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/module.h>
24 #include <linux/vmalloc.h>
26 #include <linux/mman.h>
27 #include <linux/sched.h>
28 #include <linux/kvm.h>
29 #include <trace/events/kvm.h>
30 #include <kvm/arm_pmu.h>
32 #define CREATE_TRACE_POINTS
35 #include <asm/uaccess.h>
36 #include <asm/ptrace.h>
38 #include <asm/tlbflush.h>
39 #include <asm/cacheflush.h>
41 #include <asm/kvm_arm.h>
42 #include <asm/kvm_asm.h>
43 #include <asm/kvm_mmu.h>
44 #include <asm/kvm_emulate.h>
45 #include <asm/kvm_coproc.h>
46 #include <asm/kvm_psci.h>
47 #include <asm/sections.h>
50 __asm__(".arch_extension virt");
53 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
54 static kvm_cpu_context_t __percpu
*kvm_host_cpu_state
;
55 static unsigned long hyp_default_vectors
;
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 ret
= kvm_alloc_stage2_pgd(kvm
);
125 ret
= create_hyp_mappings(kvm
, kvm
+ 1);
127 goto out_free_stage2_pgd
;
129 kvm_vgic_early_init(kvm
);
132 /* Mark the initial VMID generation invalid */
133 kvm
->arch
.vmid_gen
= 0;
135 /* The maximum number of VCPUs is limited by the host's GIC model */
136 kvm
->arch
.max_vcpus
= vgic_present
?
137 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
141 kvm_free_stage2_pgd(kvm
);
146 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
148 return VM_FAULT_SIGBUS
;
153 * kvm_arch_destroy_vm - destroy the VM data structure
154 * @kvm: pointer to the KVM struct
156 void kvm_arch_destroy_vm(struct kvm
*kvm
)
160 kvm_free_stage2_pgd(kvm
);
162 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
164 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
165 kvm
->vcpus
[i
] = NULL
;
169 kvm_vgic_destroy(kvm
);
172 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
176 case KVM_CAP_IRQCHIP
:
179 case KVM_CAP_IOEVENTFD
:
180 case KVM_CAP_DEVICE_CTRL
:
181 case KVM_CAP_USER_MEMORY
:
182 case KVM_CAP_SYNC_MMU
:
183 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
184 case KVM_CAP_ONE_REG
:
185 case KVM_CAP_ARM_PSCI
:
186 case KVM_CAP_ARM_PSCI_0_2
:
187 case KVM_CAP_READONLY_MEM
:
188 case KVM_CAP_MP_STATE
:
191 case KVM_CAP_COALESCED_MMIO
:
192 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
194 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
197 case KVM_CAP_NR_VCPUS
:
198 r
= num_online_cpus();
200 case KVM_CAP_MAX_VCPUS
:
204 r
= kvm_arch_dev_ioctl_check_extension(ext
);
210 long kvm_arch_dev_ioctl(struct file
*filp
,
211 unsigned int ioctl
, unsigned long arg
)
217 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
220 struct kvm_vcpu
*vcpu
;
222 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
227 if (id
>= kvm
->arch
.max_vcpus
) {
232 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
238 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
242 err
= create_hyp_mappings(vcpu
, vcpu
+ 1);
248 kvm_vcpu_uninit(vcpu
);
250 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
255 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
257 kvm_vgic_vcpu_early_init(vcpu
);
260 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
262 kvm_mmu_free_memory_caches(vcpu
);
263 kvm_timer_vcpu_terminate(vcpu
);
264 kvm_vgic_vcpu_destroy(vcpu
);
265 kvm_pmu_vcpu_destroy(vcpu
);
266 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
269 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
271 kvm_arch_vcpu_free(vcpu
);
274 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
276 return kvm_timer_should_fire(vcpu
);
279 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
281 kvm_timer_schedule(vcpu
);
284 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
286 kvm_timer_unschedule(vcpu
);
289 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
291 /* Force users to call KVM_ARM_VCPU_INIT */
292 vcpu
->arch
.target
= -1;
293 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
295 /* Set up the timer */
296 kvm_timer_vcpu_init(vcpu
);
298 kvm_arm_reset_debug_ptr(vcpu
);
303 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
306 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
308 kvm_arm_set_running_vcpu(vcpu
);
311 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
314 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
315 * if the vcpu is no longer assigned to a cpu. This is used for the
316 * optimized make_all_cpus_request path.
