_flush_vm_context);
503 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
504 kvm->arch.vmid = kvm_next_vmid;
506 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
508 /* update vttbr to be used with the new vmid */
509 pgd_phys = virt_to_phys(kvm->arch.pgd);
510 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
511 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
512 kvm->arch.vttbr = pgd_phys | vmid;
514 spin_unlock(&kvm_vmid_lock);
517 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
519 struct kvm *kvm = vcpu->kvm;
522 if (likely(vcpu->arch.has_run_once))
525 vcpu->arch.has_run_once = true;
528 * Map the VGIC hardware resources before running a vcpu the first
531 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
532 ret = kvm_vgic_map_resources(kvm);
537 ret = kvm_timer_enable(vcpu);
541 ret = kvm_arm_pmu_v3_enable(vcpu);
546 bool kvm_arch_intc_initialized(struct kvm *kvm)
548 return vgic_initialized(kvm);
551 void kvm_arm_halt_guest(struct kvm *kvm)
554 struct kvm_vcpu *vcpu;
556 kvm_for_each_vcpu(i, vcpu, kvm)
557 vcpu->arch.pause = true;
558 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
561 void kvm_arm_resume_guest(struct kvm *kvm)
564 struct kvm_vcpu *vcpu;
566 kvm_for_each_vcpu(i, vcpu, kvm) {
567 vcpu->arch.pause = false;
568 swake_up(kvm_arch_vcpu_wq(vcpu));
572 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
574 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
576 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
577 (!vcpu->arch.pause)));
579 if (vcpu->arch.power_off || vcpu->arch.pause) {
580 /* Awaken to handle a signal, request we sleep again later. */
581 kvm_make_request(KVM_REQ_SLEEP, vcpu);
585 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
587 return vcpu->arch.target >= 0;
590 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
592 if (kvm_request_pending(vcpu)) {
593 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
594 vcpu_req_sleep(vcpu);
597 * Clear IRQ_PENDING requests that were made to guarantee
598 * that a VCPU sees new virtual interrupts.
600 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
605 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
606 * @vcpu: The VCPU pointer
607 * @run: The kvm_run structure pointer used for userspace state exchange
609 * This function is called through the VCPU_RUN ioctl called from user space. It
610 * will execute VM code in a loop until the time slice for the process is used
611 * or some emulation is needed from user space in which case the function will
612 * return with return value 0 and with the kvm_run structure filled in with the
613 * required data for the requested emulation.
615 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
619 if (unlikely(!kvm_vcpu_initialized(vcpu)))
622 ret = kvm_vcpu_first_run_init(vcpu);
626 if (run->exit_reason == KVM_EXIT_MMIO) {
627 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
632 if (run->immediate_exit)
635 kvm_sigset_activate(vcpu);
638 run->exit_reason = KVM_EXIT_UNKNOWN;
641 * Check conditions before entering the guest
645 update_vttbr(vcpu->kvm);
647 check_vcpu_requests(vcpu);
650 * Preparing the interrupts to be injected also
651 * involves poking the GIC, which must be done in a
652 * non-preemptible context.
656 /* Flush FP/SIMD state that can't survive guest entry/exit */
657 kvm_fpsimd_flush_cpu_state();
659 kvm_pmu_flush_hwstate(vcpu);
663 kvm_vgic_flush_hwstate(vcpu);
666 * If we have a singal pending, or need to notify a userspace
667 * irqchip about timer or PMU level changes, then we exit (and
668 * update the timer level state in kvm_timer_update_run
671 if (signal_pending(current) ||
672 kvm_timer_should_notify_user(vcpu) ||
673 kvm_pmu_should_notify_user(vcpu)) {
675 run->exit_reason = KVM_EXIT_INTR;
679 * Ensure we set mode to IN_GUEST_MODE after we disable
680 * interrupts and before the final VCPU requests check.
