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 <linux/kvm_irqfd.h>
31 #include <linux/irqbypass.h>
32 #include <trace/events/kvm.h>
33 #include <kvm/arm_pmu.h>
34 #include <kvm/arm_psci.h>
36 #define CREATE_TRACE_POINTS
39 #include <linux/uaccess.h>
40 #include <asm/ptrace.h>
42 #include <asm/tlbflush.h>
43 #include <asm/cacheflush.h>
45 #include <asm/kvm_arm.h>
46 #include <asm/kvm_asm.h>
47 #include <asm/kvm_mmu.h>
48 #include <asm/kvm_emulate.h>
49 #include <asm/kvm_coproc.h>
50 #include <asm/sections.h>
53 __asm__(".arch_extension virt");
56 DEFINE_PER_CPU(kvm_cpu_context_t
, kvm_host_cpu_state
);
57 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
59 /* Per-CPU variable containing the currently running vcpu. */
60 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
62 /* The VMID used in the VTTBR */
63 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
64 static u32 kvm_next_vmid
;
65 static unsigned int kvm_vmid_bits __read_mostly
;
66 static DEFINE_SPINLOCK(kvm_vmid_lock
);
68 static bool vgic_present
;
70 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled
);
72 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
74 BUG_ON(preemptible());
75 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
79 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
80 * Must be called from non-preemptible context
82 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
84 BUG_ON(preemptible());
85 return __this_cpu_read(kvm_arm_running_vcpu
);
89 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
91 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
93 return &kvm_arm_running_vcpu
;
96 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
98 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
101 int kvm_arch_hardware_setup(void)
106 void kvm_arch_check_processor_compat(void *rtn
)
113 * kvm_arch_init_vm - initializes a VM data structure
114 * @kvm: pointer to the KVM struct
116 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
123 kvm
->arch
.last_vcpu_ran
= alloc_percpu(typeof(*kvm
->arch
.last_vcpu_ran
));
124 if (!kvm
->arch
.last_vcpu_ran
)
127 for_each_possible_cpu(cpu
)
128 *per_cpu_ptr(kvm
->arch
.last_vcpu_ran
, cpu
) = -1;
130 ret
= kvm_alloc_stage2_pgd(kvm
);
134 ret
= create_hyp_mappings(kvm
, kvm
+ 1, PAGE_HYP
);
136 goto out_free_stage2_pgd
;
138 kvm_vgic_early_init(kvm
);
140 /* Mark the initial VMID generation invalid */
141 kvm
->arch
.vmid_gen
= 0;
143 /* The maximum number of VCPUs is limited by the host's GIC model */
144 kvm
->arch
.max_vcpus
= vgic_present
?
145 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
149 kvm_free_stage2_pgd(kvm
);
151 free_percpu(kvm
->arch
.last_vcpu_ran
);
152 kvm
->arch
.last_vcpu_ran
= NULL
;
156 bool kvm_arch_has_vcpu_debugfs(void)
161 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
166 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
168 return VM_FAULT_SIGBUS
;
173 * kvm_arch_destroy_vm - destroy the VM data structure
174 * @kvm: pointer to the KVM struct
176 void kvm_arch_destroy_vm(struct kvm
*kvm
)
180 kvm_vgic_destroy(kvm
);
182 free_percpu(kvm
->arch
.last_vcpu_ran
);
183 kvm
->arch
.last_vcpu_ran
= NULL
;
185 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
187 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
188 kvm
->vcpus
[i
] = NULL
;
191 atomic_set(&kvm
->online_vcpus
, 0);
194 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
198 case KVM_CAP_IRQCHIP
:
201 case KVM_CAP_IOEVENTFD
:
202 case KVM_CAP_DEVICE_CTRL
:
203 case KVM_CAP_USER_MEMORY
:
204 case KVM_CAP_SYNC_MMU
:
205 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
206 case KVM_CAP_ONE_REG
:
207 case KVM_CAP_ARM_PSCI
:
208 case KVM_CAP_ARM_PSCI_0_2
:
209 case KVM_CAP_READONLY_MEM
:
210 case KVM_CAP_MP_STATE
:
211 case KVM_CAP_IMMEDIATE_EXIT
:
214 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
217 case KVM_CAP_NR_VCPUS
:
218 r
= num_online_cpus();
220 case KVM_CAP_MAX_VCPUS
:
223 case KVM_CAP_NR_MEMSLOTS
:
224 r
= KVM_USER_MEM_SLOTS
;
226 case KVM_CAP_MSI_DEVID
:
230 r
= kvm
->arch
.vgic
.msis_require_devid
;
232 case KVM_CAP_ARM_USER_IRQ
:
234 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
235 * (bump this number if adding more devices)
240 r
= kvm_arch_dev_ioctl_check_extension(kvm
, ext
);
246 long kvm_arch_dev_ioctl(struct file
*filp
,
247 unsigned int ioctl
, unsigned long arg
)
253 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
256 struct kvm_vcpu
*vcpu
;
258 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
263 if (id
>= kvm
->arch
.max_vcpus
) {
268 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
274 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
278 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
284 kvm_vcpu_uninit(vcpu
);
286 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
291 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
293 kvm_vgic_vcpu_early_init(vcpu
);
296 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
298 kvm_mmu_free_memory_caches(vcpu
);
299 kvm_timer_vcpu_terminate(vcpu
);
300 kvm_pmu_vcpu_destroy(vcpu
);
301 kvm_vcpu_uninit(vcpu
);
302 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
305 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
307 kvm_arch_vcpu_free(vcpu
);
310 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
312 return kvm_timer_is_pending(vcpu
);
315 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
317 kvm_timer_schedule(vcpu
);
318 kvm_vgic_v4_enable_doorbell(vcpu
);
321 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
323 kvm_timer_unschedule(vcpu
);
324 kvm_vgic_v4_disable_doorbell(vcpu
);
327 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
329 /* Force users to call KVM_ARM_VCPU_INIT */
330 vcpu
->arch
.target
= -1;
331 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
333 /* Set up the timer */
334 kvm_timer_vcpu_init(vcpu
);
336 kvm_arm_reset_debug_ptr(vcpu
);
338 return kvm_vgic_vcpu_init(vcpu
);
341 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
345 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
348 * We might get preempted before the vCPU actually runs, but
349 * over-invalidation doesn't affect correctness.
351 if (*last_ran
!= vcpu
->vcpu_id
) {
352 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
353 *last_ran
= vcpu
->vcpu_id
;
357 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(&kvm_host_cpu_state
);
359 kvm_arm_set_running_vcpu(vcpu
);
361 kvm_timer_vcpu_load(vcpu
);
364 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
366 kvm_timer_vcpu_put(vcpu
);
371 kvm_arm_set_running_vcpu(NULL
);
374 static void vcpu_power_off(struct kvm_vcpu
*vcpu
)
376 vcpu
->arch
.power_off
= true;
377 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
381 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
382 struct kvm_mp_state
*mp_state
)
384 if (vcpu
->arch
.power_off
)
385 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
387 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
392 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
393 struct kvm_mp_state
*mp_state
)
395 switch (mp_state
->mp_state
) {
396 case KVM_MP_STATE_RUNNABLE
:
397 vcpu
->arch
.power_off
= false;
399 case KVM_MP_STATE_STOPPED
:
400 vcpu_power_off(vcpu
);
410 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
411 * @v: The VCPU pointer
413 * If the guest CPU is not waiting for interrupts or an interrupt line is
414 * asserted, the CPU is by definition runnable.
