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.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/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>
32 #define CREATE_TRACE_POINTS
35 #include <asm/uaccess.h>
36 #include <asm/ptrace.h>
38 #include <asm/tlbflush.h>
39 #include <asm/cacheflush.h>
41 #include <asm/kvm_arm.h>
42 #include <asm/kvm_asm.h>
43 #include <asm/kvm_mmu.h>
44 #include <asm/kvm_emulate.h>
45 #include <asm/kvm_coproc.h>
46 #include <asm/kvm_psci.h>
47 #include <asm/sections.h>
50 __asm__(".arch_extension virt");
53 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
54 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
55 static unsigned long hyp_default_vectors;
57 /* Per-CPU variable containing the currently running vcpu. */
58 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
60 /* The VMID used in the VTTBR */
61 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
62 static u32 kvm_next_vmid;
63 static unsigned int kvm_vmid_bits __read_mostly;
64 static DEFINE_SPINLOCK(kvm_vmid_lock);
66 static bool vgic_present;
68 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
70 BUG_ON(preemptible());
71 __this_cpu_write(kvm_arm_running_vcpu, vcpu);
75 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
76 * Must be called from non-preemptible context
78 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
80 BUG_ON(preemptible());
81 return __this_cpu_read(kvm_arm_running_vcpu);
85 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
87 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
89 return &kvm_arm_running_vcpu;
92 int kvm_arch_hardware_enable(void)
97 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
99 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
102 int kvm_arch_hardware_setup(void)
107 void kvm_arch_check_processor_compat(void *rtn)
114 * kvm_arch_init_vm - initializes a VM data structure
115 * @kvm: pointer to the KVM struct
117 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
124 ret = kvm_alloc_stage2_pgd(kvm);
128 ret = create_hyp_mappings(kvm, kvm + 1);
130 goto out_free_stage2_pgd;
132 kvm_vgic_early_init(kvm);
135 /* Mark the initial VMID generation invalid */
136 kvm->arch.vmid_gen = 0;
138 /* The maximum number of VCPUs is limited by the host's GIC model */
139 kvm->arch.max_vcpus = vgic_present ?
140 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
144 kvm_free_stage2_pgd(kvm);
149 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
151 return VM_FAULT_SIGBUS;
156 * kvm_arch_destroy_vm - destroy the VM data structure
157 * @kvm: pointer to the KVM struct
159 void kvm_arch_destroy_vm(struct kvm *kvm)
163 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
165 kvm_arch_vcpu_free(kvm->vcpus[i]);
166 kvm->vcpus[i] = NULL;
170 kvm_vgic_destroy(kvm);
173 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
177 case KVM_CAP_IRQCHIP:
180 case KVM_CAP_IOEVENTFD:
181 case KVM_CAP_DEVICE_CTRL:
182 case KVM_CAP_USER_MEMORY:
183 case KVM_CAP_SYNC_MMU:
184 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
185 case KVM_CAP_ONE_REG:
186 case KVM_CAP_ARM_PSCI:
187 case KVM_CAP_ARM_PSCI_0_2:
188 case KVM_CAP_READONLY_MEM:
189 case KVM_CAP_MP_STATE:
192 case KVM_CAP_COALESCED_MMIO:
193 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
195 case KVM_CAP_ARM_SET_DEVICE_ADDR:
198 case KVM_CAP_NR_VCPUS:
199 r = num_online_cpus();
201 case KVM_CAP_MAX_VCPUS:
205 r = kvm_arch_dev_ioctl_check_extension(ext);
211 long kvm_arch_dev_ioctl(struct file *filp,
212 unsigned int ioctl, unsigned long arg)
218 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
221 struct kvm_vcpu *vcpu;
223 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
228 if (id >= kvm->arch.max_vcpus) {
233 vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
239 err = kvm_vcpu_init(vcpu, kvm, id);
243 err = create_hyp_mappings(vcpu, vcpu + 1);
249 kvm_vcpu_uninit(vcpu);
251 kmem_cache_free(kvm_vcpu_cache, vcpu);
256 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
258 kvm_vgic_vcpu_early_init(vcpu);
261 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
263 kvm_mmu_free_memory_caches(vcpu);
264 kvm_timer_vcpu_terminate(vcpu);
265 kvm_vgic_vcpu_destroy(vcpu);
266 kmem_cache_free(kvm_vcpu_cache, vcpu);
269 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
271 kvm_arch_vcpu_free(vcpu);
274 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
276 return kvm_timer_should_fire(vcpu);
279 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
281 kvm_timer_schedule(vcpu);
284 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
286 kvm_timer_unschedule(vcpu);
289 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
291 /* Force users to call KVM_ARM_VCPU_INIT */
292 vcpu->arch.target = -1;
293 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
295 /* Set up the timer */
296 kvm_timer_vcpu_init(vcpu);
298 kvm_arm_reset_debug_ptr(vcpu);
303 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
306 vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
308 kvm_arm_set_running_vcpu(vcpu);
311 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
314 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
315 * if the vcpu is no longer assigned to a cpu. This is used for the
316 * optimized make_all_cpus_request path.
