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>
31 #include <kvm/arm_pmu.h>
33 #define CREATE_TRACE_POINTS
36 #include <asm/uaccess.h>
37 #include <asm/ptrace.h>
39 #include <asm/tlbflush.h>
40 #include <asm/cacheflush.h>
42 #include <asm/kvm_arm.h>
43 #include <asm/kvm_asm.h>
44 #include <asm/kvm_mmu.h>
45 #include <asm/kvm_emulate.h>
46 #include <asm/kvm_coproc.h>
47 #include <asm/kvm_psci.h>
48 #include <asm/sections.h>
51 __asm__(".arch_extension virt");
54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
55 static kvm_cpu_context_t __percpu
*kvm_host_cpu_state
;
56 static unsigned long hyp_default_vectors
;
58 /* Per-CPU variable containing the currently running vcpu. */
59 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
61 /* The VMID used in the VTTBR */
62 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
63 static u32 kvm_next_vmid
;
64 static unsigned int kvm_vmid_bits __read_mostly
;
65 static DEFINE_SPINLOCK(kvm_vmid_lock
);
67 static bool vgic_present
;
69 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
71 BUG_ON(preemptible());
72 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
76 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
77 * Must be called from non-preemptible context
79 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
81 BUG_ON(preemptible());
82 return __this_cpu_read(kvm_arm_running_vcpu
);
86 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
88 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
90 return &kvm_arm_running_vcpu
;
93 int kvm_arch_hardware_enable(void)
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 ret
= kvm_alloc_stage2_pgd(kvm
);
129 ret
= create_hyp_mappings(kvm
, kvm
+ 1);
131 goto out_free_stage2_pgd
;
133 kvm_vgic_early_init(kvm
);
136 /* Mark the initial VMID generation invalid */
137 kvm
->arch
.vmid_gen
= 0;
139 /* The maximum number of VCPUs is limited by the host's GIC model */
140 kvm
->arch
.max_vcpus
= vgic_present
?
141 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
145 kvm_free_stage2_pgd(kvm
);
150 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
152 return VM_FAULT_SIGBUS
;
157 * kvm_arch_destroy_vm - destroy the VM data structure
158 * @kvm: pointer to the KVM struct
160 void kvm_arch_destroy_vm(struct kvm
*kvm
)
164 kvm_free_stage2_pgd(kvm
);
166 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
168 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
169 kvm
->vcpus
[i
] = NULL
;
173 kvm_vgic_destroy(kvm
);
176 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
180 case KVM_CAP_IRQCHIP
:
183 case KVM_CAP_IOEVENTFD
:
184 case KVM_CAP_DEVICE_CTRL
:
185 case KVM_CAP_USER_MEMORY
:
186 case KVM_CAP_SYNC_MMU
:
187 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
188 case KVM_CAP_ONE_REG
:
189 case KVM_CAP_ARM_PSCI
:
190 case KVM_CAP_ARM_PSCI_0_2
:
191 case KVM_CAP_READONLY_MEM
:
192 case KVM_CAP_MP_STATE
:
195 case KVM_CAP_COALESCED_MMIO
:
196 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
198 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
201 case KVM_CAP_NR_VCPUS
:
202 r
= num_online_cpus();
204 case KVM_CAP_MAX_VCPUS
:
208 r
= kvm_arch_dev_ioctl_check_extension(ext
);
214 long kvm_arch_dev_ioctl(struct file
*filp
,
215 unsigned int ioctl
, unsigned long arg
)
221 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
224 struct kvm_vcpu
*vcpu
;
226 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
231 if (id
>= kvm
->arch
.max_vcpus
) {
236 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
242 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
246 err
= create_hyp_mappings(vcpu
, vcpu
+ 1);
252 kvm_vcpu_uninit(vcpu
);
254 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
259 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
261 kvm_vgic_vcpu_early_init(vcpu
);
264 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
266 kvm_mmu_free_memory_caches(vcpu
);
267 kvm_timer_vcpu_terminate(vcpu
);
268 kvm_vgic_vcpu_destroy(vcpu
);
269 kvm_pmu_vcpu_destroy(vcpu
);
270 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
273 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
275 kvm_arch_vcpu_free(vcpu
);
278 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
280 return kvm_timer_should_fire(vcpu
);
283 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
285 kvm_timer_schedule(vcpu
);
288 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
290 kvm_timer_unschedule(vcpu
);
293 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
295 /* Force users to call KVM_ARM_VCPU_INIT */
296 vcpu
->arch
.target
= -1;
297 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
299 /* Set up the timer */
300 kvm_timer_vcpu_init(vcpu
);
302 kvm_arm_reset_debug_ptr(vcpu
);
307 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
310 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
312 kvm_arm_set_running_vcpu(vcpu
);
315 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
318 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
319 * if the vcpu is no longer assigned to a cpu. This is used for the
320 * optimized make_all_cpus_request path.
