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>
33 #define CREATE_TRACE_POINTS
36 #include <linux/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 DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled
);
71 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
73 BUG_ON(preemptible());
74 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
78 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
79 * Must be called from non-preemptible context
81 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
83 BUG_ON(preemptible());
84 return __this_cpu_read(kvm_arm_running_vcpu
);
88 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
90 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
92 return &kvm_arm_running_vcpu
;
95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
97 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
100 int kvm_arch_hardware_setup(void)
105 void kvm_arch_check_processor_compat(void *rtn
)
112 * kvm_arch_init_vm - initializes a VM data structure
113 * @kvm: pointer to the KVM struct
115 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
122 kvm
->arch
.last_vcpu_ran
= alloc_percpu(typeof(*kvm
->arch
.last_vcpu_ran
));
123 if (!kvm
->arch
.last_vcpu_ran
)
126 for_each_possible_cpu(cpu
)
127 *per_cpu_ptr(kvm
->arch
.last_vcpu_ran
, cpu
) = -1;
129 ret
= kvm_alloc_stage2_pgd(kvm
);
133 ret
= create_hyp_mappings(kvm
, kvm
+ 1, PAGE_HYP
);
135 goto out_free_stage2_pgd
;
137 kvm_vgic_early_init(kvm
);
139 /* Mark the initial VMID generation invalid */
140 kvm
->arch
.vmid_gen
= 0;
142 /* The maximum number of VCPUs is limited by the host's GIC model */
143 kvm
->arch
.max_vcpus
= vgic_present
?
144 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
148 kvm_free_stage2_pgd(kvm
);
150 free_percpu(kvm
->arch
.last_vcpu_ran
);
151 kvm
->arch
.last_vcpu_ran
= NULL
;
155 bool kvm_arch_has_vcpu_debugfs(void)
160 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
165 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
167 return VM_FAULT_SIGBUS
;
172 * kvm_arch_destroy_vm - destroy the VM data structure
173 * @kvm: pointer to the KVM struct
175 void kvm_arch_destroy_vm(struct kvm
*kvm
)
179 free_percpu(kvm
->arch
.last_vcpu_ran
);
180 kvm
->arch
.last_vcpu_ran
= NULL
;
182 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
184 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
185 kvm
->vcpus
[i
] = NULL
;
189 kvm_vgic_destroy(kvm
);
192 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
196 case KVM_CAP_IRQCHIP
:
199 case KVM_CAP_IOEVENTFD
:
200 case KVM_CAP_DEVICE_CTRL
:
201 case KVM_CAP_USER_MEMORY
:
202 case KVM_CAP_SYNC_MMU
:
203 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
204 case KVM_CAP_ONE_REG
:
205 case KVM_CAP_ARM_PSCI
:
206 case KVM_CAP_ARM_PSCI_0_2
:
207 case KVM_CAP_READONLY_MEM
:
208 case KVM_CAP_MP_STATE
:
209 case KVM_CAP_IMMEDIATE_EXIT
:
212 case KVM_CAP_COALESCED_MMIO
:
213 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
215 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
218 case KVM_CAP_NR_VCPUS
:
219 r
= num_online_cpus();
221 case KVM_CAP_MAX_VCPUS
:
224 case KVM_CAP_NR_MEMSLOTS
:
225 r
= KVM_USER_MEM_SLOTS
;
227 case KVM_CAP_MSI_DEVID
:
231 r
= kvm
->arch
.vgic
.msis_require_devid
;
234 r
= kvm_arch_dev_ioctl_check_extension(kvm
, ext
);
240 long kvm_arch_dev_ioctl(struct file
*filp
,
241 unsigned int ioctl
, unsigned long arg
)
247 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
250 struct kvm_vcpu
*vcpu
;
252 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
257 if (id
>= kvm
->arch
.max_vcpus
) {
262 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
268 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
272 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
278 kvm_vcpu_uninit(vcpu
);
280 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
285 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
287 kvm_vgic_vcpu_early_init(vcpu
);
290 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
292 kvm_mmu_free_memory_caches(vcpu
);
293 kvm_timer_vcpu_terminate(vcpu
);
294 kvm_vgic_vcpu_destroy(vcpu
);
295 kvm_pmu_vcpu_destroy(vcpu
);
296 kvm_vcpu_uninit(vcpu
);
297 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
300 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
302 kvm_arch_vcpu_free(vcpu
);
305 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
307 return kvm_timer_should_fire(vcpu_vtimer(vcpu
)) ||
308 kvm_timer_should_fire(vcpu_ptimer(vcpu
));
311 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
313 kvm_timer_schedule(vcpu
);
316 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
318 kvm_timer_unschedule(vcpu
);
321 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
323 /* Force users to call KVM_ARM_VCPU_INIT */
324 vcpu
->arch
.target
= -1;
325 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
327 /* Set up the timer */
328 kvm_timer_vcpu_init(vcpu
);
330 kvm_arm_reset_debug_ptr(vcpu
);
335 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
339 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
342 * We might get preempted before the vCPU actually runs, but
343 * over-invalidation doesn't affect correctness.