320 kvm_arm_set_running_vcpu(NULL);
323 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
324 struct kvm_mp_state *mp_state)
326 if (vcpu->arch.power_off)
327 mp_state->mp_state = KVM_MP_STATE_STOPPED;
329 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
334 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
335 struct kvm_mp_state *mp_state)
337 switch (mp_state->mp_state) {
338 case KVM_MP_STATE_RUNNABLE:
339 vcpu->arch.power_off = false;
341 case KVM_MP_STATE_STOPPED:
342 vcpu->arch.power_off = true;
352 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
353 * @v: The VCPU pointer
355 * If the guest CPU is not waiting for interrupts or an interrupt line is
356 * asserted, the CPU is by definition runnable.
358 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
360 return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
361 && !v->arch.power_off && !v->arch.pause);
364 /* Just ensure a guest exit from a particular CPU */
365 static void exit_vm_noop(void *info)
369 void force_vm_exit(const cpumask_t *mask)
371 smp_call_function_many(mask, exit_vm_noop, NULL, true);
375 * need_new_vmid_gen - check that the VMID is still valid
376 * @kvm: The VM's VMID to checkt
378 * return true if there is a new generation of VMIDs being used
380 * The hardware supports only 256 values with the value zero reserved for the
381 * host, so we check if an assigned value belongs to a previous generation,
382 * which which requires us to assign a new value. If we're the first to use a
383 * VMID for the new generation, we must flush necessary caches and TLBs on all
386 static bool need_new_vmid_gen(struct kvm *kvm)
388 return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
392 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
393 * @kvm The guest that we are about to run
395 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
396 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
399 static void update_vttbr(struct kvm *kvm)
401 phys_addr_t pgd_phys;
404 if (!need_new_vmid_gen(kvm))
407 spin_lock(&kvm_vmid_lock);
410 * We need to re-check the vmid_gen here to ensure that if another vcpu
411 * already allocated a valid vmid for this vm, then this vcpu should
414 if (!need_new_vmid_gen(kvm)) {
415 spin_unlock(&kvm_vmid_lock);
419 /* First user of a new VMID generation? */
420 if (unlikely(kvm_next_vmid == 0)) {
421 atomic64_inc(&kvm_vmid_gen);
425 * On SMP we know no other CPUs can use this CPU's or each
426 * other's VMID after force_vm_exit returns since the
427 * kvm_vmid_lock blocks them from reentry to the guest.
429 force_vm_exit(cpu_all_mask);
431 * Now broadcast TLB + ICACHE invalidation over the inner
432 * shareable domain to make sure all data structures are
435 kvm_call_hyp(__kvm_flush_vm_context);
438 kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
439 kvm->arch.vmid = kvm_next_vmid;
441 kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
443 /* update vttbr to be used with the new vmid */
444 pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
445 BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
446 vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
447 kvm->arch.vttbr = pgd_phys | vmid;
449 spin_unlock(&kvm_vmid_lock);
452 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
454 struct kvm *kvm = vcpu->kvm;
457 if (likely(vcpu->arch.has_run_once))
460 vcpu->arch.has_run_once = true;
463 * Map the VGIC hardware resources before running a vcpu the first
466 if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
467 ret = kvm_vgic_map_resources(kvm);
473 * Enable the arch timers only if we have an in-kernel VGIC
474 * and it has been properly initialized, since we cannot handle
475 * interrupts from the virtual timer with a userspace gic.
