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>
32 #include <kvm/arm_psci.h>
34 #define CREATE_TRACE_POINTS
37 #include <linux/uaccess.h>
38 #include <asm/ptrace.h>
40 #include <asm/tlbflush.h>
41 #include <asm/cacheflush.h>
43 #include <asm/kvm_arm.h>
44 #include <asm/kvm_asm.h>
45 #include <asm/kvm_mmu.h>
46 #include <asm/kvm_emulate.h>
47 #include <asm/kvm_coproc.h>
48 #include <asm/sections.h>
51 __asm__(".arch_extension virt");
54 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
55 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
57 /* Per-CPU variable containing the currently running vcpu. */
58 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
60 /* The VMID used in the VTTBR */
61 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
62 static u32 kvm_next_vmid;
63 static unsigned int kvm_vmid_bits __read_mostly;
64 static DEFINE_SPINLOCK(kvm_vmid_lock);
66 static bool vgic_present;
68 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
70 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
72 BUG_ON(preemptible());
73 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
77 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
78 * Must be called from non-preemptible context
80 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
82 BUG_ON(preemptible());
83 return __this_cpu_read(kvm_arm_running_vcpu);
87 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
89 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
91 return &kvm_arm_running_vcpu;
94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
96 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
99 int kvm_arch_hardware_setup(void)
104 void kvm_arch_check_processor_compat(void *rtn)
111 * kvm_arch_init_vm - initializes a VM data structure
112 * @kvm: pointer to the KVM struct
114 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
121 kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
122 if (!kvm->arch.last_vcpu_ran)
125 for_each_possible_cpu(cpu)
126 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
128 ret = kvm_alloc_stage2_pgd(kvm);
132 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
134 goto out_free_stage2_pgd;
136 kvm_vgic_early_init(kvm);
138 /* Mark the initial VMID generation invalid */
139 kvm->arch.vmid_gen = 0;
141 /* The maximum number of VCPUs is limited by the host's GIC model */
142 kvm->arch.max_vcpus = vgic_present ?
143 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
147 kvm_free_stage2_pgd(kvm);
149 free_percpu(kvm->arch.last_vcpu_ran);
150 kvm->arch.last_vcpu_ran = NULL;
154 bool kvm_arch_has_vcpu_debugfs(void)
159 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
164 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
166 return VM_FAULT_SIGBUS;
171 * kvm_arch_destroy_vm - destroy the VM data structure
172 * @kvm: pointer to the KVM struct
174 void kvm_arch_destroy_vm(struct kvm *kvm)
178 free_percpu(kvm->arch.last_vcpu_ran);
179 kvm->arch.last_vcpu_ran = NULL;
181 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
183 kvm_arch_vcpu_free(kvm->vcpus[i]);
184 kvm->vcpus[i] = NULL;
188 kvm_vgic_destroy(kvm);
191 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
195 case KVM_CAP_IRQCHIP:
198 case KVM_CAP_IOEVENTFD:
199 case KVM_CAP_DEVICE_CTRL:
200 case KVM_CAP_USER_MEMORY:
201 case KVM_CAP_SYNC_MMU:
202 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
203 case KVM_CAP_ONE_REG:
204 case KVM_CAP_ARM_PSCI:
205 case KVM_CAP_ARM_PSCI_0_2:
206 case KVM_CAP_READONLY_MEM:
207 case KVM_CAP_MP_STATE:
208 case KVM_CAP_IMMEDIATE_EXIT:
211 case KVM_CAP_ARM_SET_DEVICE_ADDR:
214 case KVM_CAP_NR_VCPUS:
215 r = num_online_cpus();
217 case KVM_CAP_MAX_VCPUS:
220 case KVM_CAP_MAX_VCPU_ID:
223 case KVM_CAP_NR_MEMSLOTS:
224 r = KVM_USER_MEM_SLOTS;
226 case KVM_CAP_MSI_DEVID:
230 r = kvm->arch.vgic.msis_require_devid;
232 case KVM_CAP_ARM_USER_IRQ:
234 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
235 * (bump this number if adding more devices)
240 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
246 long kvm_arch_dev_ioctl(struct file *filp,
247 unsigned int ioctl, unsigned long arg)
253 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
256 struct kvm_vcpu *vcpu;
258 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
263 if (id >= kvm->arch.max_vcpus) {
268 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
274 err = kvm_vcpu_init(vcpu, kvm, id);
278 err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
284 kvm_vcpu_uninit(vcpu);
286 kmem_cache_free(kvm_vcpu_cache, vcpu);
