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
;
57 /* Per-CPU variable containing the currently running vcpu. */
58 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
60 /* The VMID used in the VTTBR */
61 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
62 static u32 kvm_next_vmid
;
63 static unsigned int kvm_vmid_bits __read_mostly
;
64 static DEFINE_SPINLOCK(kvm_vmid_lock
);
66 static bool vgic_present
;
68 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled
);
70 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
72 BUG_ON(preemptible());
73 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
77 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
78 * Must be called from non-preemptible context
80 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
82 BUG_ON(preemptible());
83 return __this_cpu_read(kvm_arm_running_vcpu
);
87 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
89 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
91 return &kvm_arm_running_vcpu
;
94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
96 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
99 int kvm_arch_hardware_setup(void)
104 void kvm_arch_check_processor_compat(void *rtn
)
111 * kvm_arch_init_vm - initializes a VM data structure
112 * @kvm: pointer to the KVM struct
114 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
121 kvm
->arch
.last_vcpu_ran
= alloc_percpu(typeof(*kvm
->arch
.last_vcpu_ran
));
122 if (!kvm
->arch
.last_vcpu_ran
)
125 for_each_possible_cpu(cpu
)
126 *per_cpu_ptr(kvm
->arch
.last_vcpu_ran
, cpu
) = -1;
128 ret
= kvm_alloc_stage2_pgd(kvm
);
132 ret
= create_hyp_mappings(kvm
, kvm
+ 1, PAGE_HYP
);
134 goto out_free_stage2_pgd
;
136 kvm_vgic_early_init(kvm
);
138 /* Mark the initial VMID generation invalid */
139 kvm
->arch
.vmid_gen
= 0;
141 /* The maximum number of VCPUs is limited by the host's GIC model */
142 kvm
->arch
.max_vcpus
= vgic_present
?
143 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
147 kvm_free_stage2_pgd(kvm
);
149 free_percpu(kvm
->arch
.last_vcpu_ran
);
150 kvm
->arch
.last_vcpu_ran
= NULL
;
154 bool kvm_arch_has_vcpu_debugfs(void)
159 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
164 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
166 return VM_FAULT_SIGBUS
;
171 * kvm_arch_destroy_vm - destroy the VM data structure
172 * @kvm: pointer to the KVM struct
174 void kvm_arch_destroy_vm(struct kvm
*kvm
)
178 free_percpu(kvm
->arch
.last_vcpu_ran
);
179 kvm
->arch
.last_vcpu_ran
= NULL
;
181 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
183 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
184 kvm
->vcpus
[i
] = NULL
;
188 kvm_vgic_destroy(kvm
);
191 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
195 case KVM_CAP_IRQCHIP
:
198 case KVM_CAP_IOEVENTFD
:
199 case KVM_CAP_DEVICE_CTRL
:
200 case KVM_CAP_USER_MEMORY
:
201 case KVM_CAP_SYNC_MMU
:
202 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
203 case KVM_CAP_ONE_REG
:
204 case KVM_CAP_ARM_PSCI
:
205 case KVM_CAP_ARM_PSCI_0_2
:
206 case KVM_CAP_READONLY_MEM
:
207 case KVM_CAP_MP_STATE
:
208 case KVM_CAP_IMMEDIATE_EXIT
:
211 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
214 case KVM_CAP_NR_VCPUS
:
215 r
= num_online_cpus();
217 case KVM_CAP_MAX_VCPUS
:
220 case KVM_CAP_NR_MEMSLOTS
:
221 r
= KVM_USER_MEM_SLOTS
;
223 case KVM_CAP_MSI_DEVID
:
227 r
= kvm
->arch
.vgic
.msis_require_devid
;
229 case KVM_CAP_ARM_USER_IRQ
:
231 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
232 * (bump this number if adding more devices)
237 r
= kvm_arch_dev_ioctl_check_extension(kvm
, ext
);
243 long kvm_arch_dev_ioctl(struct file
*filp
,
244 unsigned int ioctl
, unsigned long arg
)
250 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
253 struct kvm_vcpu
*vcpu
;
255 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
260 if (id
>= kvm
->arch
.max_vcpus
) {
265 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
271 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
275 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
281 kvm_vcpu_uninit(vcpu
);
283 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
288 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
290 kvm_vgic_vcpu_early_init(vcpu
);
293 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
295 kvm_mmu_free_memory_caches(vcpu
);
296 kvm_timer_vcpu_terminate(vcpu
);
297 kvm_vgic_vcpu_destroy(vcpu
);
298 kvm_pmu_vcpu_destroy(vcpu
);
299 kvm_vcpu_uninit(vcpu
);
300 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
303 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
305 kvm_arch_vcpu_free(vcpu
);
308 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
310 return kvm_timer_should_fire(vcpu_vtimer(vcpu
)) ||
311 kvm_timer_should_fire(vcpu_ptimer(vcpu
));
314 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
316 kvm_timer_schedule(vcpu
);
319 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
321 kvm_timer_unschedule(vcpu
);
324 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
326 /* Force users to call KVM_ARM_VCPU_INIT */
327 vcpu
->arch
.target
= -1;
328 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
330 /* Set up the timer */
331 kvm_timer_vcpu_init(vcpu
);
333 kvm_arm_reset_debug_ptr(vcpu
);
335 return kvm_vgic_vcpu_init(vcpu
);
338 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
342 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
345 * We might get preempted before the vCPU actually runs, but
346 * over-invalidation doesn't affect correctness.