345 if (*last_ran
!= vcpu
->vcpu_id
) {
346 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
347 *last_ran
= vcpu
->vcpu_id
;
351 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
353 kvm_arm_set_running_vcpu(vcpu
);
356 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
359 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
360 * if the vcpu is no longer assigned to a cpu. This is used for the
361 * optimized make_all_cpus_request path.
365 kvm_arm_set_running_vcpu(NULL
);
366 kvm_timer_vcpu_put(vcpu
);
369 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
370 struct kvm_mp_state
*mp_state
)
372 if (vcpu
->arch
.power_off
)
373 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
375 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
380 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
381 struct kvm_mp_state
*mp_state
)
383 switch (mp_state
->mp_state
) {
384 case KVM_MP_STATE_RUNNABLE
:
385 vcpu
->arch
.power_off
= false;
387 case KVM_MP_STATE_STOPPED
:
388 vcpu
->arch
.power_off
= true;
398 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
399 * @v: The VCPU pointer
401 * If the guest CPU is not waiting for interrupts or an interrupt line is
402 * asserted, the CPU is by definition runnable.
404 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
406 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
407 && !v
->arch
.power_off
&& !v
->arch
.pause
);
410 /* Just ensure a guest exit from a particular CPU */
411 static void exit_vm_noop(void *info
)
415 void force_vm_exit(const cpumask_t
*mask
)
418 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
423 * need_new_vmid_gen - check that the VMID is still valid
424 * @kvm: The VM's VMID to check
426 * return true if there is a new generation of VMIDs being used
428 * The hardware supports only 256 values with the value zero reserved for the
429 * host, so we check if an assigned value belongs to a previous generation,
430 * which which requires us to assign a new value. If we're the first to use a
431 * VMID for the new generation, we must flush necessary caches and TLBs on all
434 static bool need_new_vmid_gen(struct kvm
*kvm
)
436 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
440 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
441 * @kvm The guest that we are about to run
443 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
444 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
447 static void update_vttbr(struct kvm
*kvm
)
449 phys_addr_t pgd_phys
;
452 if (!need_new_vmid_gen(kvm
))
455 spin_lock(&kvm_vmid_lock
);
458 * We need to re-check the vmid_gen here to ensure that if another vcpu
459 * already allocated a valid vmid for this vm, then this vcpu should
462 if (!need_new_vmid_gen(kvm
)) {
463 spin_unlock(&kvm_vmid_lock
);
467 /* First user of a new VMID generation? */
468 if (unlikely(kvm_next_vmid
== 0)) {
469 atomic64_inc(&kvm_vmid_gen
);
473 * On SMP we know no other CPUs can use this CPU's or each
474 * other's VMID after force_vm_exit returns since the
475 * kvm_vmid_lock blocks them from reentry to the guest.
477 force_vm_exit(cpu_all_mask
);
479 * Now broadcast TLB + ICACHE invalidation over the inner
480 * shareable domain to make sure all data structures are
483 kvm_call_hyp(__kvm_flush_vm_context
);
486 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
487 kvm
->arch
.vmid
= kvm_next_vmid
;
489 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
491 /* update vttbr to be used with the new vmid */
492 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
493 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
494 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
495 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
497 spin_unlock(&kvm_vmid_lock
);
500 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
502 struct kvm
*kvm
= vcpu
->kvm
;
505 if (likely(vcpu
->arch
.has_run_once
))
508 vcpu
->arch
.has_run_once
= true;
511 * Map the VGIC hardware resources before running a vcpu the first
514 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
515 ret
= kvm_vgic_map_resources(kvm
);
521 * Enable the arch timers only if we have an in-kernel VGIC
522 * and it has been properly initialized, since we cannot handle
523 * interrupts from the virtual timer with a userspace gic.