324 kvm_arm_set_running_vcpu(NULL
);
327 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
328 struct kvm_mp_state
*mp_state
)
330 if (vcpu
->arch
.power_off
)
331 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
333 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
338 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
339 struct kvm_mp_state
*mp_state
)
341 switch (mp_state
->mp_state
) {
342 case KVM_MP_STATE_RUNNABLE
:
343 vcpu
->arch
.power_off
= false;
345 case KVM_MP_STATE_STOPPED
:
346 vcpu
->arch
.power_off
= true;
356 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
357 * @v: The VCPU pointer
359 * If the guest CPU is not waiting for interrupts or an interrupt line is
360 * asserted, the CPU is by definition runnable.
362 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
364 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
365 && !v
->arch
.power_off
&& !v
->arch
.pause
);
368 /* Just ensure a guest exit from a particular CPU */
369 static void exit_vm_noop(void *info
)
373 void force_vm_exit(const cpumask_t
*mask
)
375 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
379 * need_new_vmid_gen - check that the VMID is still valid
380 * @kvm: The VM's VMID to checkt
382 * return true if there is a new generation of VMIDs being used
384 * The hardware supports only 256 values with the value zero reserved for the
385 * host, so we check if an assigned value belongs to a previous generation,
386 * which which requires us to assign a new value. If we're the first to use a
387 * VMID for the new generation, we must flush necessary caches and TLBs on all
390 static bool need_new_vmid_gen(struct kvm
*kvm
)
392 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
396 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
397 * @kvm The guest that we are about to run
399 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
400 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
403 static void update_vttbr(struct kvm
*kvm
)
405 phys_addr_t pgd_phys
;
408 if (!need_new_vmid_gen(kvm
))
411 spin_lock(&kvm_vmid_lock
);
414 * We need to re-check the vmid_gen here to ensure that if another vcpu
415 * already allocated a valid vmid for this vm, then this vcpu should
418 if (!need_new_vmid_gen(kvm
)) {
419 spin_unlock(&kvm_vmid_lock
);
423 /* First user of a new VMID generation? */
424 if (unlikely(kvm_next_vmid
== 0)) {
425 atomic64_inc(&kvm_vmid_gen
);
429 * On SMP we know no other CPUs can use this CPU's or each
430 * other's VMID after force_vm_exit returns since the
431 * kvm_vmid_lock blocks them from reentry to the guest.
433 force_vm_exit(cpu_all_mask
);
435 * Now broadcast TLB + ICACHE invalidation over the inner
436 * shareable domain to make sure all data structures are
439 kvm_call_hyp(__kvm_flush_vm_context
);
442 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
443 kvm
->arch
.vmid
= kvm_next_vmid
;
445 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
447 /* update vttbr to be used with the new vmid */
448 pgd_phys
= virt_to_phys(kvm_get_hwpgd(kvm
));
449 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
450 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
451 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
453 spin_unlock(&kvm_vmid_lock
);
456 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
458 struct kvm
*kvm
= vcpu
->kvm
;
461 if (likely(vcpu
->arch
.has_run_once
))
464 vcpu
->arch
.has_run_once
= true;
467 * Map the VGIC hardware resources before running a vcpu the first
470 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
471 ret
= kvm_vgic_map_resources(kvm
);
477 * Enable the arch timers only if we have an in-kernel VGIC
478 * and it has been properly initialized, since we cannot handle
479 * interrupts from the virtual timer with a userspace gic.
