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/bug.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/list.h>
25 #include <linux/module.h>
26 #include <linux/vmalloc.h>
28 #include <linux/mman.h>
29 #include <linux/sched.h>
30 #include <linux/kvm.h>
31 #include <linux/kvm_irqfd.h>
32 #include <linux/irqbypass.h>
33 #include <trace/events/kvm.h>
34 #include <kvm/arm_pmu.h>
35 #include <kvm/arm_psci.h>
37 #define CREATE_TRACE_POINTS
40 #include <linux/uaccess.h>
41 #include <asm/ptrace.h>
43 #include <asm/tlbflush.h>
44 #include <asm/cacheflush.h>
45 #include <asm/cpufeature.h>
47 #include <asm/kvm_arm.h>
48 #include <asm/kvm_asm.h>
49 #include <asm/kvm_mmu.h>
50 #include <asm/kvm_emulate.h>
51 #include <asm/kvm_coproc.h>
52 #include <asm/sections.h>
55 __asm__(".arch_extension virt");
58 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
59 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
61 /* Per-CPU variable containing the currently running vcpu. */
62 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
64 /* The VMID used in the VTTBR */
65 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
66 static u32 kvm_next_vmid;
67 static unsigned int kvm_vmid_bits __read_mostly;
68 static DEFINE_RWLOCK(kvm_vmid_lock);
70 static bool vgic_present;
72 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
74 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
76 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
79 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
82 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
83 * Must be called from non-preemptible context
85 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
87 return __this_cpu_read(kvm_arm_running_vcpu);
91 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
93 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
95 return &kvm_arm_running_vcpu;
98 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
100 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
103 int kvm_arch_hardware_setup(void)
108 void kvm_arch_check_processor_compat(void *rtn)
115 * kvm_arch_init_vm - initializes a VM data structure
116 * @kvm: pointer to the KVM struct
118 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
125 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
126 if (!kvm->arch.last_vcpu_ran)
129 for_each_possible_cpu(cpu)
130 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
132 ret = kvm_alloc_stage2_pgd(kvm);
136 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
138 goto out_free_stage2_pgd;
140 kvm_vgic_early_init(kvm);
142 /* Mark the initial VMID generation invalid */
143 kvm->arch.vmid_gen = 0;
145 /* The maximum number of VCPUs is limited by the host's GIC model */
146 kvm->arch.max_vcpus = vgic_present ?
147 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
151 kvm_free_stage2_pgd(kvm);
153 free_percpu(kvm->arch.last_vcpu_ran);
154 kvm->arch.last_vcpu_ran = NULL;
158 bool kvm_arch_has_vcpu_debugfs(void)
163 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
168 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
170 return VM_FAULT_SIGBUS;
175 * kvm_arch_destroy_vm - destroy the VM data structure
176 * @kvm: pointer to the KVM struct
178 void kvm_arch_destroy_vm(struct kvm *kvm)
182 kvm_vgic_destroy(kvm);
184 free_percpu(kvm->arch.last_vcpu_ran);
185 kvm->arch.last_vcpu_ran = NULL;
187 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
189 kvm_arch_vcpu_free(kvm->vcpus[i]);
190 kvm->vcpus[i] = NULL;
193 atomic_set(&kvm->online_vcpus, 0);
196 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
200 case KVM_CAP_IRQCHIP:
203 case KVM_CAP_IOEVENTFD:
204 case KVM_CAP_DEVICE_CTRL:
205 case KVM_CAP_USER_MEMORY:
206 case KVM_CAP_SYNC_MMU:
207 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
208 case KVM_CAP_ONE_REG:
209 case KVM_CAP_ARM_PSCI:
210 case KVM_CAP_ARM_PSCI_0_2:
211 case KVM_CAP_READONLY_MEM:
212 case KVM_CAP_MP_STATE:
213 case KVM_CAP_IMMEDIATE_EXIT:
216 case KVM_CAP_ARM_SET_DEVICE_ADDR:
219 case KVM_CAP_NR_VCPUS:
220 r = num_online_cpus();
222 case KVM_CAP_MAX_VCPUS:
225 case KVM_CAP_NR_MEMSLOTS:
226 r = KVM_USER_MEM_SLOTS;
228 case KVM_CAP_MSI_DEVID:
232 r = kvm->arch.vgic.msis_require_devid;
234 case KVM_CAP_ARM_USER_IRQ:
236 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
237 * (bump this number if adding more devices)
242 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
248 long kvm_arch_dev_ioctl(struct file *filp,
249 unsigned int ioctl, unsigned long arg)
255 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
258 struct kvm_vcpu *vcpu;
260 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
265 if (id >= kvm->arch.max_vcpus) {
270 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
276 err = kvm_vcpu_init(vcpu, kvm, id);
280 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