477 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
478 kvm_timer_enable(kvm);
483 bool kvm_arch_intc_initialized(struct kvm *kvm)
485 return vgic_initialized(kvm);
488 static void kvm_arm_halt_guest(struct kvm *kvm) __maybe_unused;
489 static void kvm_arm_resume_guest(struct kvm *kvm) __maybe_unused;
491 static void kvm_arm_halt_guest(struct kvm *kvm)
494 struct kvm_vcpu *vcpu;
496 kvm_for_each_vcpu(i, vcpu, kvm)
497 vcpu->arch.pause = true;
498 force_vm_exit(cpu_all_mask);
501 static void kvm_arm_resume_guest(struct kvm *kvm)
504 struct kvm_vcpu *vcpu;
506 kvm_for_each_vcpu(i, vcpu, kvm) {
507 wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
509 vcpu->arch.pause = false;
510 wake_up_interruptible(wq);
514 static void vcpu_sleep(struct kvm_vcpu *vcpu)
516 wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
518 wait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
519 (!vcpu->arch.pause)));
522 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
524 return vcpu->arch.target >= 0;
528 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
529 * @vcpu: The VCPU pointer
530 * @run: The kvm_run structure pointer used for userspace state exchange
532 * This function is called through the VCPU_RUN ioctl called from user space. It
533 * will execute VM code in a loop until the time slice for the process is used
534 * or some emulation is needed from user space in which case the function will
535 * return with return value 0 and with the kvm_run structure filled in with the
536 * required data for the requested emulation.
538 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
543 if (unlikely(!kvm_vcpu_initialized(vcpu)))
546 ret = kvm_vcpu_first_run_init(vcpu);
550 if (run->exit_reason == KVM_EXIT_MMIO) {
551 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
556 if (vcpu->sigset_active)
557 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
560 run->exit_reason = KVM_EXIT_UNKNOWN;
563 * Check conditions before entering the guest
567 update_vttbr(vcpu->kvm);
569 if (vcpu->arch.power_off || vcpu->arch.pause)
573 * Preparing the interrupts to be injected also
574 * involves poking the GIC, which must be done in a
575 * non-preemptible context.
578 kvm_timer_flush_hwstate(vcpu);
579 kvm_vgic_flush_hwstate(vcpu);
584 * Re-check atomic conditions
586 if (signal_pending(current)) {
588 run->exit_reason = KVM_EXIT_INTR;
591 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
592 vcpu->arch.power_off || vcpu->arch.pause) {
594 kvm_timer_sync_hwstate(vcpu);
595 kvm_vgic_sync_hwstate(vcpu);
600 kvm_arm_setup_debug(vcpu);
602 /**************************************************************
605 trace_kvm_entry(*vcpu_pc(vcpu));
607 vcpu->mode = IN_GUEST_MODE;
609 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
611 vcpu->mode = OUTSIDE_GUEST_MODE;
614 *************************************************************/
616 kvm_arm_clear_debug(vcpu);
619 * We may have taken a host interrupt in HYP mode (ie
620 * while executing the guest). This interrupt is still
621 * pending, as we haven't serviced it yet!