681 * See the comment in kvm_vcpu_exiting_guest_mode() and
682 * Documentation/virtual/kvm/vcpu-requests.rst
684 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
686 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
687 kvm_request_pending(vcpu)) {
688 vcpu->mode = OUTSIDE_GUEST_MODE;
689 kvm_pmu_sync_hwstate(vcpu);
690 kvm_timer_sync_hwstate(vcpu);
691 kvm_vgic_sync_hwstate(vcpu);
697 kvm_arm_setup_debug(vcpu);
699 /**************************************************************
702 trace_kvm_entry(*vcpu_pc(vcpu));
703 guest_enter_irqoff();
705 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
707 vcpu->mode = OUTSIDE_GUEST_MODE;
711 *************************************************************/
713 kvm_arm_clear_debug(vcpu);
716 * We must sync the PMU state before the vgic state so
717 * that the vgic can properly sample the updated state of the
720 kvm_pmu_sync_hwstate(vcpu);
723 * Sync the vgic state before syncing the timer state because
724 * the timer code needs to know if the virtual timer
725 * interrupts are active.
727 kvm_vgic_sync_hwstate(vcpu);
730 * Sync the timer hardware state before enabling interrupts as
731 * we don't want vtimer interrupts to race with syncing the
732 * timer virtual interrupt state.
734 kvm_timer_sync_hwstate(vcpu);
737 * We may have taken a host interrupt in HYP mode (ie
738 * while executing the guest). This interrupt is still
739 * pending, as we haven't serviced it yet!
741 * We're now back in SVC mode, with interrupts
742 * disabled. Enabling the interrupts now will have
743 * the effect of taking the interrupt again, in SVC
749 * We do local_irq_enable() before calling guest_exit() so
750 * that if a timer interrupt hits while running the guest we
751 * account that tick as being spent in the guest. We enable
752 * preemption after calling guest_exit() so that if we get
753 * preempted we make sure ticks after that is not counted as
757 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
761 ret = handle_exit(vcpu, run, ret);
764 /* Tell userspace about in-kernel device output levels */
765 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
766 kvm_timer_update_run(vcpu);
767 kvm_pmu_update_run(vcpu);
770 kvm_sigset_deactivate(vcpu);
775 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
781 if (number == KVM_ARM_IRQ_CPU_IRQ)
782 bit_index = __ffs(HCR_VI);
783 else /* KVM_ARM_IRQ_CPU_FIQ */
784 bit_index = __ffs(HCR_VF);
786 ptr = (unsigned long *)&vcpu->arch.irq_lines;
788 set = test_and_set_bit(bit_index, ptr);
790 set = test_and_clear_bit(bit_index, ptr);
793 * If we didn't change anything, no need to wake up or kick other CPUs
799 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
800 * trigger a world-switch round on the running physical CPU to set the
801 * virtual IRQ/FIQ fields in the HCR appropriately.
803 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
809 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
812 u32 irq = irq_level->irq;
813 unsigned int irq_type, vcpu_idx, irq_num;
814 int nrcpus = atomic_read(&kvm->online_vcpus);
815 struct kvm_vcpu *vcpu = NULL;
816 bool level = irq_level->level;
818 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
819 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
820 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
822 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
825 case KVM_ARM_IRQ_TYPE_CPU:
826 if (irqchip_in_kernel(kvm))
829 if (vcpu_idx >= nrcpus)
832 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
836 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
839 return vcpu_interrupt_line(vcpu, irq_num, level);
840 case KVM_ARM_IRQ_TYPE_PPI:
841 if (!irqchip_in_kernel(kvm))
844 if (vcpu_idx >= nrcpus)
847 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
851 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
854 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
855 case KVM_ARM_IRQ_TYPE_SPI:
856 if (!irqchip_in_kernel(kvm))
859 if (irq_num < VGIC_NR_PRIVATE_IRQS)
862 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
868 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
869 const struct kvm_vcpu_init *init)
872 int phys_target = kvm_target_cpu();
874 if (init->target != phys_target)
878 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
879 * use the same target.
