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_MSI_DEVID
:
228 r
= kvm
->arch
.vgic
.msis_require_devid
;
231 r
= kvm_arch_dev_ioctl_check_extension(kvm
, ext
);
237 long kvm_arch_dev_ioctl(struct file
*filp
,
238 unsigned int ioctl
, unsigned long arg
)
244 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
247 struct kvm_vcpu
*vcpu
;
249 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
254 if (id
>= kvm
->arch
.max_vcpus
) {
259 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
265 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
269 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
275 kvm_vcpu_uninit(vcpu
);
277 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
282 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
284 kvm_vgic_vcpu_early_init(vcpu
);
287 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
289 kvm_mmu_free_memory_caches(vcpu
);
290 kvm_timer_vcpu_terminate(vcpu
);
291 kvm_vgic_vcpu_destroy(vcpu
);
292 kvm_pmu_vcpu_destroy(vcpu
);
293 kvm_vcpu_uninit(vcpu
);
294 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
297 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
299 kvm_arch_vcpu_free(vcpu
);
302 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
304 return kvm_timer_should_fire(vcpu_vtimer(vcpu
)) ||
305 kvm_timer_should_fire(vcpu_ptimer(vcpu
));
308 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
310 kvm_timer_schedule(vcpu
);
313 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
315 kvm_timer_unschedule(vcpu
);
318 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
320 /* Force users to call KVM_ARM_VCPU_INIT */
321 vcpu
->arch
.target
= -1;
322 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
324 /* Set up the timer */
325 kvm_timer_vcpu_init(vcpu
);
327 kvm_arm_reset_debug_ptr(vcpu
);
332 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
336 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
339 * We might get preempted before the vCPU actually runs, but
340 * over-invalidation doesn't affect correctness.
342 if (*last_ran
!= vcpu
->vcpu_id
) {
343 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
344 *last_ran
= vcpu
->vcpu_id
;
348 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
350 kvm_arm_set_running_vcpu(vcpu
);
353 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
356 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
357 * if the vcpu is no longer assigned to a cpu. This is used for the
358 * optimized make_all_cpus_request path.
362 kvm_arm_set_running_vcpu(NULL
);
363 kvm_timer_vcpu_put(vcpu
);
366 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
367 struct kvm_mp_state
*mp_state
)
369 if (vcpu
->arch
.power_off
)
370 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
372 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
377 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
378 struct kvm_mp_state
*mp_state
)
380 switch (mp_state
->mp_state
) {
381 case KVM_MP_STATE_RUNNABLE
:
382 vcpu
->arch
.power_off
= false;
384 case KVM_MP_STATE_STOPPED
:
385 vcpu
->arch
.power_off
= true;
395 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
396 * @v: The VCPU pointer
398 * If the guest CPU is not waiting for interrupts or an interrupt line is
399 * asserted, the CPU is by definition runnable.
401 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
403 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
404 && !v
->arch
.power_off
&& !v
->arch
.pause
);
407 /* Just ensure a guest exit from a particular CPU */
408 static void exit_vm_noop(void *info
)
412 void force_vm_exit(const cpumask_t
*mask
)
415 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
420 * need_new_vmid_gen - check that the VMID is still valid
421 * @kvm: The VM's VMID to check
423 * return true if there is a new generation of VMIDs being used
425 * The hardware supports only 256 values with the value zero reserved for the
426 * host, so we check if an assigned value belongs to a previous generation,
427 * which which requires us to assign a new value. If we're the first to use a
428 * VMID for the new generation, we must flush necessary caches and TLBs on all
431 static bool need_new_vmid_gen(struct kvm
*kvm
)
433 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
437 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
438 * @kvm The guest that we are about to run
440 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
441 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
444 static void update_vttbr(struct kvm
*kvm
)
446 phys_addr_t pgd_phys
;
449 if (!need_new_vmid_gen(kvm
))
452 spin_lock(&kvm_vmid_lock
);
455 * We need to re-check the vmid_gen here to ensure that if another vcpu
456 * already allocated a valid vmid for this vm, then this vcpu should
459 if (!need_new_vmid_gen(kvm
)) {
460 spin_unlock(&kvm_vmid_lock
);
464 /* First user of a new VMID generation? */
465 if (unlikely(kvm_next_vmid
== 0)) {
466 atomic64_inc(&kvm_vmid_gen
);
470 * On SMP we know no other CPUs can use this CPU's or each
471 * other's VMID after force_vm_exit returns since the
472 * kvm_vmid_lock blocks them from reentry to the guest.