525 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
526 ret
= kvm_timer_enable(vcpu
);
531 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
533 return vgic_initialized(kvm
);
536 void kvm_arm_halt_guest(struct kvm
*kvm
)
539 struct kvm_vcpu
*vcpu
;
541 kvm_for_each_vcpu(i
, vcpu
, kvm
)
542 vcpu
->arch
.pause
= true;
543 kvm_make_all_cpus_request(kvm
, KVM_REQ_VCPU_EXIT
);
546 void kvm_arm_halt_vcpu(struct kvm_vcpu
*vcpu
)
548 vcpu
->arch
.pause
= true;
552 void kvm_arm_resume_vcpu(struct kvm_vcpu
*vcpu
)
554 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
556 vcpu
->arch
.pause
= false;
560 void kvm_arm_resume_guest(struct kvm
*kvm
)
563 struct kvm_vcpu
*vcpu
;
565 kvm_for_each_vcpu(i
, vcpu
, kvm
)
566 kvm_arm_resume_vcpu(vcpu
);
569 static void vcpu_sleep(struct kvm_vcpu
*vcpu
)
571 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
573 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
574 (!vcpu
->arch
.pause
)));
577 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
579 return vcpu
->arch
.target
>= 0;
583 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
584 * @vcpu: The VCPU pointer
585 * @run: The kvm_run structure pointer used for userspace state exchange
587 * This function is called through the VCPU_RUN ioctl called from user space. It
588 * will execute VM code in a loop until the time slice for the process is used
589 * or some emulation is needed from user space in which case the function will
590 * return with return value 0 and with the kvm_run structure filled in with the
591 * required data for the requested emulation.
593 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
598 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
601 ret
= kvm_vcpu_first_run_init(vcpu
);
605 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
606 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
611 if (run
->immediate_exit
)
614 if (vcpu
->sigset_active
)
615 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
618 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
621 * Check conditions before entering the guest
625 update_vttbr(vcpu
->kvm
);
627 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
)
631 * Preparing the interrupts to be injected also
632 * involves poking the GIC, which must be done in a
633 * non-preemptible context.
636 kvm_pmu_flush_hwstate(vcpu
);
637 kvm_timer_flush_hwstate(vcpu
);
638 kvm_vgic_flush_hwstate(vcpu
);
643 * Re-check atomic conditions
645 if (signal_pending(current
)) {
647 run
->exit_reason
= KVM_EXIT_INTR
;
650 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
651 vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
653 kvm_pmu_sync_hwstate(vcpu
);
654 kvm_timer_sync_hwstate(vcpu
);
655 kvm_vgic_sync_hwstate(vcpu
);
660 kvm_arm_setup_debug(vcpu
);
662 /**************************************************************
665 trace_kvm_entry(*vcpu_pc(vcpu
));
666 guest_enter_irqoff();
667 vcpu
->mode
= IN_GUEST_MODE
;
669 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
671 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
675 *************************************************************/
677 kvm_arm_clear_debug(vcpu
);
680 * We may have taken a host interrupt in HYP mode (ie
681 * while executing the guest). This interrupt is still
682 * pending, as we haven't serviced it yet!
684 * We're now back in SVC mode, with interrupts
685 * disabled. Enabling the interrupts now will have
686 * the effect of taking the interrupt again, in SVC
692 * We do local_irq_enable() before calling guest_exit() so
693 * that if a timer interrupt hits while running the guest we
694 * account that tick as being spent in the guest. We enable
695 * preemption after calling guest_exit() so that if we get
696 * preempted we make sure ticks after that is not counted as
700 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
703 * We must sync the PMU and timer state before the vgic state so
704 * that the vgic can properly sample the updated state of the
707 kvm_pmu_sync_hwstate(vcpu
);
708 kvm_timer_sync_hwstate(vcpu
);
710 kvm_vgic_sync_hwstate(vcpu
);
714 ret
= handle_exit(vcpu
, run
, ret
);
717 if (vcpu
->sigset_active
)
718 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
722 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
728 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
729 bit_index
= __ffs(HCR_VI
);
730 else /* KVM_ARM_IRQ_CPU_FIQ */
731 bit_index
= __ffs(HCR_VF
);
733 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
735 set
= test_and_set_bit(bit_index
, ptr
);
737 set
= test_and_clear_bit(bit_index
, ptr
);
740 * If we didn't change anything, no need to wake up or kick other CPUs
746 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
747 * trigger a world-switch round on the running physical CPU to set the
748 * virtual IRQ/FIQ fields in the HCR appropriately.