881 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
884 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
885 for (i = 0; i < sizeof(init->features) * 8; i++) {
886 bool set = (init->features[i / 32] & (1 << (i % 32)));
888 if (set && i >= KVM_VCPU_MAX_FEATURES)
892 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
893 * use the same feature set.
895 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
896 test_bit(i, vcpu->arch.features) != set)
900 set_bit(i, vcpu->arch.features);
903 vcpu->arch.target = phys_target;
905 /* Now we know what it is, we can reset it. */
906 return kvm_reset_vcpu(vcpu);
910 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
911 struct kvm_vcpu_init *init)
915 ret = kvm_vcpu_set_target(vcpu, init);
920 * Ensure a rebooted VM will fault in RAM pages and detect if the
921 * guest MMU is turned off and flush the caches as needed.
923 if (vcpu->arch.has_run_once)
924 stage2_unmap_vm(vcpu->kvm);
926 vcpu_reset_hcr(vcpu);
929 * Handle the "start in power-off" case.
931 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
932 vcpu_power_off(vcpu);
934 vcpu->arch.power_off = false;
939 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
940 struct kvm_device_attr *attr)
944 switch (attr->group) {
946 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
953 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
954 struct kvm_device_attr *attr)
958 switch (attr->group) {
960 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
967 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
968 struct kvm_device_attr *attr)
972 switch (attr->group) {
974 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
981 long kvm_arch_vcpu_ioctl(struct file *filp,
982 unsigned int ioctl, unsigned long arg)
984 struct kvm_vcpu *vcpu = filp->private_data;
985 void __user *argp = (void __user *)arg;
986 struct kvm_device_attr attr;
989 case KVM_ARM_VCPU_INIT: {
990 struct kvm_vcpu_init init;
992 if (copy_from_user(&init, argp, sizeof(init)))
995 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
997 case KVM_SET_ONE_REG:
998 case KVM_GET_ONE_REG: {
999 struct kvm_one_reg reg;
1001 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1004 if (copy_from_user(®, argp, sizeof(reg)))
1006 if (ioctl == KVM_SET_ONE_REG)
1007 return kvm_arm_set_reg(vcpu, ®);
1009 return kvm_arm_get_reg(vcpu, ®);
1011 case KVM_GET_REG_LIST: {
1012 struct kvm_reg_list __user *user_list = argp;
1013 struct kvm_reg_list reg_list;
1016 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1019 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1022 reg_list.n = kvm_arm_num_regs(vcpu);
1023 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1027 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1029 case KVM_SET_DEVICE_ATTR: {
1030 if (copy_from_user(&attr, argp, sizeof(attr)))
1032 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1034 case KVM_GET_DEVICE_ATTR: {
1035 if (copy_from_user(&attr, argp, sizeof(attr)))
1037 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1039 case KVM_HAS_DEVICE_ATTR: {
1040 if (copy_from_user(&attr, argp, sizeof(attr)))
1042 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1050 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1051 * @kvm: kvm instance
1052 * @log: slot id and address to which we copy the log
1054 * Steps 1-4 below provide general overview of dirty page logging. See
1055 * kvm_get_dirty_log_protect() function description for additional details.
1057 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1058 * always flush the TLB (step 4) even if previous step failed and the dirty
1059 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1060 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1061 * writes will be marked dirty for next log read.
1063 * 1. Take a snapshot of the bit and clear it if needed.
1064 * 2. Write protect the corresponding page.
1065 * 3. Copy the snapshot to the userspace.
1066 * 4. Flush TLB's if needed.