474 force_vm_exit(cpu_all_mask
);
476 * Now broadcast TLB + ICACHE invalidation over the inner
477 * shareable domain to make sure all data structures are
480 kvm_call_hyp(__kvm_flush_vm_context
);
483 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
484 kvm
->arch
.vmid
= kvm_next_vmid
;
486 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
488 /* update vttbr to be used with the new vmid */
489 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
490 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
491 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
492 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
494 spin_unlock(&kvm_vmid_lock
);
497 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
499 struct kvm
*kvm
= vcpu
->kvm
;
502 if (likely(vcpu
->arch
.has_run_once
))
505 vcpu
->arch
.has_run_once
= true;
508 * Map the VGIC hardware resources before running a vcpu the first
511 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
512 ret
= kvm_vgic_map_resources(kvm
);
518 * Enable the arch timers only if we have an in-kernel VGIC
519 * and it has been properly initialized, since we cannot handle
520 * interrupts from the virtual timer with a userspace gic.
522 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
523 ret
= kvm_timer_enable(vcpu
);
528 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
530 return vgic_initialized(kvm
);
533 void kvm_arm_halt_guest(struct kvm
*kvm
)
536 struct kvm_vcpu
*vcpu
;
538 kvm_for_each_vcpu(i
, vcpu
, kvm
)
539 vcpu
->arch
.pause
= true;
540 kvm_make_all_cpus_request(kvm
, KVM_REQ_VCPU_EXIT
);
543 void kvm_arm_halt_vcpu(struct kvm_vcpu
*vcpu
)
545 vcpu
->arch
.pause
= true;
549 void kvm_arm_resume_vcpu(struct kvm_vcpu
*vcpu
)
551 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
553 vcpu
->arch
.pause
= false;
557 void kvm_arm_resume_guest(struct kvm
*kvm
)
560 struct kvm_vcpu
*vcpu
;
562 kvm_for_each_vcpu(i
, vcpu
, kvm
)
563 kvm_arm_resume_vcpu(vcpu
);
566 static void vcpu_sleep(struct kvm_vcpu
*vcpu
)
568 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
570 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
571 (!vcpu
->arch
.pause
)));
574 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
576 return vcpu
->arch
.target
>= 0;
580 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
581 * @vcpu: The VCPU pointer
582 * @run: The kvm_run structure pointer used for userspace state exchange
584 * This function is called through the VCPU_RUN ioctl called from user space. It
585 * will execute VM code in a loop until the time slice for the process is used
586 * or some emulation is needed from user space in which case the function will
587 * return with return value 0 and with the kvm_run structure filled in with the
588 * required data for the requested emulation.
590 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
595 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
598 ret
= kvm_vcpu_first_run_init(vcpu
);
602 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
603 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
608 if (run
->immediate_exit
)
611 if (vcpu
->sigset_active
)
612 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
615 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
618 * Check conditions before entering the guest
622 update_vttbr(vcpu
->kvm
);
624 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
)
628 * Preparing the interrupts to be injected also
629 * involves poking the GIC, which must be done in a
630 * non-preemptible context.
633 kvm_pmu_flush_hwstate(vcpu
);
634 kvm_timer_flush_hwstate(vcpu
);
635 kvm_vgic_flush_hwstate(vcpu
);
640 * Re-check atomic conditions
642 if (signal_pending(current
)) {
644 run
->exit_reason
= KVM_EXIT_INTR
;
647 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
648 vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
650 kvm_pmu_sync_hwstate(vcpu
);
651 kvm_timer_sync_hwstate(vcpu
);
652 kvm_vgic_sync_hwstate(vcpu
);
657 kvm_arm_setup_debug(vcpu
);
659 /**************************************************************
662 trace_kvm_entry(*vcpu_pc(vcpu
));
663 guest_enter_irqoff();
664 vcpu
->mode
= IN_GUEST_MODE
;
666 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
668 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
672 *************************************************************/
674 kvm_arm_clear_debug(vcpu
);
677 * We may have taken a host interrupt in HYP mode (ie
678 * while executing the guest). This interrupt is still
679 * pending, as we haven't serviced it yet!