286 kvm_vcpu_uninit(vcpu);
288 kmem_cache_free(kvm_vcpu_cache, vcpu);
293 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
297 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
299 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
300 static_branch_dec(&userspace_irqchip_in_use);
302 kvm_mmu_free_memory_caches(vcpu);
303 kvm_timer_vcpu_terminate(vcpu);
304 kvm_pmu_vcpu_destroy(vcpu);
305 kvm_vcpu_uninit(vcpu);
306 kmem_cache_free(kvm_vcpu_cache, vcpu);
309 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
311 kvm_arch_vcpu_free(vcpu);
314 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
316 return kvm_timer_is_pending(vcpu);
319 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
321 kvm_timer_schedule(vcpu);
322 kvm_vgic_v4_enable_doorbell(vcpu);
325 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
327 kvm_timer_unschedule(vcpu);
328 kvm_vgic_v4_disable_doorbell(vcpu);
331 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
333 /* Force users to call KVM_ARM_VCPU_INIT */
334 vcpu->arch.target = -1;
335 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
337 /* Set up the timer */
338 kvm_timer_vcpu_init(vcpu);
340 kvm_arm_reset_debug_ptr(vcpu);
342 return kvm_vgic_vcpu_init(vcpu);
345 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
349 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
352 * We might get preempted before the vCPU actually runs, but
353 * over-invalidation doesn't affect correctness.
355 if (*last_ran != vcpu->vcpu_id) {
356 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
357 *last_ran = vcpu->vcpu_id;
361 vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
363 kvm_arm_set_running_vcpu(vcpu);
365 kvm_timer_vcpu_load(vcpu);
366 kvm_vcpu_load_sysregs(vcpu);
367 kvm_arch_vcpu_load_fp(vcpu);
370 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
372 kvm_arch_vcpu_put_fp(vcpu);
373 kvm_vcpu_put_sysregs(vcpu);
374 kvm_timer_vcpu_put(vcpu);
379 kvm_arm_set_running_vcpu(NULL);
382 static void vcpu_power_off(struct kvm_vcpu *vcpu)
384 vcpu->arch.power_off = true;
385 kvm_make_request(KVM_REQ_SLEEP, vcpu);
389 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
390 struct kvm_mp_state *mp_state)
392 if (vcpu->arch.power_off)
393 mp_state->mp_state = KVM_MP_STATE_STOPPED;
395 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
400 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
401 struct kvm_mp_state *mp_state)
405 switch (mp_state->mp_state) {
406 case KVM_MP_STATE_RUNNABLE:
407 vcpu->arch.power_off = false;
409 case KVM_MP_STATE_STOPPED:
410 vcpu_power_off(vcpu);
420 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
421 * @v: The VCPU pointer
423 * If the guest CPU is not waiting for interrupts or an interrupt line is
424 * asserted, the CPU is by definition runnable.
426 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
428 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
429 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
430 && !v->arch.power_off && !v->arch.pause);
433 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
435 return vcpu_mode_priv(vcpu);
438 /* Just ensure a guest exit from a particular CPU */
439 static void exit_vm_noop(void *info)
443 void force_vm_exit(const cpumask_t *mask)
446 smp_call_function_many(mask, exit_vm_noop, NULL, true);
451 * need_new_vmid_gen - check that the VMID is still valid
452 * @kvm: The VM's VMID to check
454 * return true if there is a new generation of VMIDs being used
456 * The hardware supports only 256 values with the value zero reserved for the
457 * host, so we check if an assigned value belongs to a previous generation,
458 * which which requires us to assign a new value. If we're the first to use a
459 * VMID for the new generation, we must flush necessary caches and TLBs on all
462 static bool need_new_vmid_gen(struct kvm *kvm)
464 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
468 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
469 * @kvm The guest that we are about to run
471 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
472 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
475 static void update_vttbr(struct kvm *kvm)
477 phys_addr_t pgd_phys;
481 read_lock(&kvm_vmid_lock);
482 new_gen = need_new_vmid_gen(kvm);
483 read_unlock(&kvm_vmid_lock);
488 write_lock(&kvm_vmid_lock);
491 * We need to re-check the vmid_gen here to ensure that if another vcpu
492 * already allocated a valid vmid for this vm, then this vcpu should
495 if (!need_new_vmid_gen(kvm)) {
496 write_unlock(&kvm_vmid_lock);
500 /* First user of a new VMID generation? */
501 if (unlikely(kvm_next_vmid == 0)) {
502 atomic64_inc(&kvm_vmid_gen);
506 * On SMP we know no other CPUs can use this CPU's or each
507 * other's VMID after force_vm_exit returns since the
508 * kvm_vmid_lock blocks them from reentry to the guest.