481 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
482 kvm_timer_enable(kvm
);
487 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
489 return vgic_initialized(kvm
);
492 static void kvm_arm_halt_guest(struct kvm
*kvm
) __maybe_unused
;
493 static void kvm_arm_resume_guest(struct kvm
*kvm
) __maybe_unused
;
495 static void kvm_arm_halt_guest(struct kvm
*kvm
)
498 struct kvm_vcpu
*vcpu
;
500 kvm_for_each_vcpu(i
, vcpu
, kvm
)
501 vcpu
->arch
.pause
= true;
502 force_vm_exit(cpu_all_mask
);
505 static void kvm_arm_resume_guest(struct kvm
*kvm
)
508 struct kvm_vcpu
*vcpu
;
510 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
511 wait_queue_head_t
*wq
= kvm_arch_vcpu_wq(vcpu
);
513 vcpu
->arch
.pause
= false;
514 wake_up_interruptible(wq
);
518 static void vcpu_sleep(struct kvm_vcpu
*vcpu
)
520 wait_queue_head_t
*wq
= kvm_arch_vcpu_wq(vcpu
);
522 wait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
523 (!vcpu
->arch
.pause
)));
526 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
528 return vcpu
->arch
.target
>= 0;
532 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
533 * @vcpu: The VCPU pointer
534 * @run: The kvm_run structure pointer used for userspace state exchange
536 * This function is called through the VCPU_RUN ioctl called from user space. It
537 * will execute VM code in a loop until the time slice for the process is used
538 * or some emulation is needed from user space in which case the function will
539 * return with return value 0 and with the kvm_run structure filled in with the
540 * required data for the requested emulation.
542 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
547 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
550 ret
= kvm_vcpu_first_run_init(vcpu
);
554 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
555 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
560 if (vcpu
->sigset_active
)
561 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
564 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
567 * Check conditions before entering the guest
571 update_vttbr(vcpu
->kvm
);
573 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
)
577 * Preparing the interrupts to be injected also
578 * involves poking the GIC, which must be done in a
579 * non-preemptible context.
582 kvm_pmu_flush_hwstate(vcpu
);
583 kvm_timer_flush_hwstate(vcpu
);
584 kvm_vgic_flush_hwstate(vcpu
);
589 * Re-check atomic conditions
591 if (signal_pending(current
)) {
593 run
->exit_reason
= KVM_EXIT_INTR
;
596 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
597 vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
599 kvm_pmu_sync_hwstate(vcpu
);
600 kvm_timer_sync_hwstate(vcpu
);
601 kvm_vgic_sync_hwstate(vcpu
);
606 kvm_arm_setup_debug(vcpu
);
608 /**************************************************************
611 trace_kvm_entry(*vcpu_pc(vcpu
));
613 vcpu
->mode
= IN_GUEST_MODE
;
615 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
617 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
621 *************************************************************/
623 kvm_arm_clear_debug(vcpu
);
626 * We may have taken a host interrupt in HYP mode (ie
627 * while executing the guest). This interrupt is still
628 * pending, as we haven't serviced it yet!
630 * We're now back in SVC mode, with interrupts
631 * disabled. Enabling the interrupts now will have
632 * the effect of taking the interrupt again, in SVC
638 * We do local_irq_enable() before calling kvm_guest_exit() so
639 * that if a timer interrupt hits while running the guest we
640 * account that tick as being spent in the guest. We enable
641 * preemption after calling kvm_guest_exit() so that if we get
642 * preempted we make sure ticks after that is not counted as
646 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
649 * We must sync the PMU and timer state before the vgic state so
650 * that the vgic can properly sample the updated state of the
653 kvm_pmu_sync_hwstate(vcpu
);
654 kvm_timer_sync_hwstate(vcpu
);
656 kvm_vgic_sync_hwstate(vcpu
);
660 ret
= handle_exit(vcpu
, run
, ret
);
663 if (vcpu
->sigset_active
)
664 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
668 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
674 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
675 bit_index
= __ffs(HCR_VI
);
676 else /* KVM_ARM_IRQ_CPU_FIQ */
677 bit_index
= __ffs(HCR_VF
);
679 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
681 set
= test_and_set_bit(bit_index
, ptr
);
683 set
= test_and_clear_bit(bit_index
, ptr
);
686 * If we didn't change anything, no need to wake up or kick other CPUs
692 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
693 * trigger a world-switch round on the running physical CPU to set the
694 * virtual IRQ/FIQ fields in the HCR appropriately.