755 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
758 u32 irq
= irq_level
->irq
;
759 unsigned int irq_type
, vcpu_idx
, irq_num
;
760 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
761 struct kvm_vcpu
*vcpu
= NULL
;
762 bool level
= irq_level
->level
;
764 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
765 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
766 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
768 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
771 case KVM_ARM_IRQ_TYPE_CPU
:
772 if (irqchip_in_kernel(kvm
))
775 if (vcpu_idx
>= nrcpus
)
778 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
782 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
785 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
786 case KVM_ARM_IRQ_TYPE_PPI
:
787 if (!irqchip_in_kernel(kvm
))
790 if (vcpu_idx
>= nrcpus
)
793 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
797 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
800 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
801 case KVM_ARM_IRQ_TYPE_SPI
:
802 if (!irqchip_in_kernel(kvm
))
805 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
808 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
814 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
815 const struct kvm_vcpu_init
*init
)
818 int phys_target
= kvm_target_cpu();
820 if (init
->target
!= phys_target
)
824 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
825 * use the same target.
827 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
830 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
831 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
832 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
834 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
838 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
839 * use the same feature set.
841 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
842 test_bit(i
, vcpu
->arch
.features
) != set
)
846 set_bit(i
, vcpu
->arch
.features
);
849 vcpu
->arch
.target
= phys_target
;
851 /* Now we know what it is, we can reset it. */
852 return kvm_reset_vcpu(vcpu
);
856 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
857 struct kvm_vcpu_init
*init
)
861 ret
= kvm_vcpu_set_target(vcpu
, init
);
866 * Ensure a rebooted VM will fault in RAM pages and detect if the
867 * guest MMU is turned off and flush the caches as needed.
869 if (vcpu
->arch
.has_run_once
)
870 stage2_unmap_vm(vcpu
->kvm
);
872 vcpu_reset_hcr(vcpu
);
875 * Handle the "start in power-off" case.
877 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
878 vcpu
->arch
.power_off
= true;
880 vcpu
->arch
.power_off
= false;
885 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
886 struct kvm_device_attr
*attr
)
890 switch (attr
->group
) {
892 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
899 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
900 struct kvm_device_attr
*attr
)
904 switch (attr
->group
) {
906 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
913 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
914 struct kvm_device_attr
*attr
)
918 switch (attr
->group
) {
920 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
927 long kvm_arch_vcpu_ioctl(struct file
*filp
,
928 unsigned int ioctl
, unsigned long arg
)
930 struct kvm_vcpu
*vcpu
= filp
->private_data
;
931 void __user
*argp
= (void __user
*)arg
;
932 struct kvm_device_attr attr
;
935 case KVM_ARM_VCPU_INIT
: {
936 struct kvm_vcpu_init init
;
938 if (copy_from_user(&init
, argp
, sizeof(init
)))
941 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
943 case KVM_SET_ONE_REG
:
944 case KVM_GET_ONE_REG
: {
945 struct kvm_one_reg reg
;
947 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
950 if (copy_from_user(®
, argp
, sizeof(reg
)))
952 if (ioctl
== KVM_SET_ONE_REG
)
953 return kvm_arm_set_reg(vcpu
, ®
);
955 return kvm_arm_get_reg(vcpu
, ®
);
957 case KVM_GET_REG_LIST
: {
958 struct kvm_reg_list __user
*user_list
= argp
;
959 struct kvm_reg_list reg_list
;
962 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
965 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
968 reg_list
.n
= kvm_arm_num_regs(vcpu
);
969 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
973 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
975 case KVM_SET_DEVICE_ATTR
: {
976 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
978 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
980 case KVM_GET_DEVICE_ATTR
: {
981 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
983 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
985 case KVM_HAS_DEVICE_ATTR
: {
986 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
988 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
996 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
998 * @log: slot id and address to which we copy the log
1000 * Steps 1-4 below provide general overview of dirty page logging. See
1001 * kvm_get_dirty_log_protect() function description for additional details.