291 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
293 kvm_vgic_vcpu_early_init(vcpu);
296 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
298 kvm_mmu_free_memory_caches(vcpu);
299 kvm_timer_vcpu_terminate(vcpu);
300 kvm_vgic_vcpu_destroy(vcpu);
301 kvm_pmu_vcpu_destroy(vcpu);
302 kvm_vcpu_uninit(vcpu);
303 kmem_cache_free(kvm_vcpu_cache, vcpu);
306 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
308 kvm_arch_vcpu_free(vcpu);
311 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
313 return kvm_timer_should_fire(vcpu_vtimer(vcpu)) ||
314 kvm_timer_should_fire(vcpu_ptimer(vcpu));
317 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
319 kvm_timer_schedule(vcpu);
321 * If we're about to block (most likely because we've just hit a
322 * WFI), we need to sync back the state of the GIC CPU interface
323 * so that we have the lastest PMR and group enables. This ensures
324 * that kvm_arch_vcpu_runnable has up-to-date data to decide
325 * whether we have pending interrupts.
328 kvm_vgic_vmcr_sync(vcpu);
332 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
334 kvm_timer_unschedule(vcpu);
337 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
339 /* Force users to call KVM_ARM_VCPU_INIT */
340 vcpu->arch.target = -1;
341 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
343 /* Set up the timer */
344 kvm_timer_vcpu_init(vcpu);
346 kvm_arm_reset_debug_ptr(vcpu);
348 return kvm_vgic_vcpu_init(vcpu);
351 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
355 last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
358 * We might get preempted before the vCPU actually runs, but
359 * over-invalidation doesn't affect correctness.
361 if (*last_ran != vcpu->vcpu_id) {
362 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
363 *last_ran = vcpu->vcpu_id;
367 vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
369 kvm_arm_set_running_vcpu(vcpu);
374 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
380 kvm_arm_set_running_vcpu(NULL);
381 kvm_timer_vcpu_put(vcpu);
384 static void vcpu_power_off(struct kvm_vcpu *vcpu)
386 vcpu->arch.power_off = true;
387 kvm_make_request(KVM_REQ_SLEEP, vcpu);
391 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
392 struct kvm_mp_state *mp_state)
394 if (vcpu->arch.power_off)
395 mp_state->mp_state = KVM_MP_STATE_STOPPED;
397 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
402 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
403 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 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
429 && !v->arch.power_off && !v->arch.pause);
432 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
434 return vcpu_mode_priv(vcpu);
437 /* Just ensure a guest exit from a particular CPU */
438 static void exit_vm_noop(void *info)
442 void force_vm_exit(const cpumask_t *mask)
445 smp_call_function_many(mask, exit_vm_noop, NULL, true);
450 * need_new_vmid_gen - check that the VMID is still valid
451 * @kvm: The VM's VMID to check
453 * return true if there is a new generation of VMIDs being used
455 * The hardware supports only 256 values with the value zero reserved for the
456 * host, so we check if an assigned value belongs to a previous generation,
457 * which which requires us to assign a new value. If we're the first to use a
458 * VMID for the new generation, we must flush necessary caches and TLBs on all
461 static bool need_new_vmid_gen(struct kvm *kvm)
463 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
464 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
465 return unlikely(READ_ONCE(kvm->arch.vmid_gen) != current_vmid_gen);
469 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
470 * @kvm The guest that we are about to run
472 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
473 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
476 static void update_vttbr(struct kvm *kvm)
478 phys_addr_t pgd_phys;
481 if (!need_new_vmid_gen(kvm))
484 spin_lock(&kvm_vmid_lock);
487 * We need to re-check the vmid_gen here to ensure that if another vcpu
488 * already allocated a valid vmid for this vm, then this vcpu should
491 if (!need_new_vmid_gen(kvm)) {
492 spin_unlock(&kvm_vmid_lock);
496 /* First user of a new VMID generation? */
497 if (unlikely(kvm_next_vmid == 0)) {
498 atomic64_inc(&kvm_vmid_gen);
502 * On SMP we know no other CPUs can use this CPU's or each
503 * other's VMID after force_vm_exit returns since the
504 * kvm_vmid_lock blocks them from reentry to the guest.