348 if (*last_ran
!= vcpu
->vcpu_id
) {
349 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
350 *last_ran
= vcpu
->vcpu_id
;
354 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
356 kvm_arm_set_running_vcpu(vcpu
);
361 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
367 kvm_arm_set_running_vcpu(NULL
);
368 kvm_timer_vcpu_put(vcpu
);
371 static void vcpu_power_off(struct kvm_vcpu
*vcpu
)
373 vcpu
->arch
.power_off
= true;
374 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
378 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
379 struct kvm_mp_state
*mp_state
)
381 if (vcpu
->arch
.power_off
)
382 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
384 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
389 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
390 struct kvm_mp_state
*mp_state
)
392 switch (mp_state
->mp_state
) {
393 case KVM_MP_STATE_RUNNABLE
:
394 vcpu
->arch
.power_off
= false;
396 case KVM_MP_STATE_STOPPED
:
397 vcpu_power_off(vcpu
);
407 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
408 * @v: The VCPU pointer
410 * If the guest CPU is not waiting for interrupts or an interrupt line is
411 * asserted, the CPU is by definition runnable.
413 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
415 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
416 && !v
->arch
.power_off
&& !v
->arch
.pause
);
419 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
421 return vcpu_mode_priv(vcpu
);
424 /* Just ensure a guest exit from a particular CPU */
425 static void exit_vm_noop(void *info
)
429 void force_vm_exit(const cpumask_t
*mask
)
432 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
437 * need_new_vmid_gen - check that the VMID is still valid
438 * @kvm: The VM's VMID to check
440 * return true if there is a new generation of VMIDs being used
442 * The hardware supports only 256 values with the value zero reserved for the
443 * host, so we check if an assigned value belongs to a previous generation,
444 * which which requires us to assign a new value. If we're the first to use a
445 * VMID for the new generation, we must flush necessary caches and TLBs on all
448 static bool need_new_vmid_gen(struct kvm
*kvm
)
450 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
454 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
455 * @kvm The guest that we are about to run
457 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
458 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
461 static void update_vttbr(struct kvm
*kvm
)
463 phys_addr_t pgd_phys
;
466 if (!need_new_vmid_gen(kvm
))
469 spin_lock(&kvm_vmid_lock
);
472 * We need to re-check the vmid_gen here to ensure that if another vcpu
473 * already allocated a valid vmid for this vm, then this vcpu should
476 if (!need_new_vmid_gen(kvm
)) {
477 spin_unlock(&kvm_vmid_lock
);
481 /* First user of a new VMID generation? */
482 if (unlikely(kvm_next_vmid
== 0)) {
483 atomic64_inc(&kvm_vmid_gen
);
487 * On SMP we know no other CPUs can use this CPU's or each
488 * other's VMID after force_vm_exit returns since the
489 * kvm_vmid_lock blocks them from reentry to the guest.