623 * We're now back in SVC mode, with interrupts
624 * disabled. Enabling the interrupts now will have
625 * the effect of taking the interrupt again, in SVC
631 * We do local_irq_enable() before calling kvm_guest_exit() so
632 * that if a timer interrupt hits while running the guest we
633 * account that tick as being spent in the guest. We enable
634 * preemption after calling kvm_guest_exit() so that if we get
635 * preempted we make sure ticks after that is not counted as
639 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
642 * We must sync the timer state before the vgic state so that
643 * the vgic can properly sample the updated state of the
646 kvm_timer_sync_hwstate(vcpu);
648 kvm_vgic_sync_hwstate(vcpu);
652 ret = handle_exit(vcpu, run, ret);
655 if (vcpu->sigset_active)
656 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
660 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
666 if (number == KVM_ARM_IRQ_CPU_IRQ)
667 bit_index = __ffs(HCR_VI);
668 else /* KVM_ARM_IRQ_CPU_FIQ */
669 bit_index = __ffs(HCR_VF);
671 ptr = (unsigned long *)&vcpu->arch.irq_lines;
673 set = test_and_set_bit(bit_index, ptr);
675 set = test_and_clear_bit(bit_index, ptr);
678 * If we didn't change anything, no need to wake up or kick other CPUs
684 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
685 * trigger a world-switch round on the running physical CPU to set the
686 * virtual IRQ/FIQ fields in the HCR appropriately.
693 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
696 u32 irq = irq_level->irq;
697 unsigned int irq_type, vcpu_idx, irq_num;
698 int nrcpus = atomic_read(&kvm->online_vcpus);
699 struct kvm_vcpu *vcpu = NULL;
700 bool level = irq_level->level;
702 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
703 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
704 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
706 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
709 case KVM_ARM_IRQ_TYPE_CPU:
710 if (irqchip_in_kernel(kvm))
713 if (vcpu_idx >= nrcpus)
716 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
720 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
723 return vcpu_interrupt_line(vcpu, irq_num, level);
724 case KVM_ARM_IRQ_TYPE_PPI:
725 if (!irqchip_in_kernel(kvm))
728 if (vcpu_idx >= nrcpus)
731 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
735 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
738 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
739 case KVM_ARM_IRQ_TYPE_SPI:
740 if (!irqchip_in_kernel(kvm))
743 if (irq_num < VGIC_NR_PRIVATE_IRQS)
746 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
752 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
753 const struct kvm_vcpu_init *init)
756 int phys_target = kvm_target_cpu();
758 if (init->target != phys_target)
762 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
763 * use the same target.
765 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
768 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
769 for (i = 0; i < sizeof(init->features) * 8; i++) {
770 bool set = (init->features[i / 32] & (1 << (i % 32)));
772 if (set && i >= KVM_VCPU_MAX_FEATURES)
776 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
777 * use the same feature set.
779 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
780 test_bit(i, vcpu->arch.features) != set)
784 set_bit(i, vcpu->arch.features);
787 vcpu->arch.target = phys_target;
789 /* Now we know what it is, we can reset it. */
790 return kvm_reset_vcpu(vcpu);
794 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
795 struct kvm_vcpu_init *init)
799 ret = kvm_vcpu_set_target(vcpu, init);
804 * Ensure a rebooted VM will fault in RAM pages and detect if the
805 * guest MMU is turned off and flush the caches as needed.
807 if (vcpu->arch.has_run_once)
808 stage2_unmap_vm(vcpu->kvm);
810 vcpu_reset_hcr(vcpu);
813 * Handle the "start in power-off" case.
815 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
816 vcpu->arch.power_off = true;
818 vcpu->arch.power_off = false;
823 long kvm_arch_vcpu_ioctl(struct file *filp,
824 unsigned int ioctl, unsigned long arg)
826 struct kvm_vcpu *vcpu = filp->private_data;
827 void __user *argp = (void __user *)arg;
830 case KVM_ARM_VCPU_INIT: {
831 struct kvm_vcpu_init init;
833 if (copy_from_user(&init, argp, sizeof(init)))
836 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
838 case KVM_SET_ONE_REG:
839 case KVM_GET_ONE_REG: {
840 struct kvm_one_reg reg;
842 if (unlikely(!kvm_vcpu_initialized(vcpu)))
845 if (copy_from_user(®, argp, sizeof(reg)))
847 if (ioctl == KVM_SET_ONE_REG)
848 return kvm_arm_set_reg(vcpu, ®);
850 return kvm_arm_get_reg(vcpu, ®);
852 case KVM_GET_REG_LIST: {
853 struct kvm_reg_list __user *user_list = argp;
854 struct kvm_reg_list reg_list;
857 if (unlikely(!kvm_vcpu_initialized(vcpu)))
860 if (copy_from_user(®_list, user_list, sizeof(reg_list)))
863 reg_list.n = kvm_arm_num_regs(vcpu);
864 if (copy_to_user(user_list, ®_list, sizeof(reg_list)))
868 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
876 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
878 * @log: slot id and address to which we copy the log
880 * Steps 1-4 below provide general overview of dirty page logging. See
881 * kvm_get_dirty_log_protect() function description for additional details.