506 force_vm_exit(cpu_all_mask);
508 * Now broadcast TLB + ICACHE invalidation over the inner
509 * shareable domain to make sure all data structures are
512 kvm_call_hyp(__kvm_flush_vm_context);
515 kvm->arch.vmid = kvm_next_vmid;
517 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
519 /* update vttbr to be used with the new vmid */
520 pgd_phys = virt_to_phys(kvm->arch.pgd);
521 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
522 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
523 kvm->arch.vttbr = pgd_phys | vmid;
526 WRITE_ONCE(kvm->arch.vmid_gen, atomic64_read(&kvm_vmid_gen));
528 spin_unlock(&kvm_vmid_lock);
531 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
533 struct kvm *kvm = vcpu->kvm;
536 if (likely(vcpu->arch.has_run_once))
539 vcpu->arch.has_run_once = true;
542 * Map the VGIC hardware resources before running a vcpu the first
545 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
546 ret = kvm_vgic_map_resources(kvm);
551 ret = kvm_timer_enable(vcpu);
555 ret = kvm_arm_pmu_v3_enable(vcpu);
560 bool kvm_arch_intc_initialized(struct kvm *kvm)
562 return vgic_initialized(kvm);
565 void kvm_arm_halt_guest(struct kvm *kvm)
568 struct kvm_vcpu *vcpu;
570 kvm_for_each_vcpu(i, vcpu, kvm)
571 vcpu->arch.pause = true;
572 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
575 void kvm_arm_resume_guest(struct kvm *kvm)
578 struct kvm_vcpu *vcpu;
580 kvm_for_each_vcpu(i, vcpu, kvm) {
581 vcpu->arch.pause = false;
582 swake_up(kvm_arch_vcpu_wq(vcpu));
586 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
588 struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
590 swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
591 (!vcpu->arch.pause)));
593 if (vcpu->arch.power_off || vcpu->arch.pause) {
594 /* Awaken to handle a signal, request we sleep again later. */
595 kvm_make_request(KVM_REQ_SLEEP, vcpu);
599 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
601 return vcpu->arch.target >= 0;
604 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
606 if (kvm_request_pending(vcpu)) {
607 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
608 vcpu_req_sleep(vcpu);
611 * Clear IRQ_PENDING requests that were made to guarantee
612 * that a VCPU sees new virtual interrupts.
614 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
619 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
620 * @vcpu: The VCPU pointer
621 * @run: The kvm_run structure pointer used for userspace state exchange
623 * This function is called through the VCPU_RUN ioctl called from user space. It
624 * will execute VM code in a loop until the time slice for the process is used
625 * or some emulation is needed from user space in which case the function will
626 * return with return value 0 and with the kvm_run structure filled in with the
627 * required data for the requested emulation.