681 * We're now back in SVC mode, with interrupts
682 * disabled. Enabling the interrupts now will have
683 * the effect of taking the interrupt again, in SVC
689 * We do local_irq_enable() before calling guest_exit() so
690 * that if a timer interrupt hits while running the guest we
691 * account that tick as being spent in the guest. We enable
692 * preemption after calling guest_exit() so that if we get
693 * preempted we make sure ticks after that is not counted as
697 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
700 * We must sync the PMU and timer state before the vgic state so
701 * that the vgic can properly sample the updated state of the
704 kvm_pmu_sync_hwstate(vcpu
);
705 kvm_timer_sync_hwstate(vcpu
);
707 kvm_vgic_sync_hwstate(vcpu
);
711 ret
= handle_exit(vcpu
, run
, ret
);
714 if (vcpu
->sigset_active
)
715 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
719 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
725 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
726 bit_index
= __ffs(HCR_VI
);
727 else /* KVM_ARM_IRQ_CPU_FIQ */
728 bit_index
= __ffs(HCR_VF
);
730 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
732 set
= test_and_set_bit(bit_index
, ptr
);
734 set
= test_and_clear_bit(bit_index
, ptr
);
737 * If we didn't change anything, no need to wake up or kick other CPUs
743 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
744 * trigger a world-switch round on the running physical CPU to set the
745 * virtual IRQ/FIQ fields in the HCR appropriately.
752 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
755 u32 irq
= irq_level
->irq
;
756 unsigned int irq_type
, vcpu_idx
, irq_num
;
757 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
758 struct kvm_vcpu
*vcpu
= NULL
;
759 bool level
= irq_level
->level
;
761 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
762 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
763 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
765 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
768 case KVM_ARM_IRQ_TYPE_CPU
:
769 if (irqchip_in_kernel(kvm
))
772 if (vcpu_idx
>= nrcpus
)
775 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
779 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
782 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
783 case KVM_ARM_IRQ_TYPE_PPI
:
784 if (!irqchip_in_kernel(kvm
))
787 if (vcpu_idx
>= nrcpus
)
790 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
794 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
797 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
798 case KVM_ARM_IRQ_TYPE_SPI
:
799 if (!irqchip_in_kernel(kvm
))
802 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
805 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
811 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
812 const struct kvm_vcpu_init
*init
)
815 int phys_target
= kvm_target_cpu();
817 if (init
->target
!= phys_target
)
821 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
822 * use the same target.
824 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
827 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
828 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
829 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
831 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
835 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
836 * use the same feature set.
838 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
839 test_bit(i
, vcpu
->arch
.features
) != set
)
843 set_bit(i
, vcpu
->arch
.features
);
846 vcpu
->arch
.target
= phys_target
;
848 /* Now we know what it is, we can reset it. */
849 return kvm_reset_vcpu(vcpu
);
853 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
854 struct kvm_vcpu_init
*init
)
858 ret
= kvm_vcpu_set_target(vcpu
, init
);
863 * Ensure a rebooted VM will fault in RAM pages and detect if the
864 * guest MMU is turned off and flush the caches as needed.
866 if (vcpu
->arch
.has_run_once
)
867 stage2_unmap_vm(vcpu
->kvm
);
869 vcpu_reset_hcr(vcpu
);
872 * Handle the "start in power-off" case.
874 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
875 vcpu
->arch
.power_off
= true;
877 vcpu
->arch
.power_off
= false;
882 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
883 struct kvm_device_attr
*attr
)
887 switch (attr
->group
) {
889 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
896 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
897 struct kvm_device_attr
*attr
)
901 switch (attr
->group
) {
903 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
910 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
911 struct kvm_device_attr
*attr
)
915 switch (attr
->group
) {
917 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
924 long kvm_arch_vcpu_ioctl(struct file
*filp
,
925 unsigned int ioctl
, unsigned long arg
)
927 struct kvm_vcpu
*vcpu
= filp
->private_data
;
928 void __user
*argp
= (void __user
*)arg
;
929 struct kvm_device_attr attr
;
932 case KVM_ARM_VCPU_INIT
: {
933 struct kvm_vcpu_init init
;
935 if (copy_from_user(&init
, argp
, sizeof(init
)))
938 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
940 case KVM_SET_ONE_REG
:
941 case KVM_GET_ONE_REG
: {
942 struct kvm_one_reg reg
;
944 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
947 if (copy_from_user(®
, argp
, sizeof(reg
)))
949 if (ioctl
== KVM_SET_ONE_REG
)
950 return kvm_arm_set_reg(vcpu
, ®
);
952 return kvm_arm_get_reg(vcpu
, ®
);
954 case KVM_GET_REG_LIST
: {
955 struct kvm_reg_list __user
*user_list
= argp
;
956 struct kvm_reg_list reg_list
;
959 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
962 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
965 reg_list
.n
= kvm_arm_num_regs(vcpu
);
966 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
970 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
972 case KVM_SET_DEVICE_ATTR
: {
973 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
975 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
977 case KVM_GET_DEVICE_ATTR
: {
978 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
980 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
982 case KVM_HAS_DEVICE_ATTR
: {
983 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
985 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
993 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
995 * @log: slot id and address to which we copy the log
997 * Steps 1-4 below provide general overview of dirty page logging. See
998 * kvm_get_dirty_log_protect() function description for additional details.