701 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
704 u32 irq
= irq_level
->irq
;
705 unsigned int irq_type
, vcpu_idx
, irq_num
;
706 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
707 struct kvm_vcpu
*vcpu
= NULL
;
708 bool level
= irq_level
->level
;
710 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
711 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
712 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
714 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
717 case KVM_ARM_IRQ_TYPE_CPU
:
718 if (irqchip_in_kernel(kvm
))
721 if (vcpu_idx
>= nrcpus
)
724 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
728 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
731 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
732 case KVM_ARM_IRQ_TYPE_PPI
:
733 if (!irqchip_in_kernel(kvm
))
736 if (vcpu_idx
>= nrcpus
)
739 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
743 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
746 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
747 case KVM_ARM_IRQ_TYPE_SPI
:
748 if (!irqchip_in_kernel(kvm
))
751 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
754 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
760 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
761 const struct kvm_vcpu_init
*init
)
764 int phys_target
= kvm_target_cpu();
766 if (init
->target
!= phys_target
)
770 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
771 * use the same target.
773 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
776 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
777 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
778 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
780 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
784 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
785 * use the same feature set.
787 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
788 test_bit(i
, vcpu
->arch
.features
) != set
)
792 set_bit(i
, vcpu
->arch
.features
);
795 vcpu
->arch
.target
= phys_target
;
797 /* Now we know what it is, we can reset it. */
798 return kvm_reset_vcpu(vcpu
);
802 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
803 struct kvm_vcpu_init
*init
)
807 ret
= kvm_vcpu_set_target(vcpu
, init
);
812 * Ensure a rebooted VM will fault in RAM pages and detect if the
813 * guest MMU is turned off and flush the caches as needed.
815 if (vcpu
->arch
.has_run_once
)
816 stage2_unmap_vm(vcpu
->kvm
);
818 vcpu_reset_hcr(vcpu
);
821 * Handle the "start in power-off" case.
823 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
824 vcpu
->arch
.power_off
= true;
826 vcpu
->arch
.power_off
= false;
831 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
832 struct kvm_device_attr
*attr
)
836 switch (attr
->group
) {
844 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
845 struct kvm_device_attr
*attr
)
849 switch (attr
->group
) {
857 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
858 struct kvm_device_attr
*attr
)
862 switch (attr
->group
) {
870 long kvm_arch_vcpu_ioctl(struct file
*filp
,
871 unsigned int ioctl
, unsigned long arg
)
873 struct kvm_vcpu
*vcpu
= filp
->private_data
;
874 void __user
*argp
= (void __user
*)arg
;
875 struct kvm_device_attr attr
;
878 case KVM_ARM_VCPU_INIT
: {
879 struct kvm_vcpu_init init
;
881 if (copy_from_user(&init
, argp
, sizeof(init
)))
884 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
886 case KVM_SET_ONE_REG
:
887 case KVM_GET_ONE_REG
: {
888 struct kvm_one_reg reg
;
890 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
893 if (copy_from_user(®
, argp
, sizeof(reg
)))
895 if (ioctl
== KVM_SET_ONE_REG
)
896 return kvm_arm_set_reg(vcpu
, ®
);
898 return kvm_arm_get_reg(vcpu
, ®
);
900 case KVM_GET_REG_LIST
: {
901 struct kvm_reg_list __user
*user_list
= argp
;
902 struct kvm_reg_list reg_list
;
905 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
908 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
911 reg_list
.n
= kvm_arm_num_regs(vcpu
);
912 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
916 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
918 case KVM_SET_DEVICE_ATTR
: {
919 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
921 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
923 case KVM_GET_DEVICE_ATTR
: {
924 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
926 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
928 case KVM_HAS_DEVICE_ATTR
: {
929 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
931 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
939 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
941 * @log: slot id and address to which we copy the log
943 * Steps 1-4 below provide general overview of dirty page logging. See
944 * kvm_get_dirty_log_protect() function description for additional details.