629 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
633 if (unlikely(!kvm_vcpu_initialized(vcpu)))
636 ret = kvm_vcpu_first_run_init(vcpu);
640 if (run->exit_reason == KVM_EXIT_MMIO) {
641 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
646 if (run->immediate_exit)
649 kvm_sigset_activate(vcpu);
652 run->exit_reason = KVM_EXIT_UNKNOWN;
655 * Check conditions before entering the guest
659 update_vttbr(vcpu->kvm);
661 check_vcpu_requests(vcpu);
664 * Preparing the interrupts to be injected also
665 * involves poking the GIC, which must be done in a
666 * non-preemptible context.
670 kvm_pmu_flush_hwstate(vcpu);
672 kvm_timer_flush_hwstate(vcpu);
673 kvm_vgic_flush_hwstate(vcpu);
678 * If we have a singal pending, or need to notify a userspace
679 * irqchip about timer or PMU level changes, then we exit (and
680 * update the timer level state in kvm_timer_update_run
683 if (signal_pending(current) ||
684 kvm_timer_should_notify_user(vcpu) ||
685 kvm_pmu_should_notify_user(vcpu)) {
687 run->exit_reason = KVM_EXIT_INTR;
691 * Ensure we set mode to IN_GUEST_MODE after we disable
692 * interrupts and before the final VCPU requests check.
693 * See the comment in kvm_vcpu_exiting_guest_mode() and
694 * Documentation/virtual/kvm/vcpu-requests.rst
696 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
698 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
699 kvm_request_pending(vcpu)) {
700 vcpu->mode = OUTSIDE_GUEST_MODE;
702 kvm_pmu_sync_hwstate(vcpu);
703 kvm_timer_sync_hwstate(vcpu);
704 kvm_vgic_sync_hwstate(vcpu);
709 kvm_arm_setup_debug(vcpu);
711 /**************************************************************
714 trace_kvm_entry(*vcpu_pc(vcpu));
715 guest_enter_irqoff();
717 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
719 vcpu->mode = OUTSIDE_GUEST_MODE;
723 *************************************************************/
725 kvm_arm_clear_debug(vcpu);
728 * We may have taken a host interrupt in HYP mode (ie
729 * while executing the guest). This interrupt is still
730 * pending, as we haven't serviced it yet!
732 * We're now back in SVC mode, with interrupts
733 * disabled. Enabling the interrupts now will have
734 * the effect of taking the interrupt again, in SVC
740 * We do local_irq_enable() before calling guest_exit() so
741 * that if a timer interrupt hits while running the guest we
742 * account that tick as being spent in the guest. We enable
743 * preemption after calling guest_exit() so that if we get
744 * preempted we make sure ticks after that is not counted as
748 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
751 * We must sync the PMU and timer state before the vgic state so
752 * that the vgic can properly sample the updated state of the
755 kvm_pmu_sync_hwstate(vcpu);
756 kvm_timer_sync_hwstate(vcpu);
758 kvm_vgic_sync_hwstate(vcpu);
762 ret = handle_exit(vcpu, run, ret);
765 /* Tell userspace about in-kernel device output levels */
766 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
767 kvm_timer_update_run(vcpu);
768 kvm_pmu_update_run(vcpu);
771 kvm_sigset_deactivate(vcpu);
776 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
782 if (number == KVM_ARM_IRQ_CPU_IRQ)
783 bit_index = __ffs(HCR_VI);
784 else /* KVM_ARM_IRQ_CPU_FIQ */
785 bit_index = __ffs(HCR_VF);
787 ptr = (unsigned long *)&vcpu->arch.irq_lines;
789 set = test_and_set_bit(bit_index, ptr);
791 set = test_and_clear_bit(bit_index, ptr);
794 * If we didn't change anything, no need to wake up or kick other CPUs
800 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
801 * trigger a world-switch round on the running physical CPU to set the
802 * virtual IRQ/FIQ fields in the HCR appropriately.