1000 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1001 * always flush the TLB (step 4) even if previous step failed and the dirty
1002 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1003 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1004 * writes will be marked dirty for next log read.
1006 * 1. Take a snapshot of the bit and clear it if needed.
1007 * 2. Write protect the corresponding page.
1008 * 3. Copy the snapshot to the userspace.
1009 * 4. Flush TLB's if needed.
1011 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1013 bool is_dirty
= false;
1016 mutex_lock(&kvm
->slots_lock
);
1018 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
1021 kvm_flush_remote_tlbs(kvm
);
1023 mutex_unlock(&kvm
->slots_lock
);
1027 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1028 struct kvm_arm_device_addr
*dev_addr
)
1030 unsigned long dev_id
, type
;
1032 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1033 KVM_ARM_DEVICE_ID_SHIFT
;
1034 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1035 KVM_ARM_DEVICE_TYPE_SHIFT
;
1038 case KVM_ARM_DEVICE_VGIC_V2
:
1041 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1047 long kvm_arch_vm_ioctl(struct file
*filp
,
1048 unsigned int ioctl
, unsigned long arg
)
1050 struct kvm
*kvm
= filp
->private_data
;
1051 void __user
*argp
= (void __user
*)arg
;
1054 case KVM_CREATE_IRQCHIP
: {
1058 mutex_lock(&kvm
->lock
);
1059 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1060 mutex_unlock(&kvm
->lock
);
1063 case KVM_ARM_SET_DEVICE_ADDR
: {
1064 struct kvm_arm_device_addr dev_addr
;
1066 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1068 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1070 case KVM_ARM_PREFERRED_TARGET
: {
1072 struct kvm_vcpu_init init
;
1074 err
= kvm_vcpu_preferred_target(&init
);
1078 if (copy_to_user(argp
, &init
, sizeof(init
)))
1088 static void cpu_init_hyp_mode(void *dummy
)
1090 phys_addr_t pgd_ptr
;
1091 unsigned long hyp_stack_ptr
;
1092 unsigned long stack_page
;
1093 unsigned long vector_ptr
;
1095 /* Switch from the HYP stub to our own HYP init vector */
1096 __hyp_set_vectors(kvm_get_idmap_vector());
1098 pgd_ptr
= kvm_mmu_get_httbr();
1099 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1100 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1101 vector_ptr
= (unsigned long)kvm_ksym_ref(__kvm_hyp_vector
);
1103 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1104 __cpu_init_stage2();
1106 if (is_kernel_in_hyp_mode())
1107 kvm_timer_init_vhe();
1109 kvm_arm_init_debug();
1112 static void cpu_hyp_reinit(void)
1114 if (is_kernel_in_hyp_mode()) {
1116 * __cpu_init_stage2() is safe to call even if the PM
1117 * event was cancelled before the CPU was reset.
1119 __cpu_init_stage2();
1121 if (__hyp_get_vectors() == hyp_default_vectors
)
1122 cpu_init_hyp_mode(NULL
);
1126 static void cpu_hyp_reset(void)
1128 if (!is_kernel_in_hyp_mode())
1129 __cpu_reset_hyp_mode(hyp_default_vectors
,
1130 kvm_get_idmap_start());
1133 static void _kvm_arch_hardware_enable(void *discard
)
1135 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1137 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1141 int kvm_arch_hardware_enable(void)
1143 _kvm_arch_hardware_enable(NULL
);
1147 static void _kvm_arch_hardware_disable(void *discard
)
1149 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1151 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1155 void kvm_arch_hardware_disable(void)
1157 _kvm_arch_hardware_disable(NULL
);
1160 #ifdef CONFIG_CPU_PM
1161 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1166 * kvm_arm_hardware_enabled is left with its old value over
1167 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1172 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1174 * don't update kvm_arm_hardware_enabled here
1175 * so that the hardware will be re-enabled
1176 * when we resume. See below.