320 kvm_arm_set_running_vcpu(NULL
);
321 kvm_timer_vcpu_put(vcpu
);
324 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
325 struct kvm_mp_state
*mp_state
)
327 if (vcpu
->arch
.power_off
)
328 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
330 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
335 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
336 struct kvm_mp_state
*mp_state
)
338 switch (mp_state
->mp_state
) {
339 case KVM_MP_STATE_RUNNABLE
:
340 vcpu
->arch
.power_off
= false;
342 case KVM_MP_STATE_STOPPED
:
343 vcpu
->arch
.power_off
= true;
353 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
354 * @v: The VCPU pointer
356 * If the guest CPU is not waiting for interrupts or an interrupt line is
357 * asserted, the CPU is by definition runnable.
359 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
361 return ((!!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
362 && !v
->arch
.power_off
&& !v
->arch
.pause
);
365 /* Just ensure a guest exit from a particular CPU */
366 static void exit_vm_noop(void *info
)
370 void force_vm_exit(const cpumask_t
*mask
)
373 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
378 * need_new_vmid_gen - check that the VMID is still valid
379 * @kvm: The VM's VMID to check
381 * return true if there is a new generation of VMIDs being used
383 * The hardware supports only 256 values with the value zero reserved for the
384 * host, so we check if an assigned value belongs to a previous generation,
385 * which which requires us to assign a new value. If we're the first to use a
386 * VMID for the new generation, we must flush necessary caches and TLBs on all
389 static bool need_new_vmid_gen(struct kvm
*kvm
)
391 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
395 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
396 * @kvm The guest that we are about to run
398 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
399 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
402 static void update_vttbr(struct kvm
*kvm
)
404 phys_addr_t pgd_phys
;
407 if (!need_new_vmid_gen(kvm
))
410 spin_lock(&kvm_vmid_lock
);
413 * We need to re-check the vmid_gen here to ensure that if another vcpu
414 * already allocated a valid vmid for this vm, then this vcpu should
417 if (!need_new_vmid_gen(kvm
)) {
418 spin_unlock(&kvm_vmid_lock
);
422 /* First user of a new VMID generation? */
423 if (unlikely(kvm_next_vmid
== 0)) {
424 atomic64_inc(&kvm_vmid_gen
);
428 * On SMP we know no other CPUs can use this CPU's or each
429 * other's VMID after force_vm_exit returns since the
430 * kvm_vmid_lock blocks them from reentry to the guest.
432 force_vm_exit(cpu_all_mask
);
434 * Now broadcast TLB + ICACHE invalidation over the inner
435 * shareable domain to make sure all data structures are
438 kvm_call_hyp(__kvm_flush_vm_context
);
441 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
442 kvm
->arch
.vmid
= kvm_next_vmid
;
444 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
446 /* update vttbr to be used with the new vmid */
447 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
448 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
449 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
450 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
452 spin_unlock(&kvm_vmid_lock
);
455 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
457 struct kvm
*kvm
= vcpu
->kvm
;
460 if (likely(vcpu
->arch
.has_run_once
))
463 vcpu
->arch
.has_run_once
= true;
466 * Map the VGIC hardware resources before running a vcpu the first
469 if (unlikely(irqchip_in_kernel(kvm
) && !vgic_ready(kvm
))) {
470 ret
= kvm_vgic_map_resources(kvm
);
476 * Enable the arch timers only if we have an in-kernel VGIC
477 * and it has been properly initialized, since we cannot handle
478 * interrupts from the virtual timer with a userspace gic.