804 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
810 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
813 u32 irq = irq_level->irq;
814 unsigned int irq_type, vcpu_idx, irq_num;
815 int nrcpus = atomic_read(&kvm->online_vcpus);
816 struct kvm_vcpu *vcpu = NULL;
817 bool level = irq_level->level;
819 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
820 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
821 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
823 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
826 case KVM_ARM_IRQ_TYPE_CPU:
827 if (irqchip_in_kernel(kvm))
830 if (vcpu_idx >= nrcpus)
833 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
837 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
840 return vcpu_interrupt_line(vcpu, irq_num, level);
841 case KVM_ARM_IRQ_TYPE_PPI:
842 if (!irqchip_in_kernel(kvm))
845 if (vcpu_idx >= nrcpus)
848 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
852 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
855 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
856 case KVM_ARM_IRQ_TYPE_SPI:
857 if (!irqchip_in_kernel(kvm))
860 if (irq_num < VGIC_NR_PRIVATE_IRQS)
863 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
869 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
870 const struct kvm_vcpu_init *init)
873 int phys_target = kvm_target_cpu();
875 if (init->target != phys_target)
879 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
880 * use the same target.
882 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
885 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
886 for (i = 0; i < sizeof(init->features) * 8; i++) {
887 bool set = (init->features[i / 32] & (1 << (i % 32)));
889 if (set && i >= KVM_VCPU_MAX_FEATURES)
893 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
894 * use the same feature set.
896 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
897 test_bit(i, vcpu->arch.features) != set)
901 set_bit(i, vcpu->arch.features);
904 vcpu->arch.target = phys_target;
906 /* Now we know what it is, we can reset it. */
907 ret = kvm_reset_vcpu(vcpu);
909 vcpu->arch.target = -1;
910 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
916 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
917 struct kvm_vcpu_init *init)
921 ret = kvm_vcpu_set_target(vcpu, init);
926 * Ensure a rebooted VM will fault in RAM pages and detect if the
927 * guest MMU is turned off and flush the caches as needed.
929 if (vcpu->arch.has_run_once)
930 stage2_unmap_vm(vcpu->kvm);
932 vcpu_reset_hcr(vcpu);
935 * Handle the "start in power-off" case.
937 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
938 vcpu_power_off(vcpu);
940 vcpu->arch.power_off = false;
945 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
946 struct kvm_device_attr *attr)
950 switch (attr->group) {
952 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
959 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
960 struct kvm_device_attr *attr)
964 switch (attr->group) {
966 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
973 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
974 struct kvm_device_attr *attr)
978 switch (attr->group) {
980 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
987 long kvm_arch_vcpu_ioctl(struct file *filp,
988 unsigned int ioctl, unsigned long arg)
990 struct kvm_vcpu *vcpu = filp->private_data;
991 void __user *argp = (void __user *)arg;
992 struct kvm_device_attr attr;
995 case KVM_ARM_VCPU_INIT: {
996 struct kvm_vcpu_init init;
998 if (copy_from_user(&init, argp, sizeof(init)))
1001 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1003 case KVM_SET_ONE_REG:
1004 case KVM_GET_ONE_REG: {
1005 struct kvm_one_reg reg;
1007 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1010 if (copy_from_user(®, argp, sizeof(reg)))
1012 if (ioctl == KVM_SET_ONE_REG)
1013 return kvm_arm_set_reg(vcpu, ®);
1015 return kvm_arm_get_reg(vcpu, ®);
1017 case KVM_GET_REG_LIST: {
1018 struct kvm_reg_list __user *user_list = argp;
1019 struct kvm_reg_list reg_list;
1022 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1025 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
1028 reg_list.n = kvm_arm_num_regs(vcpu);
1029 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
1033 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1035 case KVM_SET_DEVICE_ATTR: {
1036 if (copy_from_user(&attr, argp, sizeof(attr)))
1038 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1040 case KVM_GET_DEVICE_ATTR: {
1041 if (copy_from_user(&attr, argp, sizeof(attr)))
1043 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1045 case KVM_HAS_DEVICE_ATTR: {
1046 if (copy_from_user(&attr, argp, sizeof(attr)))
1048 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1056 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1057 * @kvm: kvm instance
1058 * @log: slot id and address to which we copy the log
1060 * Steps 1-4 below provide general overview of dirty page logging. See
1061 * kvm_get_dirty_log_protect() function description for additional details.