510 force_vm_exit(cpu_all_mask);
512 * Now broadcast TLB + ICACHE invalidation over the inner
513 * shareable domain to make sure all data structures are
516 kvm_call_hyp(__kvm_flush_vm_context);
519 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
520 kvm->arch.vmid = kvm_next_vmid;
522 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
524 /* update vttbr to be used with the new vmid */
525 pgd_phys = virt_to_phys(kvm->arch.pgd);
526 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
527 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
528 kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid;
530 write_unlock(&kvm_vmid_lock);
533 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
535 struct kvm *kvm = vcpu->kvm;
538 if (likely(vcpu->arch.has_run_once))
541 vcpu->arch.has_run_once = true;
543 if (likely(irqchip_in_kernel(kvm))) {
545 * Map the VGIC hardware resources before running a vcpu the
546 * first time on this VM.
548 if (unlikely(!vgic_ready(kvm))) {
549 ret = kvm_vgic_map_resources(kvm);
555 * Tell the rest of the code that there are userspace irqchip
558 static_branch_inc(&userspace_irqchip_in_use);
561 ret = kvm_timer_enable(vcpu);
565 ret = kvm_arm_pmu_v3_enable(vcpu);
570 bool kvm_arch_intc_initialized(struct kvm *kvm)
572 return vgic_initialized(kvm);
575 void kvm_arm_halt_guest(struct kvm *kvm)
578 struct kvm_vcpu *vcpu;
580 kvm_for_each_vcpu(i, vcpu, kvm)
581 vcpu->arch.pause = true;
582 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
585 void kvm_arm_resume_guest(struct kvm *kvm)
588 struct kvm_vcpu *vcpu;
590 kvm_for_each_vcpu(i, vcpu, kvm) {
591 vcpu->arch.pause = false;
592 swake_up(kvm_arch_vcpu_wq(vcpu));
596 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
598 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
600 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
601 (!vcpu->arch.pause)));
603 if (vcpu->arch.power_off || vcpu->arch.pause) {
604 /* Awaken to handle a signal, request we sleep again later. */
605 kvm_make_request(KVM_REQ_SLEEP, vcpu);
609 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
611 return vcpu->arch.target >= 0;
614 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
616 if (kvm_request_pending(vcpu)) {
617 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
618 vcpu_req_sleep(vcpu);
621 * Clear IRQ_PENDING requests that were made to guarantee
622 * that a VCPU sees new virtual interrupts.
624 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
629 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
630 * @vcpu: The VCPU pointer
631 * @run: The kvm_run structure pointer used for userspace state exchange
633 * This function is called through the VCPU_RUN ioctl called from user space. It
634 * will execute VM code in a loop until the time slice for the process is used
635 * or some emulation is needed from user space in which case the function will
636 * return with return value 0 and with the kvm_run structure filled in with the
637 * required data for the requested emulation.
639 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
643 if (unlikely(!kvm_vcpu_initialized(vcpu)))
646 ret = kvm_vcpu_first_run_init(vcpu);
650 if (run->exit_reason == KVM_EXIT_MMIO) {
651 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
654 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
658 if (run->immediate_exit)
663 kvm_sigset_activate(vcpu);
666 run->exit_reason = KVM_EXIT_UNKNOWN;
669 * Check conditions before entering the guest
673 update_vttbr(vcpu->kvm);
675 check_vcpu_requests(vcpu);
678 * Preparing the interrupts to be injected also
679 * involves poking the GIC, which must be done in a
680 * non-preemptible context.