480 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
481 ret
= kvm_timer_enable(vcpu
);
486 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
488 return vgic_initialized(kvm
);
491 void kvm_arm_halt_guest(struct kvm
*kvm
)
494 struct kvm_vcpu
*vcpu
;
496 kvm_for_each_vcpu(i
, vcpu
, kvm
)
497 vcpu
->arch
.pause
= true;
498 kvm_make_all_cpus_request(kvm
, KVM_REQ_VCPU_EXIT
);
501 void kvm_arm_halt_vcpu(struct kvm_vcpu
*vcpu
)
503 vcpu
->arch
.pause
= true;
507 void kvm_arm_resume_vcpu(struct kvm_vcpu
*vcpu
)
509 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
511 vcpu
->arch
.pause
= false;
515 void kvm_arm_resume_guest(struct kvm
*kvm
)
518 struct kvm_vcpu
*vcpu
;
520 kvm_for_each_vcpu(i
, vcpu
, kvm
)
521 kvm_arm_resume_vcpu(vcpu
);
524 static void vcpu_sleep(struct kvm_vcpu
*vcpu
)
526 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
528 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
529 (!vcpu
->arch
.pause
)));
532 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
534 return vcpu
->arch
.target
>= 0;
538 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
539 * @vcpu: The VCPU pointer
540 * @run: The kvm_run structure pointer used for userspace state exchange
542 * This function is called through the VCPU_RUN ioctl called from user space. It
543 * will execute VM code in a loop until the time slice for the process is used
544 * or some emulation is needed from user space in which case the function will
545 * return with return value 0 and with the kvm_run structure filled in with the
546 * required data for the requested emulation.
548 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
553 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
556 ret
= kvm_vcpu_first_run_init(vcpu
);
560 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
561 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
566 if (vcpu
->sigset_active
)
567 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
570 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
573 * Check conditions before entering the guest
577 update_vttbr(vcpu
->kvm
);
579 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
)
583 * Preparing the interrupts to be injected also
584 * involves poking the GIC, which must be done in a
585 * non-preemptible context.
588 kvm_pmu_flush_hwstate(vcpu
);
589 kvm_timer_flush_hwstate(vcpu
);
590 kvm_vgic_flush_hwstate(vcpu
);
595 * Re-check atomic conditions
597 if (signal_pending(current
)) {
599 run
->exit_reason
= KVM_EXIT_INTR
;
602 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
603 vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
605 kvm_pmu_sync_hwstate(vcpu
);
606 kvm_timer_sync_hwstate(vcpu
);
607 kvm_vgic_sync_hwstate(vcpu
);
612 kvm_arm_setup_debug(vcpu
);
614 /**************************************************************
617 trace_kvm_entry(*vcpu_pc(vcpu
));
619 vcpu
->mode
= IN_GUEST_MODE
;
621 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
623 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
627 *************************************************************/
629 kvm_arm_clear_debug(vcpu
);
632 * We may have taken a host interrupt in HYP mode (ie
633 * while executing the guest). This interrupt is still
634 * pending, as we haven't serviced it yet!
636 * We're now back in SVC mode, with interrupts
637 * disabled. Enabling the interrupts now will have
638 * the effect of taking the interrupt again, in SVC
644 * We do local_irq_enable() before calling kvm_guest_exit() so
645 * that if a timer interrupt hits while running the guest we
646 * account that tick as being spent in the guest. We enable
647 * preemption after calling kvm_guest_exit() so that if we get
648 * preempted we make sure ticks after that is not counted as
652 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
655 * We must sync the PMU and timer state before the vgic state so
656 * that the vgic can properly sample the updated state of the
659 kvm_pmu_sync_hwstate(vcpu
);
660 kvm_timer_sync_hwstate(vcpu
);
662 kvm_vgic_sync_hwstate(vcpu
);
666 ret
= handle_exit(vcpu
, run
, ret
);
669 if (vcpu
->sigset_active
)
670 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
674 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
680 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
681 bit_index
= __ffs(HCR_VI
);
682 else /* KVM_ARM_IRQ_CPU_FIQ */
683 bit_index
= __ffs(HCR_VF
);
685 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
687 set
= test_and_set_bit(bit_index
, ptr
);
689 set
= test_and_clear_bit(bit_index
, ptr
);
692 * If we didn't change anything, no need to wake up or kick other CPUs
698 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
699 * trigger a world-switch round on the running physical CPU to set the
700 * virtual IRQ/FIQ fields in the HCR appropriately.