1063 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1064 * always flush the TLB (step 4) even if previous step failed and the dirty
1065 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1066 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1067 * writes will be marked dirty for next log read.
1069 * 1. Take a snapshot of the bit and clear it if needed.
1070 * 2. Write protect the corresponding page.
1071 * 3. Copy the snapshot to the userspace.
1072 * 4. Flush TLB's if needed.
1074 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1076 bool is_dirty = false;
1079 mutex_lock(&kvm->slots_lock);
1081 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1084 kvm_flush_remote_tlbs(kvm);
1086 mutex_unlock(&kvm->slots_lock);
1090 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1091 struct kvm_arm_device_addr *dev_addr)
1093 unsigned long dev_id, type;
1095 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1096 KVM_ARM_DEVICE_ID_SHIFT;
1097 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1098 KVM_ARM_DEVICE_TYPE_SHIFT;
1101 case KVM_ARM_DEVICE_VGIC_V2:
1104 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1110 long kvm_arch_vm_ioctl(struct file *filp,
1111 unsigned int ioctl, unsigned long arg)
1113 struct kvm *kvm = filp->private_data;
1114 void __user *argp = (void __user *)arg;
1117 case KVM_CREATE_IRQCHIP: {
1121 mutex_lock(&kvm->lock);
1122 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1123 mutex_unlock(&kvm->lock);
1126 case KVM_ARM_SET_DEVICE_ADDR: {
1127 struct kvm_arm_device_addr dev_addr;
1129 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1131 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1133 case KVM_ARM_PREFERRED_TARGET: {
1135 struct kvm_vcpu_init init;
1137 err = kvm_vcpu_preferred_target(&init);
1141 if (copy_to_user(argp, &init, sizeof(init)))
1151 static void cpu_init_hyp_mode(void *dummy)
1153 phys_addr_t pgd_ptr;
1154 unsigned long hyp_stack_ptr;
1155 unsigned long stack_page;
1156 unsigned long vector_ptr;
1158 /* Switch from the HYP stub to our own HYP init vector */
1159 __hyp_set_vectors(kvm_get_idmap_vector());
1161 pgd_ptr = kvm_mmu_get_httbr();
1162 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1163 hyp_stack_ptr = stack_page + PAGE_SIZE;
1164 vector_ptr = (unsigned long)kvm_get_hyp_vector();
1166 __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1167 __cpu_init_stage2();
1170 static void cpu_hyp_reset(void)
1172 if (!is_kernel_in_hyp_mode())
1173 __hyp_reset_vectors();
1176 static void cpu_hyp_reinit(void)
1180 if (is_kernel_in_hyp_mode()) {
1182 * __cpu_init_stage2() is safe to call even if the PM
1183 * event was cancelled before the CPU was reset.
1185 __cpu_init_stage2();
1186 kvm_timer_init_vhe();
1188 cpu_init_hyp_mode(NULL);
1191 kvm_arm_init_debug();
1194 kvm_vgic_init_cpu_hardware();
1197 static void _kvm_arch_hardware_enable(void *discard)
1199 if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1201 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1205 int kvm_arch_hardware_enable(void)
1207 _kvm_arch_hardware_enable(NULL);
1211 static void _kvm_arch_hardware_disable(void *discard)
1213 if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1215 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1219 void kvm_arch_hardware_disable(void)
1221 _kvm_arch_hardware_disable(NULL);
1224 #ifdef CONFIG_CPU_PM
1225 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1230 * kvm_arm_hardware_enabled is left with its old value over
1231 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1236 if (__this_cpu_read(kvm_arm_hardware_enabled))
1238 * don't update kvm_arm_hardware_enabled here
1239 * so that the hardware will be re-enabled
1240 * when we resume. See below.