416 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
418 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
419 && !v
->arch
.power_off
&& !v
->arch
.pause
);
422 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
424 return vcpu_mode_priv(vcpu
);
427 /* Just ensure a guest exit from a particular CPU */
428 static void exit_vm_noop(void *info
)
432 void force_vm_exit(const cpumask_t
*mask
)
435 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
440 * need_new_vmid_gen - check that the VMID is still valid
441 * @kvm: The VM's VMID to check
443 * return true if there is a new generation of VMIDs being used
445 * The hardware supports only 256 values with the value zero reserved for the
446 * host, so we check if an assigned value belongs to a previous generation,
447 * which which requires us to assign a new value. If we're the first to use a
448 * VMID for the new generation, we must flush necessary caches and TLBs on all
451 static bool need_new_vmid_gen(struct kvm
*kvm
)
453 u64 current_vmid_gen
= atomic64_read(&kvm_vmid_gen
);
454 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
455 return unlikely(READ_ONCE(kvm
->arch
.vmid_gen
) != current_vmid_gen
);
459 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
460 * @kvm The guest that we are about to run
462 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
463 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
466 static void update_vttbr(struct kvm
*kvm
)
468 phys_addr_t pgd_phys
;
471 if (!need_new_vmid_gen(kvm
))
474 spin_lock(&kvm_vmid_lock
);
477 * We need to re-check the vmid_gen here to ensure that if another vcpu
478 * already allocated a valid vmid for this vm, then this vcpu should
481 if (!need_new_vmid_gen(kvm
)) {
482 spin_unlock(&kvm_vmid_lock
);
486 /* First user of a new VMID generation? */
487 if (unlikely(kvm_next_vmid
== 0)) {
488 atomic64_inc(&kvm_vmid_gen
);
492 * On SMP we know no other CPUs can use this CPU's or each
493 * other's VMID after force_vm_exit returns since the
494 * kvm_vmid_lock blocks them from reentry to the guest.
496 force_vm_exit(cpu_all_mask
);
498 * Now broadcast TLB + ICACHE invalidation over the inner
499 * shareable domain to make sure all data structures are
502 kvm_call_hyp(__kvm_flush_vm_context
);
505 kvm
->arch
.vmid
= kvm_next_vmid
;
507 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
509 /* update vttbr to be used with the new vmid */
510 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
511 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
512 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
513 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
516 WRITE_ONCE(kvm
->arch
.vmid_gen
, atomic64_read(&kvm_vmid_gen
));
518 spin_unlock(&kvm_vmid_lock
);
521 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
523 struct kvm
*kvm
= vcpu
->kvm
;
526 if (likely(vcpu
->arch
.has_run_once
))
529 vcpu
->arch
.has_run_once
= true;
532 * Map the VGIC hardware resources before running a vcpu the first
535 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
536 ret
= kvm_vgic_map_resources(kvm
);
541 ret
= kvm_timer_enable(vcpu
);
545 ret
= kvm_arm_pmu_v3_enable(vcpu
);
550 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
552 return vgic_initialized(kvm
);
555 void kvm_arm_halt_guest(struct kvm
*kvm
)
558 struct kvm_vcpu
*vcpu
;
560 kvm_for_each_vcpu(i
, vcpu
, kvm
)
561 vcpu
->arch
.pause
= true;
562 kvm_make_all_cpus_request(kvm
, KVM_REQ_SLEEP
);
565 void kvm_arm_resume_guest(struct kvm
*kvm
)
568 struct kvm_vcpu
*vcpu
;
570 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
571 vcpu
->arch
.pause
= false;
572 swake_up(kvm_arch_vcpu_wq(vcpu
));
576 static void vcpu_req_sleep(struct kvm_vcpu
*vcpu
)
578 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
580 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
581 (!vcpu
->arch
.pause
)));
583 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
584 /* Awaken to handle a signal, request we sleep again later. */
585 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
589 * Make sure we will observe a potential reset request if we've
590 * observed a change to the power state. Pairs with the smp_wmb() in
591 * kvm_psci_vcpu_on().
596 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
598 return vcpu
->arch
.target
>= 0;
601 static void check_vcpu_requests(struct kvm_vcpu
*vcpu
)
603 if (kvm_request_pending(vcpu
)) {
604 if (kvm_check_request(KVM_REQ_SLEEP
, vcpu
))
605 vcpu_req_sleep(vcpu
);
607 if (kvm_check_request(KVM_REQ_VCPU_RESET
, vcpu
))
608 kvm_reset_vcpu(vcpu
);
611 * Clear IRQ_PENDING requests that were made to guarantee
612 * that a VCPU sees new virtual interrupts.
614 kvm_check_request(KVM_REQ_IRQ_PENDING
, vcpu
);
619 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
620 * @vcpu: The VCPU pointer
621 * @run: The kvm_run structure pointer used for userspace state exchange
623 * This function is called through the VCPU_RUN ioctl called from user space. It
624 * will execute VM code in a loop until the time slice for the process is used
625 * or some emulation is needed from user space in which case the function will
626 * return with return value 0 and with the kvm_run structure filled in with the
627 * required data for the requested emulation.