491 force_vm_exit(cpu_all_mask
);
493 * Now broadcast TLB + ICACHE invalidation over the inner
494 * shareable domain to make sure all data structures are
497 kvm_call_hyp(__kvm_flush_vm_context
);
500 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
501 kvm
->arch
.vmid
= kvm_next_vmid
;
503 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
505 /* update vttbr to be used with the new vmid */
506 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
507 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
508 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
509 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
511 spin_unlock(&kvm_vmid_lock
);
514 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
516 struct kvm
*kvm
= vcpu
->kvm
;
519 if (likely(vcpu
->arch
.has_run_once
))
522 vcpu
->arch
.has_run_once
= true;
525 * Map the VGIC hardware resources before running a vcpu the first
528 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
529 ret
= kvm_vgic_map_resources(kvm
);
534 ret
= kvm_timer_enable(vcpu
);
538 ret
= kvm_arm_pmu_v3_enable(vcpu
);
543 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
545 return vgic_initialized(kvm
);
548 void kvm_arm_halt_guest(struct kvm
*kvm
)
551 struct kvm_vcpu
*vcpu
;
553 kvm_for_each_vcpu(i
, vcpu
, kvm
)
554 vcpu
->arch
.pause
= true;
555 kvm_make_all_cpus_request(kvm
, KVM_REQ_SLEEP
);
558 void kvm_arm_resume_guest(struct kvm
*kvm
)
561 struct kvm_vcpu
*vcpu
;
563 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
564 vcpu
->arch
.pause
= false;
565 swake_up(kvm_arch_vcpu_wq(vcpu
));
569 static void vcpu_req_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
)));
576 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
577 /* Awaken to handle a signal, request we sleep again later. */
578 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
582 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
584 return vcpu
->arch
.target
>= 0;
587 static void check_vcpu_requests(struct kvm_vcpu
*vcpu
)
589 if (kvm_request_pending(vcpu
)) {
590 if (kvm_check_request(KVM_REQ_SLEEP
, vcpu
))
591 vcpu_req_sleep(vcpu
);
594 * Clear IRQ_PENDING requests that were made to guarantee
595 * that a VCPU sees new virtual interrupts.
597 kvm_check_request(KVM_REQ_IRQ_PENDING
, vcpu
);
602 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
603 * @vcpu: The VCPU pointer
604 * @run: The kvm_run structure pointer used for userspace state exchange
606 * This function is called through the VCPU_RUN ioctl called from user space. It
607 * will execute VM code in a loop until the time slice for the process is used
608 * or some emulation is needed from user space in which case the function will
609 * return with return value 0 and with the kvm_run structure filled in with the
610 * required data for the requested emulation.
612 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
617 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
620 ret
= kvm_vcpu_first_run_init(vcpu
);
624 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
625 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
630 if (run
->immediate_exit
)
633 if (vcpu
->sigset_active
)
634 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
637 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
640 * Check conditions before entering the guest
644 update_vttbr(vcpu
->kvm
);
646 check_vcpu_requests(vcpu
);
649 * Preparing the interrupts to be injected also
650 * involves poking the GIC, which must be done in a
651 * non-preemptible context.
655 /* Flush FP/SIMD state that can't survive guest entry/exit */
656 kvm_fpsimd_flush_cpu_state();
658 kvm_pmu_flush_hwstate(vcpu
);
660 kvm_timer_flush_hwstate(vcpu
);
661 kvm_vgic_flush_hwstate(vcpu
);
666 * If we have a singal pending, or need to notify a userspace
667 * irqchip about timer or PMU level changes, then we exit (and
668 * update the timer level state in kvm_timer_update_run
671 if (signal_pending(current
) ||
672 kvm_timer_should_notify_user(vcpu
) ||
673 kvm_pmu_should_notify_user(vcpu
)) {
675 run
->exit_reason
= KVM_EXIT_INTR
;
679 * Ensure we set mode to IN_GUEST_MODE after we disable
680 * interrupts and before the final VCPU requests check.
681 * See the comment in kvm_vcpu_exiting_guest_mode() and
682 * Documentation/virtual/kvm/vcpu-requests.rst
684 smp_store_mb(vcpu
->mode
, IN_GUEST_MODE
);
686 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
687 kvm_request_pending(vcpu
)) {
688 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
690 kvm_pmu_sync_hwstate(vcpu
);
691 kvm_timer_sync_hwstate(vcpu
);
692 kvm_vgic_sync_hwstate(vcpu
);
697 kvm_arm_setup_debug(vcpu
);
699 /**************************************************************
702 trace_kvm_entry(*vcpu_pc(vcpu
));
703 guest_enter_irqoff();
705 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
707 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
711 *************************************************************/
713 kvm_arm_clear_debug(vcpu
);
716 * We may have taken a host interrupt in HYP mode (ie
717 * while executing the guest). This interrupt is still
718 * pending, as we haven't serviced it yet!