1245 case CPU_PM_ENTER_FAILED:
1247 if (__this_cpu_read(kvm_arm_hardware_enabled))
1248 /* The hardware was enabled before suspend. */
1258 static struct notifier_block hyp_init_cpu_pm_nb = {
1259 .notifier_call = hyp_init_cpu_pm_notifier,
1262 static void __init hyp_cpu_pm_init(void)
1264 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1266 static void __init hyp_cpu_pm_exit(void)
1268 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1271 static inline void hyp_cpu_pm_init(void)
1274 static inline void hyp_cpu_pm_exit(void)
1279 static int init_common_resources(void)
1281 /* set size of VMID supported by CPU */
1282 kvm_vmid_bits = kvm_get_vmid_bits();
1283 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1288 static int init_subsystems(void)
1293 * Enable hardware so that subsystem initialisation can access EL2.
1295 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1298 * Register CPU lower-power notifier
1303 * Init HYP view of VGIC
1305 err = kvm_vgic_hyp_init();
1308 vgic_present = true;
1312 vgic_present = false;
1320 * Init HYP architected timer support
1322 err = kvm_timer_hyp_init();
1327 kvm_coproc_table_init();
1330 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1335 static void teardown_hyp_mode(void)
1340 for_each_possible_cpu(cpu)
1341 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1346 * Inits Hyp-mode on all online CPUs
1348 static int init_hyp_mode(void)
1354 * Allocate Hyp PGD and setup Hyp identity mapping
1356 err = kvm_mmu_init();
1361 * Allocate stack pages for Hypervisor-mode
1363 for_each_possible_cpu(cpu) {
1364 unsigned long stack_page;
1366 stack_page = __get_free_page(GFP_KERNEL);
1372 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1376 * Map the Hyp-code called directly from the host
1378 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1379 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1381 kvm_err("Cannot map world-switch code\n");
1385 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1386 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1388 kvm_err("Cannot map rodata section\n");
1392 err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1393 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1395 kvm_err("Cannot map bss section\n");
1399 err = kvm_map_vectors();
1401 kvm_err("Cannot map vectors\n");
1406 * Map the Hyp stack pages
1408 for_each_possible_cpu(cpu) {
1409 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1410 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1414 kvm_err("Cannot map hyp stack\n");
1419 for_each_possible_cpu(cpu) {
1420 kvm_cpu_context_t *cpu_ctxt;
1422 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1423 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1426 kvm_err("Cannot map host CPU state: %d\n", err);
1431 err = hyp_map_aux_data();
1433 kvm_err("Cannot map host auxilary data: %d\n", err);
1438 teardown_hyp_mode();
1439 kvm_err("error initializing Hyp mode: %d\n", err);
1443 static void check_kvm_target_cpu(void *ret)
1445 *(int *)ret = kvm_target_cpu();
1448 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1450 struct kvm_vcpu *vcpu;
1453 mpidr &= MPIDR_HWID_BITMASK;
1454 kvm_for_each_vcpu(i, vcpu, kvm) {
1455 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1462 * Initialize Hyp-mode and memory mappings on all CPUs.
1464 int kvm_arch_init(void *opaque)
1470 if (!is_hyp_mode_available()) {
1471 kvm_info("HYP mode not available\n");
1475 for_each_online_cpu(cpu) {
1476 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1478 kvm_err("Error, CPU %d not supported!\n", cpu);
1483 err = init_common_resources();
1487 in_hyp_mode = is_kernel_in_hyp_mode();
1490 err = init_hyp_mode();
1495 err = init_subsystems();
1500 kvm_info("VHE mode initialized successfully\n");
1502 kvm_info("Hyp mode initialized successfully\n");
1508 teardown_hyp_mode();
1513 /* NOP: Compiling as a module not supported */
1514 void kvm_arch_exit(void)
1516 kvm_perf_teardown();
1519 static int arm_init(void)
1521 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1525 module_init(arm_init);