684 kvm_pmu_flush_hwstate(vcpu);
688 kvm_vgic_flush_hwstate(vcpu);
691 * Exit if we have a signal pending so that we can deliver the
692 * signal to user space.
694 if (signal_pending(current)) {
696 run->exit_reason = KVM_EXIT_INTR;
700 * If we're using a userspace irqchip, then check if we need
701 * to tell a userspace irqchip about timer or PMU level
702 * changes and if so, exit to userspace (the actual level
703 * state gets updated in kvm_timer_update_run and
704 * kvm_pmu_update_run below).
706 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
707 if (kvm_timer_should_notify_user(vcpu) ||
708 kvm_pmu_should_notify_user(vcpu)) {
710 run->exit_reason = KVM_EXIT_INTR;
715 * Ensure we set mode to IN_GUEST_MODE after we disable
716 * interrupts and before the final VCPU requests check.
717 * See the comment in kvm_vcpu_exiting_guest_mode() and
718 * Documentation/virtual/kvm/vcpu-requests.rst
720 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
722 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
723 kvm_request_pending(vcpu)) {
724 vcpu->mode = OUTSIDE_GUEST_MODE;
725 isb(); /* Ensure work in x_flush_hwstate is committed */
726 kvm_pmu_sync_hwstate(vcpu);
727 if (static_branch_unlikely(&userspace_irqchip_in_use))
728 kvm_timer_sync_hwstate(vcpu);
729 kvm_vgic_sync_hwstate(vcpu);
735 kvm_arm_setup_debug(vcpu);
737 /**************************************************************
740 trace_kvm_entry(*vcpu_pc(vcpu));
741 guest_enter_irqoff();
744 kvm_arm_vhe_guest_enter();
745 ret = kvm_vcpu_run_vhe(vcpu);
746 kvm_arm_vhe_guest_exit();
748 ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
751 vcpu->mode = OUTSIDE_GUEST_MODE;
755 *************************************************************/
757 kvm_arm_clear_debug(vcpu);
760 * We must sync the PMU state before the vgic state so
761 * that the vgic can properly sample the updated state of the
764 kvm_pmu_sync_hwstate(vcpu);
767 * Sync the vgic state before syncing the timer state because
768 * the timer code needs to know if the virtual timer
769 * interrupts are active.
771 kvm_vgic_sync_hwstate(vcpu);
774 * Sync the timer hardware state before enabling interrupts as
775 * we don't want vtimer interrupts to race with syncing the
776 * timer virtual interrupt state.
778 if (static_branch_unlikely(&userspace_irqchip_in_use))
779 kvm_timer_sync_hwstate(vcpu);
781 kvm_arch_vcpu_ctxsync_fp(vcpu);
784 * We may have taken a host interrupt in HYP mode (ie
785 * while executing the guest). This interrupt is still
786 * pending, as we haven't serviced it yet!
788 * We're now back in SVC mode, with interrupts
789 * disabled. Enabling the interrupts now will have
790 * the effect of taking the interrupt again, in SVC
796 * We do local_irq_enable() before calling guest_exit() so
797 * that if a timer interrupt hits while running the guest we
798 * account that tick as being spent in the guest. We enable
799 * preemption after calling guest_exit() so that if we get
800 * preempted we make sure ticks after that is not counted as
804 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
806 /* Exit types that need handling before we can be preempted */
807 handle_exit_early(vcpu, run, ret);
811 ret = handle_exit(vcpu, run, ret);
814 /* Tell userspace about in-kernel device output levels */
815 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
816 kvm_timer_update_run(vcpu);
817 kvm_pmu_update_run(vcpu);
820 kvm_sigset_deactivate(vcpu);
826 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
832 if (number == KVM_ARM_IRQ_CPU_IRQ)
833 bit_index = __ffs(HCR_VI);
834 else /* KVM_ARM_IRQ_CPU_FIQ */
835 bit_index = __ffs(HCR_VF);
837 hcr = vcpu_hcr(vcpu);
839 set = test_and_set_bit(bit_index, hcr);
841 set = test_and_clear_bit(bit_index, hcr);
844 * If we didn't change anything, no need to wake up or kick other CPUs
850 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
851 * trigger a world-switch round on the running physical CPU to set the
852 * virtual IRQ/FIQ fields in the HCR appropriately.
