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 checkt
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 kvm_timer_enable(kvm
);
486 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
488 return vgic_initialized(kvm
);
491 static void kvm_arm_halt_guest(struct kvm
*kvm
) __maybe_unused
;
492 static void kvm_arm_resume_guest(struct kvm
*kvm
) __maybe_unused
;
494 static void kvm_arm_halt_guest(struct kvm
*kvm
)
497 struct kvm_vcpu
*vcpu
;
499 kvm_for_each_vcpu(i
, vcpu
, kvm
)
500 vcpu
->arch
.pause
= true;
501 force_vm_exit(cpu_all_mask
);
504 static void kvm_arm_resume_guest(struct kvm
*kvm
)
507 struct kvm_vcpu
*vcpu
;
509 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
510 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
512 vcpu
->arch
.pause
= false;
517 static void vcpu_sleep(struct kvm_vcpu
*vcpu
)
519 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
521 swait_event_interruptible(*wq
, ((!vcpu
->arch
.power_off
) &&
522 (!vcpu
->arch
.pause
)));
525 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
527 return vcpu
->arch
.target
>= 0;
531 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
532 * @vcpu: The VCPU pointer
533 * @run: The kvm_run structure pointer used for userspace state exchange
535 * This function is called through the VCPU_RUN ioctl called from user space. It
536 * will execute VM code in a loop until the time slice for the process is used
537 * or some emulation is needed from user space in which case the function will
538 * return with return value 0 and with the kvm_run structure filled in with the
539 * required data for the requested emulation.
541 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
546 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
549 ret
= kvm_vcpu_first_run_init(vcpu
);
553 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
554 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
559 if (vcpu
->sigset_active
)
560 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
563 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
566 * Check conditions before entering the guest
570 update_vttbr(vcpu
->kvm
);
572 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
)
576 * Preparing the interrupts to be injected also
577 * involves poking the GIC, which must be done in a
578 * non-preemptible context.
581 kvm_pmu_flush_hwstate(vcpu
);
582 kvm_timer_flush_hwstate(vcpu
);
583 kvm_vgic_flush_hwstate(vcpu
);
588 * Re-check atomic conditions
590 if (signal_pending(current
)) {
592 run
->exit_reason
= KVM_EXIT_INTR
;
595 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
596 vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
598 kvm_pmu_sync_hwstate(vcpu
);
599 kvm_timer_sync_hwstate(vcpu
);
600 kvm_vgic_sync_hwstate(vcpu
);
605 kvm_arm_setup_debug(vcpu
);
607 /**************************************************************
610 trace_kvm_entry(*vcpu_pc(vcpu
));
612 vcpu
->mode
= IN_GUEST_MODE
;
614 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
616 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
620 *************************************************************/
622 kvm_arm_clear_debug(vcpu
);
625 * We may have taken a host interrupt in HYP mode (ie
626 * while executing the guest). This interrupt is still
627 * pending, as we haven't serviced it yet!
629 * We're now back in SVC mode, with interrupts
630 * disabled. Enabling the interrupts now will have
631 * the effect of taking the interrupt again, in SVC
637 * We do local_irq_enable() before calling kvm_guest_exit() so
638 * that if a timer interrupt hits while running the guest we
639 * account that tick as being spent in the guest. We enable
640 * preemption after calling kvm_guest_exit() so that if we get
641 * preempted we make sure ticks after that is not counted as
645 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
648 * We must sync the PMU and timer state before the vgic state so
649 * that the vgic can properly sample the updated state of the
652 kvm_pmu_sync_hwstate(vcpu
);
653 kvm_timer_sync_hwstate(vcpu
);
655 kvm_vgic_sync_hwstate(vcpu
);
659 ret
= handle_exit(vcpu
, run
, ret
);
662 if (vcpu
->sigset_active
)
663 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
667 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
673 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
674 bit_index
= __ffs(HCR_VI
);
675 else /* KVM_ARM_IRQ_CPU_FIQ */
676 bit_index
= __ffs(HCR_VF
);
678 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
680 set
= test_and_set_bit(bit_index
, ptr
);
682 set
= test_and_clear_bit(bit_index
, ptr
);
685 * If we didn't change anything, no need to wake up or kick other CPUs
691 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
692 * trigger a world-switch round on the running physical CPU to set the
693 * virtual IRQ/FIQ fields in the HCR appropriately.
700 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
703 u32 irq
= irq_level
->irq
;
704 unsigned int irq_type
, vcpu_idx
, irq_num
;
705 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
706 struct kvm_vcpu
*vcpu
= NULL
;
707 bool level
= irq_level
->level
;
709 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
710 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
711 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
713 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
716 case KVM_ARM_IRQ_TYPE_CPU
:
717 if (irqchip_in_kernel(kvm
))
720 if (vcpu_idx
>= nrcpus
)
723 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
727 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
730 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
731 case KVM_ARM_IRQ_TYPE_PPI
:
732 if (!irqchip_in_kernel(kvm
))
735 if (vcpu_idx
>= nrcpus
)
738 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
742 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
745 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
746 case KVM_ARM_IRQ_TYPE_SPI
:
747 if (!irqchip_in_kernel(kvm
))
750 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
753 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
759 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
760 const struct kvm_vcpu_init
*init
)
763 int phys_target
= kvm_target_cpu();
765 if (init
->target
!= phys_target
)
769 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
770 * use the same target.