629 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
633 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
636 ret
= kvm_vcpu_first_run_init(vcpu
);
640 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
641 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
644 if (kvm_arm_handle_step_debug(vcpu
, vcpu
->run
))
649 if (run
->immediate_exit
)
652 kvm_sigset_activate(vcpu
);
655 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
658 * Check conditions before entering the guest
662 update_vttbr(vcpu
->kvm
);
664 check_vcpu_requests(vcpu
);
667 * Preparing the interrupts to be injected also
668 * involves poking the GIC, which must be done in a
669 * non-preemptible context.
673 /* Flush FP/SIMD state that can't survive guest entry/exit */
674 kvm_fpsimd_flush_cpu_state();
676 kvm_pmu_flush_hwstate(vcpu
);
680 kvm_vgic_flush_hwstate(vcpu
);
683 * If we have a singal pending, or need to notify a userspace
684 * irqchip about timer or PMU level changes, then we exit (and
685 * update the timer level state in kvm_timer_update_run
688 if (signal_pending(current
) ||
689 kvm_timer_should_notify_user(vcpu
) ||
690 kvm_pmu_should_notify_user(vcpu
)) {
692 run
->exit_reason
= KVM_EXIT_INTR
;
696 * Ensure we set mode to IN_GUEST_MODE after we disable
697 * interrupts and before the final VCPU requests check.
698 * See the comment in kvm_vcpu_exiting_guest_mode() and
699 * Documentation/virtual/kvm/vcpu-requests.rst
701 smp_store_mb(vcpu
->mode
, IN_GUEST_MODE
);
703 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
704 kvm_request_pending(vcpu
)) {
705 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
706 kvm_pmu_sync_hwstate(vcpu
);
707 kvm_timer_sync_hwstate(vcpu
);
708 kvm_vgic_sync_hwstate(vcpu
);
714 kvm_arm_setup_debug(vcpu
);
716 /**************************************************************
719 trace_kvm_entry(*vcpu_pc(vcpu
));
720 guest_enter_irqoff();
722 kvm_arm_vhe_guest_enter();
724 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
727 kvm_arm_vhe_guest_exit();
728 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
732 *************************************************************/
734 kvm_arm_clear_debug(vcpu
);
737 * We must sync the PMU state before the vgic state so
738 * that the vgic can properly sample the updated state of the
741 kvm_pmu_sync_hwstate(vcpu
);
744 * Sync the vgic state before syncing the timer state because
745 * the timer code needs to know if the virtual timer
746 * interrupts are active.
748 kvm_vgic_sync_hwstate(vcpu
);
751 * Sync the timer hardware state before enabling interrupts as
752 * we don't want vtimer interrupts to race with syncing the
753 * timer virtual interrupt state.
755 kvm_timer_sync_hwstate(vcpu
);
758 * We may have taken a host interrupt in HYP mode (ie
759 * while executing the guest). This interrupt is still
760 * pending, as we haven't serviced it yet!
762 * We're now back in SVC mode, with interrupts
763 * disabled. Enabling the interrupts now will have
764 * the effect of taking the interrupt again, in SVC
770 * We do local_irq_enable() before calling guest_exit() so
771 * that if a timer interrupt hits while running the guest we
772 * account that tick as being spent in the guest. We enable
773 * preemption after calling guest_exit() so that if we get
774 * preempted we make sure ticks after that is not counted as
778 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
780 /* Exit types that need handling before we can be preempted */
781 handle_exit_early(vcpu
, run
, ret
);
785 ret
= handle_exit(vcpu
, run
, ret
);
788 /* Tell userspace about in-kernel device output levels */
789 if (unlikely(!irqchip_in_kernel(vcpu
->kvm
))) {
790 kvm_timer_update_run(vcpu
);
791 kvm_pmu_update_run(vcpu
);
794 kvm_sigset_deactivate(vcpu
);
799 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
805 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
806 bit_index
= __ffs(HCR_VI
);
807 else /* KVM_ARM_IRQ_CPU_FIQ */
808 bit_index
= __ffs(HCR_VF
);
810 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
812 set
= test_and_set_bit(bit_index
, ptr
);
814 set
= test_and_clear_bit(bit_index
, ptr
);
817 * If we didn't change anything, no need to wake up or kick other CPUs
823 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
824 * trigger a world-switch round on the running physical CPU to set the
825 * virtual IRQ/FIQ fields in the HCR appropriately.