707 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
710 u32 irq
= irq_level
->irq
;
711 unsigned int irq_type
, vcpu_idx
, irq_num
;
712 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
713 struct kvm_vcpu
*vcpu
= NULL
;
714 bool level
= irq_level
->level
;
716 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
717 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
718 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
720 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
723 case KVM_ARM_IRQ_TYPE_CPU
:
724 if (irqchip_in_kernel(kvm
))
727 if (vcpu_idx
>= nrcpus
)
730 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
734 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
737 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
738 case KVM_ARM_IRQ_TYPE_PPI
:
739 if (!irqchip_in_kernel(kvm
))
742 if (vcpu_idx
>= nrcpus
)
745 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
749 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
752 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
753 case KVM_ARM_IRQ_TYPE_SPI
:
754 if (!irqchip_in_kernel(kvm
))
757 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
760 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
766 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
767 const struct kvm_vcpu_init
*init
)
770 int phys_target
= kvm_target_cpu();
772 if (init
->target
!= phys_target
)
776 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
777 * use the same target.
779 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
782 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
783 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
784 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
786 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
790 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
791 * use the same feature set.
793 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
794 test_bit(i
, vcpu
->arch
.features
) != set
)
798 set_bit(i
, vcpu
->arch
.features
);
801 vcpu
->arch
.target
= phys_target
;
803 /* Now we know what it is, we can reset it. */
804 return kvm_reset_vcpu(vcpu
);
808 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
809 struct kvm_vcpu_init
*init
)
813 ret
= kvm_vcpu_set_target(vcpu
, init
);
818 * Ensure a rebooted VM will fault in RAM pages and detect if the
819 * guest MMU is turned off and flush the caches as needed.
821 if (vcpu
->arch
.has_run_once
)
822 stage2_unmap_vm(vcpu
->kvm
);
824 vcpu_reset_hcr(vcpu
);
827 * Handle the "start in power-off" case.
829 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
830 vcpu
->arch
.power_off
= true;
832 vcpu
->arch
.power_off
= false;
837 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
838 struct kvm_device_attr
*attr
)
842 switch (attr
->group
) {
844 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
851 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
852 struct kvm_device_attr
*attr
)
856 switch (attr
->group
) {
858 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
865 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
866 struct kvm_device_attr
*attr
)
870 switch (attr
->group
) {
872 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
879 long kvm_arch_vcpu_ioctl(struct file
*filp
,
880 unsigned int ioctl
, unsigned long arg
)
882 struct kvm_vcpu
*vcpu
= filp
->private_data
;
883 void __user
*argp
= (void __user
*)arg
;
884 struct kvm_device_attr attr
;
887 case KVM_ARM_VCPU_INIT
: {
888 struct kvm_vcpu_init init
;
890 if (copy_from_user(&init
, argp
, sizeof(init
)))
893 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
895 case KVM_SET_ONE_REG
:
896 case KVM_GET_ONE_REG
: {
897 struct kvm_one_reg reg
;
899 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
902 if (copy_from_user(®
, argp
, sizeof(reg
)))
904 if (ioctl
== KVM_SET_ONE_REG
)
905 return kvm_arm_set_reg(vcpu
, ®
);
907 return kvm_arm_get_reg(vcpu
, ®
);
909 case KVM_GET_REG_LIST
: {
910 struct kvm_reg_list __user
*user_list
= argp
;
911 struct kvm_reg_list reg_list
;
914 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
917 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
920 reg_list
.n
= kvm_arm_num_regs(vcpu
);
921 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
925 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
927 case KVM_SET_DEVICE_ATTR
: {
928 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
930 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
932 case KVM_GET_DEVICE_ATTR
: {
933 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
935 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
937 case KVM_HAS_DEVICE_ATTR
: {
938 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
940 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
948 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
950 * @log: slot id and address to which we copy the log
952 * Steps 1-4 below provide general overview of dirty page logging. See
953 * kvm_get_dirty_log_protect() function description for additional details.
955 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
956 * always flush the TLB (step 4) even if previous step failed and the dirty
957 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
958 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
959 * writes will be marked dirty for next log read.