1068 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1070 bool is_dirty = false;
1073 mutex_lock(&kvm->slots_lock);
1075 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1078 kvm_flush_remote_tlbs(kvm);
1080 mutex_unlock(&kvm->slots_lock);
1084 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1085 struct kvm_arm_device_addr *dev_addr)
1087 unsigned long dev_id, type;
1089 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1090 KVM_ARM_DEVICE_ID_SHIFT;
1091 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1092 KVM_ARM_DEVICE_TYPE_SHIFT;
1095 case KVM_ARM_DEVICE_VGIC_V2:
1098 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1104 long kvm_arch_vm_ioctl(struct file *filp,
1105 unsigned int ioctl, unsigned long arg)
1107 struct kvm *kvm = filp->private_data;
1108 void __user *argp = (void __user *)arg;
1111 case KVM_CREATE_IRQCHIP: {
1115 mutex_lock(&kvm->lock);
1116 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1117 mutex_unlock(&kvm->lock);
1120 case KVM_ARM_SET_DEVICE_ADDR: {
1121 struct kvm_arm_device_addr dev_addr;
1123 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1125 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1127 case KVM_ARM_PREFERRED_TARGET: {
1129 struct kvm_vcpu_init init;
1131 err = kvm_vcpu_preferred_target(&init);
1135 if (copy_to_user(argp, &init, sizeof(init)))
1145 static void cpu_init_hyp_mode(void *dummy)
1147 phys_addr_t pgd_ptr;
1148 unsigned long hyp_stack_ptr;
1149 unsigned long stack_page;
1150 unsigned long vector_ptr;
1152 /* Switch from the HYP stub to our own HYP init vector */
1153 __hyp_set_vectors(kvm_get_idmap_vector());
1155 pgd_ptr = kvm_mmu_get_httbr();
1156 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1157 hyp_stack_ptr = stack_page + PAGE_SIZE;
1158 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1160 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1161 __cpu_init_stage2();
1163 kvm_arm_init_debug();
1166 static void cpu_hyp_reset(void)
1168 if (!is_kernel_in_hyp_mode())
1169 __hyp_reset_vectors();
1172 static void cpu_hyp_reinit(void)
1176 if (is_kernel_in_hyp_mode()) {
1178 * __cpu_init_stage2() is safe to call even if the PM
1179 * event was cancelled before the CPU was reset.
1181 __cpu_init_stage2();
1182 kvm_timer_init_vhe();
1184 cpu_init_hyp_mode(NULL);
1188 kvm_vgic_init_cpu_hardware();
1191 static void _kvm_arch_hardware_enable(void *discard)
1193 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1195 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1199 int kvm_arch_hardware_enable(void)
1201 _kvm_arch_hardware_enable(NULL);
1205 static void _kvm_arch_hardware_disable(void *discard)
1207 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1209 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1213 void kvm_arch_hardware_disable(void)
1215 _kvm_arch_hardware_disable(NULL);
1218 #ifdef CONFIG_CPU_PM
1219 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1224 * kvm_arm_hardware_enabled is left with its old value over
1225 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1230 if (__this_cpu_read(kvm_arm_hardware_enabled))
1232 * don't update kvm_arm_hardware_enabled here
1233 * so that the hardware will be re-enabled
1234 * when we resume. See below.
1240 if (__this_cpu_read(kvm_arm_hardware_enabled))
1241 /* The hardware was enabled before suspend. */
1251 static struct notifier_block hyp_init_cpu_pm_nb = {
1252 .notifier_call = hyp_init_cpu_pm_notifier,
1255 static void __init hyp_cpu_pm_init(void)
1257 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1259 static void __init hyp_cpu_pm_exit(void)
1261 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1264 static inline void hyp_cpu_pm_init(void)
1267 static inline void hyp_cpu_pm_exit(void)
1272 static void teardown_common_resources(void)
1274 free_percpu(kvm_host_cpu_state);
1277 static int init_common_resources(void)
1279 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1280 if (!kvm_host_cpu_state) {
1281 kvm_err("Cannot allocate host CPU state\n");
1285 /* set size of VMID supported by CPU */
1286 kvm_vmid_bits = kvm_get_vmid_bits();
1287 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1292 static int init_subsystems(void)
1297 * Enable hardware so that subsystem initialisation can access EL2.