883 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
884 * always flush the TLB (step 4) even if previous step failed and the dirty
885 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
886 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
887 * writes will be marked dirty for next log read.
889 * 1. Take a snapshot of the bit and clear it if needed.
890 * 2. Write protect the corresponding page.
891 * 3. Copy the snapshot to the userspace.
892 * 4. Flush TLB's if needed.
894 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
896 bool is_dirty = false;
899 mutex_lock(&kvm->slots_lock);
901 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
904 kvm_flush_remote_tlbs(kvm);
906 mutex_unlock(&kvm->slots_lock);
910 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
911 struct kvm_arm_device_addr *dev_addr)
913 unsigned long dev_id, type;
915 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
916 KVM_ARM_DEVICE_ID_SHIFT;
917 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
918 KVM_ARM_DEVICE_TYPE_SHIFT;
921 case KVM_ARM_DEVICE_VGIC_V2:
924 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
930 long kvm_arch_vm_ioctl(struct file *filp,
931 unsigned int ioctl, unsigned long arg)
933 struct kvm *kvm = filp->private_data;
934 void __user *argp = (void __user *)arg;
937 case KVM_CREATE_IRQCHIP: {
940 return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
942 case KVM_ARM_SET_DEVICE_ADDR: {
943 struct kvm_arm_device_addr dev_addr;
945 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
947 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
949 case KVM_ARM_PREFERRED_TARGET: {
951 struct kvm_vcpu_init init;
953 err = kvm_vcpu_preferred_target(&init);
957 if (copy_to_user(argp, &init, sizeof(init)))
967 static void cpu_init_stage2(void *dummy)
972 static void cpu_init_hyp_mode(void *dummy)
974 phys_addr_t boot_pgd_ptr;
976 unsigned long hyp_stack_ptr;
977 unsigned long stack_page;
978 unsigned long vector_ptr;
980 /* Switch from the HYP stub to our own HYP init vector */
981 __hyp_set_vectors(kvm_get_idmap_vector());
983 boot_pgd_ptr = kvm_mmu_get_boot_httbr();
984 pgd_ptr = kvm_mmu_get_httbr();
985 stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
986 hyp_stack_ptr = stack_page + PAGE_SIZE;
987 vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
989 __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
992 kvm_arm_init_debug();
995 static void cpu_hyp_reinit(void)
997 if (is_kernel_in_hyp_mode()) {
999 * cpu_init_stage2() is safe to call even if the PM
1000 * event was cancelled before the CPU was reset.
1002 cpu_init_stage2(NULL);
1004 if (__hyp_get_vectors() == hyp_default_vectors)
1005 cpu_init_hyp_mode(NULL);
1009 static int hyp_init_cpu_notify(struct notifier_block *self,
1010 unsigned long action, void *cpu)
1014 case CPU_STARTING_FROZEN:
1021 static struct notifier_block hyp_init_cpu_nb = {
1022 .notifier_call = hyp_init_cpu_notify,
1025 #ifdef CONFIG_CPU_PM
1026 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1030 if (cmd == CPU_PM_EXIT) {
1038 static struct notifier_block hyp_init_cpu_pm_nb = {
1039 .notifier_call = hyp_init_cpu_pm_notifier,
1042 static void __init hyp_cpu_pm_init(void)
1044 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1047 static inline void hyp_cpu_pm_init(void)
1052 static void teardown_common_resources(void)
1054 free_percpu(kvm_host_cpu_state);
1057 static int init_common_resources(void)
1059 kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1060 if (!kvm_host_cpu_state) {
1061 kvm_err("Cannot allocate host CPU state\n");
1068 static int init_subsystems(void)
1073 * Register CPU Hotplug notifier
1075 cpu_notifier_register_begin();
1076 err = __register_cpu_notifier(&hyp_init_cpu_nb);
1077 cpu_notifier_register_done();
1079 kvm_err("Cannot register KVM init CPU notifier (%d)\n", err);
1084 * Register CPU lower-power notifier
1089 * Init HYP view of VGIC
1091 err = kvm_vgic_hyp_init();
1094 vgic_present = true;
1098 vgic_present = false;
1105 * Init HYP architected timer support
1107 err = kvm_timer_hyp_init();
1112 kvm_coproc_table_init();
1117 static void teardown_hyp_mode(void)
1121 if (is_kernel_in_hyp_mode())
1125 for_each_possible_cpu(cpu)
1126 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1129 static int init_vhe_mode(void)
1132 * Execute the init code on each CPU.