827 kvm_make_request(KVM_REQ_IRQ_PENDING
, vcpu
);
833 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
836 u32 irq
= irq_level
->irq
;
837 unsigned int irq_type
, vcpu_idx
, irq_num
;
838 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
839 struct kvm_vcpu
*vcpu
= NULL
;
840 bool level
= irq_level
->level
;
842 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
843 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
844 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
846 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
849 case KVM_ARM_IRQ_TYPE_CPU
:
850 if (irqchip_in_kernel(kvm
))
853 if (vcpu_idx
>= nrcpus
)
856 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
860 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
863 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
864 case KVM_ARM_IRQ_TYPE_PPI
:
865 if (!irqchip_in_kernel(kvm
))
868 if (vcpu_idx
>= nrcpus
)
871 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
875 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
878 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
, NULL
);
879 case KVM_ARM_IRQ_TYPE_SPI
:
880 if (!irqchip_in_kernel(kvm
))
883 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
886 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
, NULL
);
892 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
893 const struct kvm_vcpu_init
*init
)
896 int phys_target
= kvm_target_cpu();
898 if (init
->target
!= phys_target
)
902 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
903 * use the same target.
905 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
908 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
909 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
910 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
912 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
916 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
917 * use the same feature set.
919 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
920 test_bit(i
, vcpu
->arch
.features
) != set
)
924 set_bit(i
, vcpu
->arch
.features
);
927 vcpu
->arch
.target
= phys_target
;
929 /* Now we know what it is, we can reset it. */
930 return kvm_reset_vcpu(vcpu
);
934 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
935 struct kvm_vcpu_init
*init
)
939 ret
= kvm_vcpu_set_target(vcpu
, init
);
944 * Ensure a rebooted VM will fault in RAM pages and detect if the
945 * guest MMU is turned off and flush the caches as needed.
947 if (vcpu
->arch
.has_run_once
)
948 stage2_unmap_vm(vcpu
->kvm
);
950 vcpu_reset_hcr(vcpu
);
953 * Handle the "start in power-off" case.
955 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
956 vcpu_power_off(vcpu
);
958 vcpu
->arch
.power_off
= false;
963 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
964 struct kvm_device_attr
*attr
)
968 switch (attr
->group
) {
970 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
977 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
978 struct kvm_device_attr
*attr
)
982 switch (attr
->group
) {
984 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
991 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
992 struct kvm_device_attr
*attr
)
996 switch (attr
->group
) {
998 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
1005 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1006 unsigned int ioctl
, unsigned long arg
)
1008 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1009 void __user
*argp
= (void __user
*)arg
;
1010 struct kvm_device_attr attr
;
1013 case KVM_ARM_VCPU_INIT
: {
1014 struct kvm_vcpu_init init
;
1016 if (copy_from_user(&init
, argp
, sizeof(init
)))
1019 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
1021 case KVM_SET_ONE_REG
:
1022 case KVM_GET_ONE_REG
: {
1023 struct kvm_one_reg reg
;
1025 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1028 if (copy_from_user(®
, argp
, sizeof(reg
)))
1030 if (ioctl
== KVM_SET_ONE_REG
)
1031 return kvm_arm_set_reg(vcpu
, ®
);
1033 return kvm_arm_get_reg(vcpu
, ®
);
1035 case KVM_GET_REG_LIST
: {
1036 struct kvm_reg_list __user
*user_list
= argp
;
1037 struct kvm_reg_list reg_list
;
1040 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1043 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
1046 reg_list
.n
= kvm_arm_num_regs(vcpu
);
1047 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
1051 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
1053 case KVM_SET_DEVICE_ATTR
: {
1054 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1056 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
1058 case KVM_GET_DEVICE_ATTR
: {
1059 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1061 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
1063 case KVM_HAS_DEVICE_ATTR
: {
1064 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1066 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
1074 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1075 * @kvm: kvm instance
1076 * @log: slot id and address to which we copy the log
1078 * Steps 1-4 below provide general overview of dirty page logging. See
1079 * kvm_get_dirty_log_protect() function description for additional details.
1081 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1082 * always flush the TLB (step 4) even if previous step failed and the dirty
1083 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1084 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1085 * writes will be marked dirty for next log read.