1299 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1302 * Register CPU lower-power notifier
1307 * Init HYP view of VGIC
1309 err = kvm_vgic_hyp_init();
1312 vgic_present = true;
1316 vgic_present = false;
1324 * Init HYP architected timer support
1326 err = kvm_timer_hyp_init();
1331 kvm_coproc_table_init();
1334 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1339 static void teardown_hyp_mode(void)
1344 for_each_possible_cpu(cpu)
1345 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1350 * Inits Hyp-mode on all online CPUs
1352 static int init_hyp_mode(void)
1358 * Allocate Hyp PGD and setup Hyp identity mapping
1360 err = kvm_mmu_init();
1365 * Allocate stack pages for Hypervisor-mode
1367 for_each_possible_cpu(cpu) {
1368 unsigned long stack_page;
1370 stack_page = __get_free_page(GFP_KERNEL);
1376 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1380 * Map the Hyp-code called directly from the host
1382 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1383 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1385 kvm_err("Cannot map world-switch code\n");
1389 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1390 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1392 kvm_err("Cannot map rodata section\n");
1396 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1397 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1399 kvm_err("Cannot map bss section\n");
1404 * Map the Hyp stack pages
1406 for_each_possible_cpu(cpu) {
1407 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1408 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1412 kvm_err("Cannot map hyp stack\n");
1417 for_each_possible_cpu(cpu) {
1418 kvm_cpu_context_t *cpu_ctxt;
1420 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1421 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1424 kvm_err("Cannot map host CPU state: %d\n", err);
1432 teardown_hyp_mode();
1433 kvm_err("error initializing Hyp mode: %d\n", err);
1437 static void check_kvm_target_cpu(void *ret)
1439 *(int *)ret = kvm_target_cpu();
1442 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1444 struct kvm_vcpu *vcpu;
1447 mpidr &= MPIDR_HWID_BITMASK;
1448 kvm_for_each_vcpu(i, vcpu, kvm) {
1449 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1455 bool kvm_arch_has_irq_bypass(void)
1460 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1461 struct irq_bypass_producer *prod)
1463 struct kvm_kernel_irqfd *irqfd =
1464 container_of(cons, struct kvm_kernel_irqfd, consumer);
1466 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1469 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1470 struct irq_bypass_producer *prod)
1472 struct kvm_kernel_irqfd *irqfd =
1473 container_of(cons, struct kvm_kernel_irqfd, consumer);
1475 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1479 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1481 struct kvm_kernel_irqfd *irqfd =
1482 container_of(cons, struct kvm_kernel_irqfd, consumer);
1484 kvm_arm_halt_guest(irqfd->kvm);
1487 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1489 struct kvm_kernel_irqfd *irqfd =
1490 container_of(cons, struct kvm_kernel_irqfd, consumer);
1492 kvm_arm_resume_guest(irqfd->kvm);
1496 * Initialize Hyp-mode and memory mappings on all CPUs.
1498 int kvm_arch_init(void *opaque)
1504 if (!is_hyp_mode_available()) {
1505 kvm_err("HYP mode not available\n");
1509 for_each_online_cpu(cpu) {
1510 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1512 kvm_err("Error, CPU %d not supported!\n", cpu);
1517 err = init_common_resources();
1521 in_hyp_mode = is_kernel_in_hyp_mode();
1524 err = init_hyp_mode();
1529 err = init_subsystems();
1534 kvm_info("VHE mode initialized successfully\n");
1536 kvm_info("Hyp mode initialized successfully\n");
1542 teardown_hyp_mode();
1544 teardown_common_resources();
1548 /* NOP: Compiling as a module not supported */
1549 void kvm_arch_exit(void)
1551 kvm_perf_teardown();
1554 static int arm_init(void)
1556 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1560 module_init(arm_init);