720 * We're now back in SVC mode, with interrupts
721 * disabled. Enabling the interrupts now will have
722 * the effect of taking the interrupt again, in SVC
728 * We do local_irq_enable() before calling guest_exit() so
729 * that if a timer interrupt hits while running the guest we
730 * account that tick as being spent in the guest. We enable
731 * preemption after calling guest_exit() so that if we get
732 * preempted we make sure ticks after that is not counted as
736 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
739 * We must sync the PMU and timer state before the vgic state so
740 * that the vgic can properly sample the updated state of the
743 kvm_pmu_sync_hwstate(vcpu
);
744 kvm_timer_sync_hwstate(vcpu
);
746 kvm_vgic_sync_hwstate(vcpu
);
750 ret
= handle_exit(vcpu
, run
, ret
);
753 /* Tell userspace about in-kernel device output levels */
754 if (unlikely(!irqchip_in_kernel(vcpu
->kvm
))) {
755 kvm_timer_update_run(vcpu
);
756 kvm_pmu_update_run(vcpu
);
759 if (vcpu
->sigset_active
)
760 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
764 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
770 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
771 bit_index
= __ffs(HCR_VI
);
772 else /* KVM_ARM_IRQ_CPU_FIQ */
773 bit_index
= __ffs(HCR_VF
);
775 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
777 set
= test_and_set_bit(bit_index
, ptr
);
779 set
= test_and_clear_bit(bit_index
, ptr
);
782 * If we didn't change anything, no need to wake up or kick other CPUs
788 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
789 * trigger a world-switch round on the running physical CPU to set the
790 * virtual IRQ/FIQ fields in the HCR appropriately.
792 kvm_make_request(KVM_REQ_IRQ_PENDING
, vcpu
);
798 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
801 u32 irq
= irq_level
->irq
;
802 unsigned int irq_type
, vcpu_idx
, irq_num
;
803 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
804 struct kvm_vcpu
*vcpu
= NULL
;
805 bool level
= irq_level
->level
;
807 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
808 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
809 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
811 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
814 case KVM_ARM_IRQ_TYPE_CPU
:
815 if (irqchip_in_kernel(kvm
))
818 if (vcpu_idx
>= nrcpus
)
821 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
825 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
828 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
829 case KVM_ARM_IRQ_TYPE_PPI
:
830 if (!irqchip_in_kernel(kvm
))
833 if (vcpu_idx
>= nrcpus
)
836 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
840 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
843 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
, NULL
);
844 case KVM_ARM_IRQ_TYPE_SPI
:
845 if (!irqchip_in_kernel(kvm
))
848 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
851 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
, NULL
);
857 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
858 const struct kvm_vcpu_init
*init
)
861 int phys_target
= kvm_target_cpu();
863 if (init
->target
!= phys_target
)
867 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
868 * use the same target.
870 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
873 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
874 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
875 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
877 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
881 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
882 * use the same feature set.
884 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
885 test_bit(i
, vcpu
->arch
.features
) != set
)
889 set_bit(i
, vcpu
->arch
.features
);
892 vcpu
->arch
.target
= phys_target
;
894 /* Now we know what it is, we can reset it. */
895 return kvm_reset_vcpu(vcpu
);
899 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
900 struct kvm_vcpu_init
*init
)
904 ret
= kvm_vcpu_set_target(vcpu
, init
);
909 * Ensure a rebooted VM will fault in RAM pages and detect if the
910 * guest MMU is turned off and flush the caches as needed.
912 if (vcpu
->arch
.has_run_once
)
913 stage2_unmap_vm(vcpu
->kvm
);
915 vcpu_reset_hcr(vcpu
);
918 * Handle the "start in power-off" case.