1182 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1183 /* The hardware was enabled before suspend. */
1193 static struct notifier_block hyp_init_cpu_pm_nb
= {
1194 .notifier_call
= hyp_init_cpu_pm_notifier
,
1197 static void __init
hyp_cpu_pm_init(void)
1199 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1201 static void __init
hyp_cpu_pm_exit(void)
1203 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1206 static inline void hyp_cpu_pm_init(void)
1209 static inline void hyp_cpu_pm_exit(void)
1214 static void teardown_common_resources(void)
1216 free_percpu(kvm_host_cpu_state
);
1219 static int init_common_resources(void)
1221 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1222 if (!kvm_host_cpu_state
) {
1223 kvm_err("Cannot allocate host CPU state\n");
1227 /* set size of VMID supported by CPU */
1228 kvm_vmid_bits
= kvm_get_vmid_bits();
1229 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1234 static int init_subsystems(void)
1239 * Enable hardware so that subsystem initialisation can access EL2.
1241 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1244 * Register CPU lower-power notifier
1249 * Init HYP view of VGIC
1251 err
= kvm_vgic_hyp_init();
1254 vgic_present
= true;
1258 vgic_present
= false;
1266 * Init HYP architected timer support
1268 err
= kvm_timer_hyp_init();
1273 kvm_coproc_table_init();
1276 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1281 static void teardown_hyp_mode(void)
1285 if (is_kernel_in_hyp_mode())
1289 for_each_possible_cpu(cpu
)
1290 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1294 static int init_vhe_mode(void)
1296 kvm_info("VHE mode initialized successfully\n");
1301 * Inits Hyp-mode on all online CPUs
1303 static int init_hyp_mode(void)
1309 * Allocate Hyp PGD and setup Hyp identity mapping
1311 err
= kvm_mmu_init();
1316 * It is probably enough to obtain the default on one
1317 * CPU. It's unlikely to be different on the others.
1319 hyp_default_vectors
= __hyp_get_vectors();
1322 * Allocate stack pages for Hypervisor-mode
1324 for_each_possible_cpu(cpu
) {
1325 unsigned long stack_page
;
1327 stack_page
= __get_free_page(GFP_KERNEL
);
1333 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1337 * Map the Hyp-code called directly from the host
1339 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1340 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1342 kvm_err("Cannot map world-switch code\n");
1346 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1347 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1349 kvm_err("Cannot map rodata section\n");
1353 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1354 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1356 kvm_err("Cannot map bss section\n");
1361 * Map the Hyp stack pages
1363 for_each_possible_cpu(cpu
) {
1364 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1365 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1369 kvm_err("Cannot map hyp stack\n");
1374 for_each_possible_cpu(cpu
) {
1375 kvm_cpu_context_t
*cpu_ctxt
;
1377 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1378 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1381 kvm_err("Cannot map host CPU state: %d\n", err
);
1386 kvm_info("Hyp mode initialized successfully\n");
1391 teardown_hyp_mode();
1392 kvm_err("error initializing Hyp mode: %d\n", err
);
1396 static void check_kvm_target_cpu(void *ret
)
1398 *(int *)ret
= kvm_target_cpu();
1401 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1403 struct kvm_vcpu
*vcpu
;
1406 mpidr
&= MPIDR_HWID_BITMASK
;
1407 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1408 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1415 * Initialize Hyp-mode and memory mappings on all CPUs.
1417 int kvm_arch_init(void *opaque
)
1422 if (!is_hyp_mode_available()) {
1423 kvm_err("HYP mode not available\n");
1427 for_each_online_cpu(cpu
) {
1428 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1430 kvm_err("Error, CPU %d not supported!\n", cpu
);
1435 err
= init_common_resources();
1439 if (is_kernel_in_hyp_mode())
1440 err
= init_vhe_mode();
1442 err
= init_hyp_mode();
1446 err
= init_subsystems();
1453 teardown_hyp_mode();
1455 teardown_common_resources();
1459 /* NOP: Compiling as a module not supported */
1460 void kvm_arch_exit(void)
1462 kvm_perf_teardown();
1465 static int arm_init(void)
1467 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1471 module_init(arm_init
);