1087 * 1. Take a snapshot of the bit and clear it if needed.
1088 * 2. Write protect the corresponding page.
1089 * 3. Copy the snapshot to the userspace.
1090 * 4. Flush TLB's if needed.
1092 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1094 bool is_dirty
= false;
1097 mutex_lock(&kvm
->slots_lock
);
1099 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
1102 kvm_flush_remote_tlbs(kvm
);
1104 mutex_unlock(&kvm
->slots_lock
);
1108 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1109 struct kvm_arm_device_addr
*dev_addr
)
1111 unsigned long dev_id
, type
;
1113 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1114 KVM_ARM_DEVICE_ID_SHIFT
;
1115 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1116 KVM_ARM_DEVICE_TYPE_SHIFT
;
1119 case KVM_ARM_DEVICE_VGIC_V2
:
1122 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1128 long kvm_arch_vm_ioctl(struct file
*filp
,
1129 unsigned int ioctl
, unsigned long arg
)
1131 struct kvm
*kvm
= filp
->private_data
;
1132 void __user
*argp
= (void __user
*)arg
;
1135 case KVM_CREATE_IRQCHIP
: {
1139 mutex_lock(&kvm
->lock
);
1140 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1141 mutex_unlock(&kvm
->lock
);
1144 case KVM_ARM_SET_DEVICE_ADDR
: {
1145 struct kvm_arm_device_addr dev_addr
;
1147 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1149 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1151 case KVM_ARM_PREFERRED_TARGET
: {
1153 struct kvm_vcpu_init init
;
1155 err
= kvm_vcpu_preferred_target(&init
);
1159 if (copy_to_user(argp
, &init
, sizeof(init
)))
1169 static void cpu_init_hyp_mode(void *dummy
)
1171 phys_addr_t pgd_ptr
;
1172 unsigned long hyp_stack_ptr
;
1173 unsigned long stack_page
;
1174 unsigned long vector_ptr
;
1176 /* Switch from the HYP stub to our own HYP init vector */
1177 __hyp_set_vectors(kvm_get_idmap_vector());
1179 pgd_ptr
= kvm_mmu_get_httbr();
1180 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1181 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1182 vector_ptr
= (unsigned long)kvm_get_hyp_vector();
1184 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1185 __cpu_init_stage2();
1188 static void cpu_hyp_reset(void)
1190 if (!is_kernel_in_hyp_mode())
1191 __hyp_reset_vectors();
1194 static void cpu_hyp_reinit(void)
1198 if (is_kernel_in_hyp_mode()) {
1200 * __cpu_init_stage2() is safe to call even if the PM
1201 * event was cancelled before the CPU was reset.
1203 __cpu_init_stage2();
1204 kvm_timer_init_vhe();
1206 cpu_init_hyp_mode(NULL
);
1209 kvm_arm_init_debug();
1212 kvm_vgic_init_cpu_hardware();
1215 static void _kvm_arch_hardware_enable(void *discard
)
1217 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1219 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1223 int kvm_arch_hardware_enable(void)
1225 _kvm_arch_hardware_enable(NULL
);
1229 static void _kvm_arch_hardware_disable(void *discard
)
1231 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1233 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1237 void kvm_arch_hardware_disable(void)
1239 _kvm_arch_hardware_disable(NULL
);
1242 #ifdef CONFIG_CPU_PM
1243 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1248 * kvm_arm_hardware_enabled is left with its old value over
1249 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1254 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1256 * don't update kvm_arm_hardware_enabled here
1257 * so that the hardware will be re-enabled
1258 * when we resume. See below.
1263 case CPU_PM_ENTER_FAILED
:
1265 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1266 /* The hardware was enabled before suspend. */
1276 static struct notifier_block hyp_init_cpu_pm_nb
= {
1277 .notifier_call
= hyp_init_cpu_pm_notifier
,
1280 static void __init
hyp_cpu_pm_init(void)
1282 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1284 static void __init
hyp_cpu_pm_exit(void)
1286 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1289 static inline void hyp_cpu_pm_init(void)
1292 static inline void hyp_cpu_pm_exit(void)
1297 static int init_common_resources(void)
1299 /* set size of VMID supported by CPU */
1300 kvm_vmid_bits
= kvm_get_vmid_bits();
1301 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1306 static int init_subsystems(void)
1311 * Enable hardware so that subsystem initialisation can access EL2.