1003 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1004 * always flush the TLB (step 4) even if previous step failed and the dirty
1005 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1006 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1007 * writes will be marked dirty for next log read.
1009 * 1. Take a snapshot of the bit and clear it if needed.
1010 * 2. Write protect the corresponding page.
1011 * 3. Copy the snapshot to the userspace.
1012 * 4. Flush TLB's if needed.
1014 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1016 bool is_dirty
= false;
1019 mutex_lock(&kvm
->slots_lock
);
1021 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
1024 kvm_flush_remote_tlbs(kvm
);
1026 mutex_unlock(&kvm
->slots_lock
);
1030 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1031 struct kvm_arm_device_addr
*dev_addr
)
1033 unsigned long dev_id
, type
;
1035 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1036 KVM_ARM_DEVICE_ID_SHIFT
;
1037 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1038 KVM_ARM_DEVICE_TYPE_SHIFT
;
1041 case KVM_ARM_DEVICE_VGIC_V2
:
1044 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1050 long kvm_arch_vm_ioctl(struct file
*filp
,
1051 unsigned int ioctl
, unsigned long arg
)
1053 struct kvm
*kvm
= filp
->private_data
;
1054 void __user
*argp
= (void __user
*)arg
;
1057 case KVM_CREATE_IRQCHIP
: {
1061 mutex_lock(&kvm
->lock
);
1062 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1063 mutex_unlock(&kvm
->lock
);
1066 case KVM_ARM_SET_DEVICE_ADDR
: {
1067 struct kvm_arm_device_addr dev_addr
;
1069 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1071 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1073 case KVM_ARM_PREFERRED_TARGET
: {
1075 struct kvm_vcpu_init init
;
1077 err
= kvm_vcpu_preferred_target(&init
);
1081 if (copy_to_user(argp
, &init
, sizeof(init
)))
1091 static void cpu_init_hyp_mode(void *dummy
)
1093 phys_addr_t pgd_ptr
;
1094 unsigned long hyp_stack_ptr
;
1095 unsigned long stack_page
;
1096 unsigned long vector_ptr
;
1098 /* Switch from the HYP stub to our own HYP init vector */
1099 __hyp_set_vectors(kvm_get_idmap_vector());
1101 pgd_ptr
= kvm_mmu_get_httbr();
1102 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1103 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1104 vector_ptr
= (unsigned long)kvm_ksym_ref(__kvm_hyp_vector
);
1106 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1107 __cpu_init_stage2();
1109 if (is_kernel_in_hyp_mode())
1110 kvm_timer_init_vhe();
1112 kvm_arm_init_debug();
1115 static void cpu_hyp_reinit(void)
1117 if (is_kernel_in_hyp_mode()) {
1119 * __cpu_init_stage2() is safe to call even if the PM
1120 * event was cancelled before the CPU was reset.
1122 __cpu_init_stage2();
1124 if (__hyp_get_vectors() == hyp_default_vectors
)
1125 cpu_init_hyp_mode(NULL
);
1129 static void cpu_hyp_reset(void)
1131 if (!is_kernel_in_hyp_mode())
1132 __cpu_reset_hyp_mode(hyp_default_vectors
,
1133 kvm_get_idmap_start());
1136 static void _kvm_arch_hardware_enable(void *discard
)
1138 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1140 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1144 int kvm_arch_hardware_enable(void)
1146 _kvm_arch_hardware_enable(NULL
);
1150 static void _kvm_arch_hardware_disable(void *discard
)
1152 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1154 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1158 void kvm_arch_hardware_disable(void)
1160 _kvm_arch_hardware_disable(NULL
);
1163 #ifdef CONFIG_CPU_PM
1164 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1169 * kvm_arm_hardware_enabled is left with its old value over
1170 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1175 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1177 * don't update kvm_arm_hardware_enabled here
1178 * so that the hardware will be re-enabled
1179 * when we resume. See below.