920 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
921 vcpu_power_off(vcpu
);
923 vcpu
->arch
.power_off
= false;
928 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
929 struct kvm_device_attr
*attr
)
933 switch (attr
->group
) {
935 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
942 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
943 struct kvm_device_attr
*attr
)
947 switch (attr
->group
) {
949 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
956 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
957 struct kvm_device_attr
*attr
)
961 switch (attr
->group
) {
963 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
970 long kvm_arch_vcpu_ioctl(struct file
*filp
,
971 unsigned int ioctl
, unsigned long arg
)
973 struct kvm_vcpu
*vcpu
= filp
->private_data
;
974 void __user
*argp
= (void __user
*)arg
;
975 struct kvm_device_attr attr
;
978 case KVM_ARM_VCPU_INIT
: {
979 struct kvm_vcpu_init init
;
981 if (copy_from_user(&init
, argp
, sizeof(init
)))
984 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
986 case KVM_SET_ONE_REG
:
987 case KVM_GET_ONE_REG
: {
988 struct kvm_one_reg reg
;
990 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
993 if (copy_from_user(®
, argp
, sizeof(reg
)))
995 if (ioctl
== KVM_SET_ONE_REG
)
996 return kvm_arm_set_reg(vcpu
, ®
);
998 return kvm_arm_get_reg(vcpu
, ®
);
1000 case KVM_GET_REG_LIST
: {
1001 struct kvm_reg_list __user
*user_list
= argp
;
1002 struct kvm_reg_list reg_list
;
1005 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1008 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
1011 reg_list
.n
= kvm_arm_num_regs(vcpu
);
1012 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
1016 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
1018 case KVM_SET_DEVICE_ATTR
: {
1019 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1021 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
1023 case KVM_GET_DEVICE_ATTR
: {
1024 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1026 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
1028 case KVM_HAS_DEVICE_ATTR
: {
1029 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1031 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
1039 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1040 * @kvm: kvm instance
1041 * @log: slot id and address to which we copy the log
1043 * Steps 1-4 below provide general overview of dirty page logging. See
1044 * kvm_get_dirty_log_protect() function description for additional details.
1046 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1047 * always flush the TLB (step 4) even if previous step failed and the dirty
1048 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1049 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1050 * writes will be marked dirty for next log read.
1052 * 1. Take a snapshot of the bit and clear it if needed.
1053 * 2. Write protect the corresponding page.
1054 * 3. Copy the snapshot to the userspace.
1055 * 4. Flush TLB's if needed.
1057 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1059 bool is_dirty
= false;
1062 mutex_lock(&kvm
->slots_lock
);
1064 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
1067 kvm_flush_remote_tlbs(kvm
);
1069 mutex_unlock(&kvm
->slots_lock
);
1073 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1074 struct kvm_arm_device_addr
*dev_addr
)
1076 unsigned long dev_id
, type
;
1078 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1079 KVM_ARM_DEVICE_ID_SHIFT
;
1080 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1081 KVM_ARM_DEVICE_TYPE_SHIFT
;
1084 case KVM_ARM_DEVICE_VGIC_V2
:
1087 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1093 long kvm_arch_vm_ioctl(struct file
*filp
,
1094 unsigned int ioctl
, unsigned long arg
)
1096 struct kvm
*kvm
= filp
->private_data
;
1097 void __user
*argp
= (void __user
*)arg
;
1100 case KVM_CREATE_IRQCHIP
: {
1104 mutex_lock(&kvm
->lock
);
1105 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1106 mutex_unlock(&kvm
->lock
);
1109 case KVM_ARM_SET_DEVICE_ADDR
: {
1110 struct kvm_arm_device_addr dev_addr
;
1112 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1114 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1116 case KVM_ARM_PREFERRED_TARGET
: {
1118 struct kvm_vcpu_init init
;
1120 err
= kvm_vcpu_preferred_target(&init
);
1124 if (copy_to_user(argp
, &init
, sizeof(init
)))
1134 static void cpu_init_hyp_mode(void *dummy
)
1136 phys_addr_t pgd_ptr
;
1137 unsigned long hyp_stack_ptr
;
1138 unsigned long stack_page
;
1139 unsigned long vector_ptr
;
1141 /* Switch from the HYP stub to our own HYP init vector */
1142 __hyp_set_vectors(kvm_get_idmap_vector());
1144 pgd_ptr
= kvm_mmu_get_httbr();
1145 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1146 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1147 vector_ptr
= (unsigned long)kvm_ksym_ref(__kvm_hyp_vector
);
1149 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1150 __cpu_init_stage2();
1152 kvm_arm_init_debug();
1155 static void cpu_hyp_reset(void)
1157 if (!is_kernel_in_hyp_mode())
1158 __hyp_reset_vectors();
1161 static void cpu_hyp_reinit(void)
1165 if (is_kernel_in_hyp_mode()) {
1167 * __cpu_init_stage2() is safe to call even if the PM
1168 * event was cancelled before the CPU was reset.