854 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
860 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
863 u32 irq = irq_level->irq;
864 unsigned int irq_type, vcpu_idx, irq_num;
865 int nrcpus = atomic_read(&kvm->online_vcpus);
866 struct kvm_vcpu *vcpu = NULL;
867 bool level = irq_level->level;
869 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
870 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
871 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
873 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
876 case KVM_ARM_IRQ_TYPE_CPU:
877 if (irqchip_in_kernel(kvm))
880 if (vcpu_idx >= nrcpus)
883 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
887 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
890 return vcpu_interrupt_line(vcpu, irq_num, level);
891 case KVM_ARM_IRQ_TYPE_PPI:
892 if (!irqchip_in_kernel(kvm))
895 if (vcpu_idx >= nrcpus)
898 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
902 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
905 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
906 case KVM_ARM_IRQ_TYPE_SPI:
907 if (!irqchip_in_kernel(kvm))
910 if (irq_num < VGIC_NR_PRIVATE_IRQS)
913 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
919 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
920 const struct kvm_vcpu_init *init)
923 int phys_target = kvm_target_cpu();
925 if (init->target != phys_target)
929 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
930 * use the same target.
932 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
935 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
936 for (i = 0; i < sizeof(init->features) * 8; i++) {
937 bool set = (init->features[i / 32] & (1 << (i % 32)));
939 if (set && i >= KVM_VCPU_MAX_FEATURES)
943 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
944 * use the same feature set.
946 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
947 test_bit(i, vcpu->arch.features) != set)
951 set_bit(i, vcpu->arch.features);
954 vcpu->arch.target = phys_target;
956 /* Now we know what it is, we can reset it. */
957 return kvm_reset_vcpu(vcpu);
961 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
962 struct kvm_vcpu_init *init)
966 ret = kvm_vcpu_set_target(vcpu, init);
971 * Ensure a rebooted VM will fault in RAM pages and detect if the
972 * guest MMU is turned off and flush the caches as needed.
974 if (vcpu->arch.has_run_once)
975 stage2_unmap_vm(vcpu->kvm);
977 vcpu_reset_hcr(vcpu);
980 * Handle the "start in power-off" case.
982 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
983 vcpu_power_off(vcpu);
985 vcpu->arch.power_off = false;
990 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
991 struct kvm_device_attr *attr)
995 switch (attr->group) {
997 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1004 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1005 struct kvm_device_attr *attr)
1009 switch (attr->group) {
1011 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1018 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1019 struct kvm_device_attr *attr)
1023 switch (attr->group) {
1025 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1032 long kvm_arch_vcpu_ioctl(struct file *filp,
1033 unsigned int ioctl, unsigned long arg)
1035 struct kvm_vcpu *vcpu = filp->private_data;
1036 void __user *argp = (void __user *)arg;
1037 struct kvm_device_attr attr;
1041 case KVM_ARM_VCPU_INIT: {
1042 struct kvm_vcpu_init init;
1045 if (copy_from_user(&init, argp, sizeof(init)))
1048 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1051 case KVM_SET_ONE_REG:
1052 case KVM_GET_ONE_REG: {
1053 struct kvm_one_reg reg;
1056 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1060 if (copy_from_user(®, argp, sizeof(reg)))
1063 if (ioctl == KVM_SET_ONE_REG)
1064 r = kvm_arm_set_reg(vcpu, ®);
1066 r = kvm_arm_get_reg(vcpu, ®);
1069 case KVM_GET_REG_LIST: {
1070 struct kvm_reg_list __user *user_list = argp;
1071 struct kvm_reg_list reg_list;
1075 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1079 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1082 reg_list.n = kvm_arm_num_regs(vcpu);
1083 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1088 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1091 case KVM_SET_DEVICE_ATTR: {
1093 if (copy_from_user(&attr, argp, sizeof(attr)))
1095 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1098 case KVM_GET_DEVICE_ATTR: {
1100 if (copy_from_user(&attr, argp, sizeof(attr)))
1102 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1105 case KVM_HAS_DEVICE_ATTR: {
1107 if (copy_from_user(&attr, argp, sizeof(attr)))
1109 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1120 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1121 * @kvm: kvm instance
1122 * @log: slot id and address to which we copy the log
1124 * Steps 1-4 below provide general overview of dirty page logging. See
1125 * kvm_get_dirty_log_protect() function description for additional details.