1185 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1186 /* The hardware was enabled before suspend. */
1196 static struct notifier_block hyp_init_cpu_pm_nb
= {
1197 .notifier_call
= hyp_init_cpu_pm_notifier
,
1200 static void __init
hyp_cpu_pm_init(void)
1202 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1204 static void __init
hyp_cpu_pm_exit(void)
1206 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1209 static inline void hyp_cpu_pm_init(void)
1212 static inline void hyp_cpu_pm_exit(void)
1217 static void teardown_common_resources(void)
1219 free_percpu(kvm_host_cpu_state
);
1222 static int init_common_resources(void)
1224 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1225 if (!kvm_host_cpu_state
) {
1226 kvm_err("Cannot allocate host CPU state\n");
1230 /* set size of VMID supported by CPU */
1231 kvm_vmid_bits
= kvm_get_vmid_bits();
1232 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1237 static int init_subsystems(void)
1242 * Enable hardware so that subsystem initialisation can access EL2.
1244 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1247 * Register CPU lower-power notifier
1252 * Init HYP view of VGIC
1254 err
= kvm_vgic_hyp_init();
1257 vgic_present
= true;
1261 vgic_present
= false;
1269 * Init HYP architected timer support
1271 err
= kvm_timer_hyp_init();
1276 kvm_coproc_table_init();
1279 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1284 static void teardown_hyp_mode(void)
1288 if (is_kernel_in_hyp_mode())
1292 for_each_possible_cpu(cpu
)
1293 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1297 static int init_vhe_mode(void)
1299 kvm_info("VHE mode initialized successfully\n");
1304 * Inits Hyp-mode on all online CPUs
1306 static int init_hyp_mode(void)
1312 * Allocate Hyp PGD and setup Hyp identity mapping
1314 err
= kvm_mmu_init();
1319 * It is probably enough to obtain the default on one
1320 * CPU. It's unlikely to be different on the others.
1322 hyp_default_vectors
= __hyp_get_vectors();
1325 * Allocate stack pages for Hypervisor-mode
1327 for_each_possible_cpu(cpu
) {
1328 unsigned long stack_page
;
1330 stack_page
= __get_free_page(GFP_KERNEL
);
1336 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1340 * Map the Hyp-code called directly from the host
1342 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1343 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1345 kvm_err("Cannot map world-switch code\n");
1349 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1350 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1352 kvm_err("Cannot map rodata section\n");
1356 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1357 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1359 kvm_err("Cannot map bss section\n");
1364 * Map the Hyp stack pages
1366 for_each_possible_cpu(cpu
) {
1367 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1368 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1372 kvm_err("Cannot map hyp stack\n");
1377 for_each_possible_cpu(cpu
) {
1378 kvm_cpu_context_t
*cpu_ctxt
;
1380 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1381 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1384 kvm_err("Cannot map host CPU state: %d\n", err
);
1389 kvm_info("Hyp mode initialized successfully\n");
1394 teardown_hyp_mode();
1395 kvm_err("error initializing Hyp mode: %d\n", err
);
1399 static void check_kvm_target_cpu(void *ret
)
1401 *(int *)ret
= kvm_target_cpu();
1404 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1406 struct kvm_vcpu
*vcpu
;
1409 mpidr
&= MPIDR_HWID_BITMASK
;
1410 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1411 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1418 * Initialize Hyp-mode and memory mappings on all CPUs.
1420 int kvm_arch_init(void *opaque
)
1425 if (!is_hyp_mode_available()) {
1426 kvm_err("HYP mode not available\n");
1430 for_each_online_cpu(cpu
) {
1431 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1433 kvm_err("Error, CPU %d not supported!\n", cpu
);
1438 err
= init_common_resources();
1442 if (is_kernel_in_hyp_mode())
1443 err
= init_vhe_mode();
1445 err
= init_hyp_mode();
1449 err
= init_subsystems();
1456 teardown_hyp_mode();
1458 teardown_common_resources();
1462 /* NOP: Compiling as a module not supported */
1463 void kvm_arch_exit(void)
1465 kvm_perf_teardown();
1468 static int arm_init(void)
1470 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1474 module_init(arm_init
);