961 * 1. Take a snapshot of the bit and clear it if needed.
962 * 2. Write protect the corresponding page.
963 * 3. Copy the snapshot to the userspace.
964 * 4. Flush TLB's if needed.
966 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
968 bool is_dirty
= false;
971 mutex_lock(&kvm
->slots_lock
);
973 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
976 kvm_flush_remote_tlbs(kvm
);
978 mutex_unlock(&kvm
->slots_lock
);
982 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
983 struct kvm_arm_device_addr
*dev_addr
)
985 unsigned long dev_id
, type
;
987 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
988 KVM_ARM_DEVICE_ID_SHIFT
;
989 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
990 KVM_ARM_DEVICE_TYPE_SHIFT
;
993 case KVM_ARM_DEVICE_VGIC_V2
:
996 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1002 long kvm_arch_vm_ioctl(struct file
*filp
,
1003 unsigned int ioctl
, unsigned long arg
)
1005 struct kvm
*kvm
= filp
->private_data
;
1006 void __user
*argp
= (void __user
*)arg
;
1009 case KVM_CREATE_IRQCHIP
: {
1012 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1014 case KVM_ARM_SET_DEVICE_ADDR
: {
1015 struct kvm_arm_device_addr dev_addr
;
1017 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1019 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1021 case KVM_ARM_PREFERRED_TARGET
: {
1023 struct kvm_vcpu_init init
;
1025 err
= kvm_vcpu_preferred_target(&init
);
1029 if (copy_to_user(argp
, &init
, sizeof(init
)))
1039 static void cpu_init_hyp_mode(void *dummy
)
1041 phys_addr_t boot_pgd_ptr
;
1042 phys_addr_t pgd_ptr
;
1043 unsigned long hyp_stack_ptr
;
1044 unsigned long stack_page
;
1045 unsigned long vector_ptr
;
1047 /* Switch from the HYP stub to our own HYP init vector */
1048 __hyp_set_vectors(kvm_get_idmap_vector());
1050 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
1051 pgd_ptr
= kvm_mmu_get_httbr();
1052 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1053 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1054 vector_ptr
= (unsigned long)kvm_ksym_ref(__kvm_hyp_vector
);
1056 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1057 __cpu_init_stage2();
1059 kvm_arm_init_debug();
1062 static void cpu_hyp_reinit(void)
1064 if (is_kernel_in_hyp_mode()) {
1066 * __cpu_init_stage2() is safe to call even if the PM
1067 * event was cancelled before the CPU was reset.
1069 __cpu_init_stage2();
1071 if (__hyp_get_vectors() == hyp_default_vectors
)
1072 cpu_init_hyp_mode(NULL
);
1076 static void cpu_hyp_reset(void)
1078 phys_addr_t boot_pgd_ptr
;
1079 phys_addr_t phys_idmap_start
;
1081 if (!is_kernel_in_hyp_mode()) {
1082 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
1083 phys_idmap_start
= kvm_get_idmap_start();
1085 __cpu_reset_hyp_mode(boot_pgd_ptr
, phys_idmap_start
);
1089 static void _kvm_arch_hardware_enable(void *discard
)
1091 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1093 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1097 int kvm_arch_hardware_enable(void)
1099 _kvm_arch_hardware_enable(NULL
);
1103 static void _kvm_arch_hardware_disable(void *discard
)
1105 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1107 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1111 void kvm_arch_hardware_disable(void)
1113 _kvm_arch_hardware_disable(NULL
);
1116 #ifdef CONFIG_CPU_PM
1117 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1122 * kvm_arm_hardware_enabled is left with its old value over
1123 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1128 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1130 * don't update kvm_arm_hardware_enabled here
1131 * so that the hardware will be re-enabled
1132 * when we resume. See below.