1127 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1128 * always flush the TLB (step 4) even if previous step failed and the dirty
1129 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1130 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1131 * writes will be marked dirty for next log read.
1133 * 1. Take a snapshot of the bit and clear it if needed.
1134 * 2. Write protect the corresponding page.
1135 * 3. Copy the snapshot to the userspace.
1136 * 4. Flush TLB's if needed.
1138 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1140 bool is_dirty = false;
1143 mutex_lock(&kvm->slots_lock);
1145 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1148 kvm_flush_remote_tlbs(kvm);
1150 mutex_unlock(&kvm->slots_lock);
1154 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1155 struct kvm_arm_device_addr *dev_addr)
1157 unsigned long dev_id, type;
1159 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1160 KVM_ARM_DEVICE_ID_SHIFT;
1161 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1162 KVM_ARM_DEVICE_TYPE_SHIFT;
1165 case KVM_ARM_DEVICE_VGIC_V2:
1168 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1174 long kvm_arch_vm_ioctl(struct file *filp,
1175 unsigned int ioctl, unsigned long arg)
1177 struct kvm *kvm = filp->private_data;
1178 void __user *argp = (void __user *)arg;
1181 case KVM_CREATE_IRQCHIP: {
1185 mutex_lock(&kvm->lock);
1186 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1187 mutex_unlock(&kvm->lock);
1190 case KVM_ARM_SET_DEVICE_ADDR: {
1191 struct kvm_arm_device_addr dev_addr;
1193 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1195 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1197 case KVM_ARM_PREFERRED_TARGET: {
1199 struct kvm_vcpu_init init;
1201 err = kvm_vcpu_preferred_target(&init);
1205 if (copy_to_user(argp, &init, sizeof(init)))
1215 static void cpu_init_hyp_mode(void *dummy)
1217 phys_addr_t pgd_ptr;
1218 unsigned long hyp_stack_ptr;
1219 unsigned long stack_page;
1220 unsigned long vector_ptr;
1222 /* Switch from the HYP stub to our own HYP init vector */
1223 __hyp_set_vectors(kvm_get_idmap_vector());
1225 pgd_ptr = kvm_mmu_get_httbr();
1226 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1227 hyp_stack_ptr = stack_page + PAGE_SIZE;
1228 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1230 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1231 __cpu_init_stage2();
1233 kvm_arm_init_debug();
1236 static void cpu_hyp_reset(void)
1238 if (!is_kernel_in_hyp_mode())
1239 __hyp_reset_vectors();
1242 static void cpu_hyp_reinit(void)
1246 if (is_kernel_in_hyp_mode()) {
1248 * __cpu_init_stage2() is safe to call even if the PM
1249 * event was cancelled before the CPU was reset.
1251 __cpu_init_stage2();
1252 kvm_timer_init_vhe();
1254 cpu_init_hyp_mode(NULL);
1258 kvm_vgic_init_cpu_hardware();
1261 static void _kvm_arch_hardware_enable(void *discard)
1263 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1265 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1269 int kvm_arch_hardware_enable(void)
1271 _kvm_arch_hardware_enable(NULL);
1275 static void _kvm_arch_hardware_disable(void *discard)
1277 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1279 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1283 void kvm_arch_hardware_disable(void)
1285 _kvm_arch_hardware_disable(NULL);
1288 #ifdef CONFIG_CPU_PM
1289 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1294 * kvm_arm_hardware_enabled is left with its old value over
1295 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1300 if (__this_cpu_read(kvm_arm_hardware_enabled))
1302 * don't update kvm_arm_hardware_enabled here
1303 * so that the hardware will be re-enabled
1304 * when we resume. See below.
1309 case CPU_PM_ENTER_FAILED:
1311 if (__this_cpu_read(kvm_arm_hardware_enabled))
1312 /* The hardware was enabled before suspend. */
1322 static struct notifier_block hyp_init_cpu_pm_nb = {
1323 .notifier_call = hyp_init_cpu_pm_notifier,
1326 static void __init hyp_cpu_pm_init(void)
1328 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1330 static void __init hyp_cpu_pm_exit(void)
1332 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1335 static inline void hyp_cpu_pm_init(void)
1338 static inline void hyp_cpu_pm_exit(void)
1343 static int init_common_resources(void)
1345 /* set size of VMID supported by CPU */
1346 kvm_vmid_bits = kvm_get_vmid_bits();
1347 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1352 static int init_subsystems(void)
1357 * Enable hardware so that subsystem initialisation can access EL2.