1134 on_each_cpu(cpu_init_stage2, NULL, 1);
1136 /* set size of VMID supported by CPU */
1137 kvm_vmid_bits = kvm_get_vmid_bits();
1138 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1140 kvm_info("VHE mode initialized successfully\n");
1145 * Inits Hyp-mode on all online CPUs
1147 static int init_hyp_mode(void)
1153 * Allocate Hyp PGD and setup Hyp identity mapping
1155 err = kvm_mmu_init();
1160 * It is probably enough to obtain the default on one
1161 * CPU. It's unlikely to be different on the others.
1163 hyp_default_vectors = __hyp_get_vectors();
1166 * Allocate stack pages for Hypervisor-mode
1168 for_each_possible_cpu(cpu) {
1169 unsigned long stack_page;
1171 stack_page = __get_free_page(GFP_KERNEL);
1177 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1181 * Map the Hyp-code called directly from the host
1183 err = create_hyp_mappings(kvm_ksym_ref(__kvm_hyp_code_start),
1184 kvm_ksym_ref(__kvm_hyp_code_end));
1186 kvm_err("Cannot map world-switch code\n");
1190 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1191 kvm_ksym_ref(__end_rodata));
1193 kvm_err("Cannot map rodata section\n");
1198 * Map the Hyp stack pages
1200 for_each_possible_cpu(cpu) {
1201 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1202 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
1205 kvm_err("Cannot map hyp stack\n");
1210 for_each_possible_cpu(cpu) {
1211 kvm_cpu_context_t *cpu_ctxt;
1213 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1214 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1217 kvm_err("Cannot map host CPU state: %d\n", err);
1223 * Execute the init code on each CPU.
1225 on_each_cpu(cpu_init_hyp_mode, NULL, 1);
1227 #ifndef CONFIG_HOTPLUG_CPU
1228 free_boot_hyp_pgd();
1231 /* set size of VMID supported by CPU */
1232 kvm_vmid_bits = kvm_get_vmid_bits();
1233 kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1235 kvm_info("Hyp mode initialized successfully\n");
1240 teardown_hyp_mode();
1241 kvm_err("error initializing Hyp mode: %d\n", err);
1245 static void check_kvm_target_cpu(void *ret)
1247 *(int *)ret = kvm_target_cpu();
1250 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1252 struct kvm_vcpu *vcpu;
1255 mpidr &= MPIDR_HWID_BITMASK;
1256 kvm_for_each_vcpu(i, vcpu, kvm) {
1257 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1264 * Initialize Hyp-mode and memory mappings on all CPUs.
1266 int kvm_arch_init(void *opaque)
1271 if (!is_hyp_mode_available()) {
1272 kvm_err("HYP mode not available\n");
1276 for_each_online_cpu(cpu) {
1277 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1279 kvm_err("Error, CPU %d not supported!\n", cpu);
1284 err = init_common_resources();
1288 if (is_kernel_in_hyp_mode())
1289 err = init_vhe_mode();
1291 err = init_hyp_mode();
1295 err = init_subsystems();
1302 teardown_hyp_mode();
1304 teardown_common_resources();
1308 /* NOP: Compiling as a module not supported */
1309 void kvm_arch_exit(void)
1311 kvm_perf_teardown();
1314 static int arm_init(void)
1316 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1320 module_init(arm_init);