1138 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1139 /* The hardware was enabled before suspend. */
1149 static struct notifier_block hyp_init_cpu_pm_nb
= {
1150 .notifier_call
= hyp_init_cpu_pm_notifier
,
1153 static void __init
hyp_cpu_pm_init(void)
1155 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1157 static void __init
hyp_cpu_pm_exit(void)
1159 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1162 static inline void hyp_cpu_pm_init(void)
1165 static inline void hyp_cpu_pm_exit(void)
1170 static void teardown_common_resources(void)
1172 free_percpu(kvm_host_cpu_state
);
1175 static int init_common_resources(void)
1177 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1178 if (!kvm_host_cpu_state
) {
1179 kvm_err("Cannot allocate host CPU state\n");
1186 static int init_subsystems(void)
1191 * Enable hardware so that subsystem initialisation can access EL2.
1193 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1196 * Register CPU lower-power notifier
1201 * Init HYP view of VGIC
1203 err
= kvm_vgic_hyp_init();
1206 vgic_present
= true;
1210 vgic_present
= false;
1218 * Init HYP architected timer support
1220 err
= kvm_timer_hyp_init();
1225 kvm_coproc_table_init();
1228 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1233 static void teardown_hyp_mode(void)
1237 if (is_kernel_in_hyp_mode())
1241 for_each_possible_cpu(cpu
)
1242 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1246 static int init_vhe_mode(void)
1248 /* set size of VMID supported by CPU */
1249 kvm_vmid_bits
= kvm_get_vmid_bits();
1250 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1252 kvm_info("VHE mode initialized successfully\n");
1257 * Inits Hyp-mode on all online CPUs
1259 static int init_hyp_mode(void)
1265 * Allocate Hyp PGD and setup Hyp identity mapping
1267 err
= kvm_mmu_init();
1272 * It is probably enough to obtain the default on one
1273 * CPU. It's unlikely to be different on the others.
1275 hyp_default_vectors
= __hyp_get_vectors();
1278 * Allocate stack pages for Hypervisor-mode
1280 for_each_possible_cpu(cpu
) {
1281 unsigned long stack_page
;
1283 stack_page
= __get_free_page(GFP_KERNEL
);
1289 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1293 * Map the Hyp-code called directly from the host
1295 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1296 kvm_ksym_ref(__hyp_text_end
));
1298 kvm_err("Cannot map world-switch code\n");
1302 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1303 kvm_ksym_ref(__end_rodata
));
1305 kvm_err("Cannot map rodata section\n");
1310 * Map the Hyp stack pages
1312 for_each_possible_cpu(cpu
) {
1313 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1314 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
1317 kvm_err("Cannot map hyp stack\n");
1322 for_each_possible_cpu(cpu
) {
1323 kvm_cpu_context_t
*cpu_ctxt
;
1325 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1326 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1329 kvm_err("Cannot map host CPU state: %d\n", err
);
1334 #ifndef CONFIG_HOTPLUG_CPU
1335 free_boot_hyp_pgd();
1338 /* set size of VMID supported by CPU */
1339 kvm_vmid_bits
= kvm_get_vmid_bits();
1340 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1342 kvm_info("Hyp mode initialized successfully\n");
1347 teardown_hyp_mode();
1348 kvm_err("error initializing Hyp mode: %d\n", err
);
1352 static void check_kvm_target_cpu(void *ret
)
1354 *(int *)ret
= kvm_target_cpu();
1357 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1359 struct kvm_vcpu
*vcpu
;
1362 mpidr
&= MPIDR_HWID_BITMASK
;
1363 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1364 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1371 * Initialize Hyp-mode and memory mappings on all CPUs.
1373 int kvm_arch_init(void *opaque
)
1378 if (!is_hyp_mode_available()) {
1379 kvm_err("HYP mode not available\n");
1383 for_each_online_cpu(cpu
) {
1384 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1386 kvm_err("Error, CPU %d not supported!\n", cpu
);
1391 err
= init_common_resources();
1395 if (is_kernel_in_hyp_mode())
1396 err
= init_vhe_mode();
1398 err
= init_hyp_mode();
1402 err
= init_subsystems();
1409 teardown_hyp_mode();
1411 teardown_common_resources();
1415 /* NOP: Compiling as a module not supported */
1416 void kvm_arch_exit(void)
1418 kvm_perf_teardown();
1421 static int arm_init(void)
1423 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1427 module_init(arm_init
);