946 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
947 * always flush the TLB (step 4) even if previous step failed and the dirty
948 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
949 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
950 * writes will be marked dirty for next log read.
952 * 1. Take a snapshot of the bit and clear it if needed.
953 * 2. Write protect the corresponding page.
954 * 3. Copy the snapshot to the userspace.
955 * 4. Flush TLB's if needed.
957 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
959 bool is_dirty
= false;
962 mutex_lock(&kvm
->slots_lock
);
964 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
967 kvm_flush_remote_tlbs(kvm
);
969 mutex_unlock(&kvm
->slots_lock
);
973 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
974 struct kvm_arm_device_addr
*dev_addr
)
976 unsigned long dev_id
, type
;
978 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
979 KVM_ARM_DEVICE_ID_SHIFT
;
980 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
981 KVM_ARM_DEVICE_TYPE_SHIFT
;
984 case KVM_ARM_DEVICE_VGIC_V2
:
987 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
993 long kvm_arch_vm_ioctl(struct file
*filp
,
994 unsigned int ioctl
, unsigned long arg
)
996 struct kvm
*kvm
= filp
->private_data
;
997 void __user
*argp
= (void __user
*)arg
;
1000 case KVM_CREATE_IRQCHIP
: {
1003 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1005 case KVM_ARM_SET_DEVICE_ADDR
: {
1006 struct kvm_arm_device_addr dev_addr
;
1008 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1010 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1012 case KVM_ARM_PREFERRED_TARGET
: {
1014 struct kvm_vcpu_init init
;
1016 err
= kvm_vcpu_preferred_target(&init
);
1020 if (copy_to_user(argp
, &init
, sizeof(init
)))
1030 static void cpu_init_stage2(void *dummy
)
1032 __cpu_init_stage2();
1035 static void cpu_init_hyp_mode(void *dummy
)
1037 phys_addr_t boot_pgd_ptr
;
1038 phys_addr_t pgd_ptr
;
1039 unsigned long hyp_stack_ptr
;
1040 unsigned long stack_page
;
1041 unsigned long vector_ptr
;
1043 /* Switch from the HYP stub to our own HYP init vector */
1044 __hyp_set_vectors(kvm_get_idmap_vector());
1046 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
1047 pgd_ptr
= kvm_mmu_get_httbr();
1048 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1049 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1050 vector_ptr
= (unsigned long)__kvm_hyp_vector
;
1052 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1053 __cpu_init_stage2();
1055 kvm_arm_init_debug();
1058 static int hyp_init_cpu_notify(struct notifier_block
*self
,
1059 unsigned long action
, void *cpu
)
1063 case CPU_STARTING_FROZEN
:
1064 if (__hyp_get_vectors() == hyp_default_vectors
)
1065 cpu_init_hyp_mode(NULL
);
1072 static struct notifier_block hyp_init_cpu_nb
= {
1073 .notifier_call
= hyp_init_cpu_notify
,
1076 #ifdef CONFIG_CPU_PM
1077 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1081 if (cmd
== CPU_PM_EXIT
&&
1082 __hyp_get_vectors() == hyp_default_vectors
) {
1083 cpu_init_hyp_mode(NULL
);
1090 static struct notifier_block hyp_init_cpu_pm_nb
= {
1091 .notifier_call
= hyp_init_cpu_pm_notifier
,
1094 static void __init
hyp_cpu_pm_init(void)
1096 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1099 static inline void hyp_cpu_pm_init(void)
1104 static void teardown_common_resources(void)
1106 free_percpu(kvm_host_cpu_state
);
1109 static int init_common_resources(void)
1111 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1112 if (!kvm_host_cpu_state
) {
1113 kvm_err("Cannot allocate host CPU state\n");
1120 static int init_subsystems(void)
1125 * Init HYP view of VGIC
1127 err
= kvm_vgic_hyp_init();
1130 vgic_present
= true;
1134 vgic_present
= false;
1141 * Init HYP architected timer support
1143 err
= kvm_timer_hyp_init();
1148 kvm_coproc_table_init();
1153 static void teardown_hyp_mode(void)
1157 if (is_kernel_in_hyp_mode())
1161 for_each_possible_cpu(cpu
)
1162 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1165 static int init_vhe_mode(void)
1168 * Execute the init code on each CPU.