1313 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1316 * Register CPU lower-power notifier
1321 * Init HYP view of VGIC
1323 err
= kvm_vgic_hyp_init();
1326 vgic_present
= true;
1330 vgic_present
= false;
1338 * Init HYP architected timer support
1340 err
= kvm_timer_hyp_init(vgic_present
);
1345 kvm_coproc_table_init();
1348 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1353 static void teardown_hyp_mode(void)
1358 for_each_possible_cpu(cpu
)
1359 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1364 * Inits Hyp-mode on all online CPUs
1366 static int init_hyp_mode(void)
1372 * Allocate Hyp PGD and setup Hyp identity mapping
1374 err
= kvm_mmu_init();
1379 * Allocate stack pages for Hypervisor-mode
1381 for_each_possible_cpu(cpu
) {
1382 unsigned long stack_page
;
1384 stack_page
= __get_free_page(GFP_KERNEL
);
1390 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1394 * Map the Hyp-code called directly from the host
1396 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1397 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1399 kvm_err("Cannot map world-switch code\n");
1403 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1404 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1406 kvm_err("Cannot map rodata section\n");
1410 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1411 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1413 kvm_err("Cannot map bss section\n");
1417 err
= kvm_map_vectors();
1419 kvm_err("Cannot map vectors\n");
1424 * Map the Hyp stack pages
1426 for_each_possible_cpu(cpu
) {
1427 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1428 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1432 kvm_err("Cannot map hyp stack\n");
1437 for_each_possible_cpu(cpu
) {
1438 kvm_cpu_context_t
*cpu_ctxt
;
1440 cpu_ctxt
= per_cpu_ptr(&kvm_host_cpu_state
, cpu
);
1441 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1444 kvm_err("Cannot map host CPU state: %d\n", err
);
1449 err
= hyp_map_aux_data();
1451 kvm_err("Cannot map host auxilary data: %d\n", err
);
1456 teardown_hyp_mode();
1457 kvm_err("error initializing Hyp mode: %d\n", err
);
1461 static void check_kvm_target_cpu(void *ret
)
1463 *(int *)ret
= kvm_target_cpu();
1466 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1468 struct kvm_vcpu
*vcpu
;
1471 mpidr
&= MPIDR_HWID_BITMASK
;
1472 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1473 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1479 bool kvm_arch_has_irq_bypass(void)
1484 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
1485 struct irq_bypass_producer
*prod
)
1487 struct kvm_kernel_irqfd
*irqfd
=
1488 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1490 return kvm_vgic_v4_set_forwarding(irqfd
->kvm
, prod
->irq
,
1493 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
1494 struct irq_bypass_producer
*prod
)
1496 struct kvm_kernel_irqfd
*irqfd
=
1497 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1499 kvm_vgic_v4_unset_forwarding(irqfd
->kvm
, prod
->irq
,
1503 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer
*cons
)
1505 struct kvm_kernel_irqfd
*irqfd
=
1506 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1508 kvm_arm_halt_guest(irqfd
->kvm
);
1511 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer
*cons
)
1513 struct kvm_kernel_irqfd
*irqfd
=
1514 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1516 kvm_arm_resume_guest(irqfd
->kvm
);
1520 * Initialize Hyp-mode and memory mappings on all CPUs.
1522 int kvm_arch_init(void *opaque
)
1528 if (!is_hyp_mode_available()) {
1529 kvm_info("HYP mode not available\n");
1533 for_each_online_cpu(cpu
) {
1534 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1536 kvm_err("Error, CPU %d not supported!\n", cpu
);
1541 err
= init_common_resources();
1545 in_hyp_mode
= is_kernel_in_hyp_mode();
1548 err
= init_hyp_mode();
1553 err
= init_subsystems();
1558 kvm_info("VHE mode initialized successfully\n");
1560 kvm_info("Hyp mode initialized successfully\n");
1566 teardown_hyp_mode();
1571 /* NOP: Compiling as a module not supported */
1572 void kvm_arch_exit(void)
1574 kvm_perf_teardown();
1577 static int arm_init(void)
1579 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1583 module_init(arm_init
);