1359 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1362 * Register CPU lower-power notifier
1367 * Init HYP view of VGIC
1369 err = kvm_vgic_hyp_init();
1372 vgic_present = true;
1376 vgic_present = false;
1384 * Init HYP architected timer support
1386 err = kvm_timer_hyp_init(vgic_present);
1391 kvm_coproc_table_init();
1394 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1399 static void teardown_hyp_mode(void)
1404 for_each_possible_cpu(cpu)
1405 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1410 * Inits Hyp-mode on all online CPUs
1412 static int init_hyp_mode(void)
1418 * Allocate Hyp PGD and setup Hyp identity mapping
1420 err = kvm_mmu_init();
1425 * Allocate stack pages for Hypervisor-mode
1427 for_each_possible_cpu(cpu) {
1428 unsigned long stack_page;
1430 stack_page = __get_free_page(GFP_KERNEL);
1436 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1440 * Map the Hyp-code called directly from the host
1442 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1443 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1445 kvm_err("Cannot map world-switch code\n");
1449 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1450 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1452 kvm_err("Cannot map rodata section\n");
1456 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1457 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1459 kvm_err("Cannot map bss section\n");
1463 err = kvm_map_vectors();
1465 kvm_err("Cannot map vectors\n");
1470 * Map the Hyp stack pages
1472 for_each_possible_cpu(cpu) {
1473 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1474 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1478 kvm_err("Cannot map hyp stack\n");
1483 for_each_possible_cpu(cpu) {
1484 kvm_cpu_context_t *cpu_ctxt;
1486 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1487 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1490 kvm_err("Cannot map host CPU state: %d\n", err);
1498 teardown_hyp_mode();
1499 kvm_err("error initializing Hyp mode: %d\n", err);
1503 static void check_kvm_target_cpu(void *ret)
1505 *(int *)ret = kvm_target_cpu();
1508 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1510 struct kvm_vcpu *vcpu;
1513 mpidr &= MPIDR_HWID_BITMASK;
1514 kvm_for_each_vcpu(i, vcpu, kvm) {
1515 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1521 bool kvm_arch_has_irq_bypass(void)
1526 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1527 struct irq_bypass_producer *prod)
1529 struct kvm_kernel_irqfd *irqfd =
1530 container_of(cons, struct kvm_kernel_irqfd, consumer);
1532 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1535 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1536 struct irq_bypass_producer *prod)
1538 struct kvm_kernel_irqfd *irqfd =
1539 container_of(cons, struct kvm_kernel_irqfd, consumer);
1541 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1545 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1547 struct kvm_kernel_irqfd *irqfd =
1548 container_of(cons, struct kvm_kernel_irqfd, consumer);
1550 kvm_arm_halt_guest(irqfd->kvm);
1553 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1555 struct kvm_kernel_irqfd *irqfd =
1556 container_of(cons, struct kvm_kernel_irqfd, consumer);
1558 kvm_arm_resume_guest(irqfd->kvm);
1562 * Initialize Hyp-mode and memory mappings on all CPUs.
1564 int kvm_arch_init(void *opaque)
1570 if (!is_hyp_mode_available()) {
1571 kvm_info("HYP mode not available\n");
1575 if (!kvm_arch_check_sve_has_vhe()) {
1576 kvm_pr_unimpl("SVE system without VHE unsupported. Broken cpu?");
1580 for_each_online_cpu(cpu) {
1581 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1583 kvm_err("Error, CPU %d not supported!\n", cpu);
1588 err = init_common_resources();
1592 in_hyp_mode = is_kernel_in_hyp_mode();
1595 err = init_hyp_mode();
1600 err = init_subsystems();
1605 kvm_info("VHE mode initialized successfully\n");
1607 kvm_info("Hyp mode initialized successfully\n");
1613 teardown_hyp_mode();
1618 /* NOP: Compiling as a module not supported */
1619 void kvm_arch_exit(void)
1621 kvm_perf_teardown();
1624 static int arm_init(void)
1626 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1630 module_init(arm_init);