1170 on_each_cpu(cpu_init_stage2
, NULL
, 1);
1172 /* set size of VMID supported by CPU */
1173 kvm_vmid_bits
= kvm_get_vmid_bits();
1174 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1176 kvm_info("VHE mode initialized successfully\n");
1181 * Inits Hyp-mode on all online CPUs
1183 static int init_hyp_mode(void)
1189 * Allocate Hyp PGD and setup Hyp identity mapping
1191 err
= kvm_mmu_init();
1196 * It is probably enough to obtain the default on one
1197 * CPU. It's unlikely to be different on the others.
1199 hyp_default_vectors
= __hyp_get_vectors();
1202 * Allocate stack pages for Hypervisor-mode
1204 for_each_possible_cpu(cpu
) {
1205 unsigned long stack_page
;
1207 stack_page
= __get_free_page(GFP_KERNEL
);
1213 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1217 * Map the Hyp-code called directly from the host
1219 err
= create_hyp_mappings(__hyp_text_start
, __hyp_text_end
);
1221 kvm_err("Cannot map world-switch code\n");
1225 err
= create_hyp_mappings(__start_rodata
, __end_rodata
);
1227 kvm_err("Cannot map rodata section\n");
1232 * Map the Hyp stack pages
1234 for_each_possible_cpu(cpu
) {
1235 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1236 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
1239 kvm_err("Cannot map hyp stack\n");
1244 for_each_possible_cpu(cpu
) {
1245 kvm_cpu_context_t
*cpu_ctxt
;
1247 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1248 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1251 kvm_err("Cannot map host CPU state: %d\n", err
);
1257 * Execute the init code on each CPU.
1259 on_each_cpu(cpu_init_hyp_mode
, NULL
, 1);
1261 #ifndef CONFIG_HOTPLUG_CPU
1262 free_boot_hyp_pgd();
1265 cpu_notifier_register_begin();
1267 err
= __register_cpu_notifier(&hyp_init_cpu_nb
);
1269 cpu_notifier_register_done();
1272 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err
);
1278 /* set size of VMID supported by CPU */
1279 kvm_vmid_bits
= kvm_get_vmid_bits();
1280 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1282 kvm_info("Hyp mode initialized successfully\n");
1287 teardown_hyp_mode();
1288 kvm_err("error initializing Hyp mode: %d\n", err
);
1292 static void check_kvm_target_cpu(void *ret
)
1294 *(int *)ret
= kvm_target_cpu();
1297 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1299 struct kvm_vcpu
*vcpu
;
1302 mpidr
&= MPIDR_HWID_BITMASK
;
1303 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1304 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1311 * Initialize Hyp-mode and memory mappings on all CPUs.
1313 int kvm_arch_init(void *opaque
)
1318 if (!is_hyp_mode_available()) {
1319 kvm_err("HYP mode not available\n");
1323 for_each_online_cpu(cpu
) {
1324 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1326 kvm_err("Error, CPU %d not supported!\n", cpu
);
1331 err
= init_common_resources();
1335 if (is_kernel_in_hyp_mode())
1336 err
= init_vhe_mode();
1338 err
= init_hyp_mode();
1342 err
= init_subsystems();
1349 teardown_hyp_mode();
1351 teardown_common_resources();
1355 /* NOP: Compiling as a module not supported */
1356 void kvm_arch_exit(void)
1358 kvm_perf_teardown();
1361 static int arm_init(void)
1363 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1367 module_init(arm_init
);