1170 __cpu_init_stage2();
1171 kvm_timer_init_vhe();
1173 cpu_init_hyp_mode(NULL
);
1177 kvm_vgic_init_cpu_hardware();
1180 static void _kvm_arch_hardware_enable(void *discard
)
1182 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1184 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1188 int kvm_arch_hardware_enable(void)
1190 _kvm_arch_hardware_enable(NULL
);
1194 static void _kvm_arch_hardware_disable(void *discard
)
1196 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1198 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1202 void kvm_arch_hardware_disable(void)
1204 _kvm_arch_hardware_disable(NULL
);
1207 #ifdef CONFIG_CPU_PM
1208 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1213 * kvm_arm_hardware_enabled is left with its old value over
1214 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1219 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1221 * don't update kvm_arm_hardware_enabled here
1222 * so that the hardware will be re-enabled
1223 * when we resume. See below.
1229 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1230 /* The hardware was enabled before suspend. */
1240 static struct notifier_block hyp_init_cpu_pm_nb
= {
1241 .notifier_call
= hyp_init_cpu_pm_notifier
,
1244 static void __init
hyp_cpu_pm_init(void)
1246 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1248 static void __init
hyp_cpu_pm_exit(void)
1250 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1253 static inline void hyp_cpu_pm_init(void)
1256 static inline void hyp_cpu_pm_exit(void)
1261 static void teardown_common_resources(void)
1263 free_percpu(kvm_host_cpu_state
);
1266 static int init_common_resources(void)
1268 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1269 if (!kvm_host_cpu_state
) {
1270 kvm_err("Cannot allocate host CPU state\n");
1274 /* set size of VMID supported by CPU */
1275 kvm_vmid_bits
= kvm_get_vmid_bits();
1276 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1281 static int init_subsystems(void)
1286 * Enable hardware so that subsystem initialisation can access EL2.
1288 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1291 * Register CPU lower-power notifier
1296 * Init HYP view of VGIC
1298 err
= kvm_vgic_hyp_init();
1301 vgic_present
= true;
1305 vgic_present
= false;
1313 * Init HYP architected timer support
1315 err
= kvm_timer_hyp_init();
1320 kvm_coproc_table_init();
1323 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1328 static void teardown_hyp_mode(void)
1333 for_each_possible_cpu(cpu
)
1334 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1339 * Inits Hyp-mode on all online CPUs
1341 static int init_hyp_mode(void)
1347 * Allocate Hyp PGD and setup Hyp identity mapping
1349 err
= kvm_mmu_init();
1354 * Allocate stack pages for Hypervisor-mode
1356 for_each_possible_cpu(cpu
) {
1357 unsigned long stack_page
;
1359 stack_page
= __get_free_page(GFP_KERNEL
);
1365 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1369 * Map the Hyp-code called directly from the host
1371 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1372 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1374 kvm_err("Cannot map world-switch code\n");
1378 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1379 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1381 kvm_err("Cannot map rodata section\n");
1385 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1386 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1388 kvm_err("Cannot map bss section\n");
1393 * Map the Hyp stack pages
1395 for_each_possible_cpu(cpu
) {
1396 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1397 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1401 kvm_err("Cannot map hyp stack\n");
1406 for_each_possible_cpu(cpu
) {
1407 kvm_cpu_context_t
*cpu_ctxt
;
1409 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1410 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1413 kvm_err("Cannot map host CPU state: %d\n", err
);
1421 teardown_hyp_mode();
1422 kvm_err("error initializing Hyp mode: %d\n", err
);
1426 static void check_kvm_target_cpu(void *ret
)
1428 *(int *)ret
= kvm_target_cpu();
1431 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1433 struct kvm_vcpu
*vcpu
;
1436 mpidr
&= MPIDR_HWID_BITMASK
;
1437 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1438 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1445 * Initialize Hyp-mode and memory mappings on all CPUs.
1447 int kvm_arch_init(void *opaque
)
1453 if (!is_hyp_mode_available()) {
1454 kvm_err("HYP mode not available\n");
1458 for_each_online_cpu(cpu
) {
1459 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1461 kvm_err("Error, CPU %d not supported!\n", cpu
);
1466 err
= init_common_resources();
1470 in_hyp_mode
= is_kernel_in_hyp_mode();
1473 err
= init_hyp_mode();
1478 err
= init_subsystems();
1483 kvm_info("VHE mode initialized successfully\n");
1485 kvm_info("Hyp mode initialized successfully\n");
1491 teardown_hyp_mode();
1493 teardown_common_resources();
1497 /* NOP: Compiling as a module not supported */
1498 void kvm_arch_exit(void)
1500 kvm_perf_teardown();
1503 static int arm_init(void)
1505 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1509 module_init(arm_init
);