772 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
775 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
776 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
777 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
779 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
783 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
784 * use the same feature set.
786 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
787 test_bit(i
, vcpu
->arch
.features
) != set
)
791 set_bit(i
, vcpu
->arch
.features
);
794 vcpu
->arch
.target
= phys_target
;
796 /* Now we know what it is, we can reset it. */
797 return kvm_reset_vcpu(vcpu
);
801 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
802 struct kvm_vcpu_init
*init
)
806 ret
= kvm_vcpu_set_target(vcpu
, init
);
811 * Ensure a rebooted VM will fault in RAM pages and detect if the
812 * guest MMU is turned off and flush the caches as needed.
814 if (vcpu
->arch
.has_run_once
)
815 stage2_unmap_vm(vcpu
->kvm
);
817 vcpu_reset_hcr(vcpu
);
820 * Handle the "start in power-off" case.
822 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
823 vcpu
->arch
.power_off
= true;
825 vcpu
->arch
.power_off
= false;
830 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
831 struct kvm_device_attr
*attr
)
835 switch (attr
->group
) {
837 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
844 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
845 struct kvm_device_attr
*attr
)
849 switch (attr
->group
) {
851 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
858 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
859 struct kvm_device_attr
*attr
)
863 switch (attr
->group
) {
865 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
872 long kvm_arch_vcpu_ioctl(struct file
*filp
,
873 unsigned int ioctl
, unsigned long arg
)
875 struct kvm_vcpu
*vcpu
= filp
->private_data
;
876 void __user
*argp
= (void __user
*)arg
;
877 struct kvm_device_attr attr
;
880 case KVM_ARM_VCPU_INIT
: {
881 struct kvm_vcpu_init init
;
883 if (copy_from_user(&init
, argp
, sizeof(init
)))
886 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
888 case KVM_SET_ONE_REG
:
889 case KVM_GET_ONE_REG
: {
890 struct kvm_one_reg reg
;
892 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
895 if (copy_from_user(®
, argp
, sizeof(reg
)))
897 if (ioctl
== KVM_SET_ONE_REG
)
898 return kvm_arm_set_reg(vcpu
, ®
);
900 return kvm_arm_get_reg(vcpu
, ®
);
902 case KVM_GET_REG_LIST
: {
903 struct kvm_reg_list __user
*user_list
= argp
;
904 struct kvm_reg_list reg_list
;
907 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
910 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
913 reg_list
.n
= kvm_arm_num_regs(vcpu
);
914 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
918 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
920 case KVM_SET_DEVICE_ATTR
: {
921 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
923 return kvm_arm_vcpu_set_attr(vcpu
, &attr
);
925 case KVM_GET_DEVICE_ATTR
: {
926 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
928 return kvm_arm_vcpu_get_attr(vcpu
, &attr
);
930 case KVM_HAS_DEVICE_ATTR
: {
931 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
933 return kvm_arm_vcpu_has_attr(vcpu
, &attr
);
941 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
943 * @log: slot id and address to which we copy the log
945 * Steps 1-4 below provide general overview of dirty page logging. See
946 * kvm_get_dirty_log_protect() function description for additional details.
948 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
949 * always flush the TLB (step 4) even if previous step failed and the dirty
950 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
951 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
952 * writes will be marked dirty for next log read.
954 * 1. Take a snapshot of the bit and clear it if needed.
955 * 2. Write protect the corresponding page.
956 * 3. Copy the snapshot to the userspace.
957 * 4. Flush TLB's if needed.
959 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
961 bool is_dirty
= false;
964 mutex_lock(&kvm
->slots_lock
);
966 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
969 kvm_flush_remote_tlbs(kvm
);
971 mutex_unlock(&kvm
->slots_lock
);
975 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
976 struct kvm_arm_device_addr
*dev_addr
)
978 unsigned long dev_id
, type
;
980 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
981 KVM_ARM_DEVICE_ID_SHIFT
;
982 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
983 KVM_ARM_DEVICE_TYPE_SHIFT
;
986 case KVM_ARM_DEVICE_VGIC_V2
:
989 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
995 long kvm_arch_vm_ioctl(struct file
*filp
,
996 unsigned int ioctl
, unsigned long arg
)
998 struct kvm
*kvm
= filp
->private_data
;
999 void __user
*argp
= (void __user
*)arg
;
1002 case KVM_CREATE_IRQCHIP
: {
1005 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1007 case KVM_ARM_SET_DEVICE_ADDR
: {
1008 struct kvm_arm_device_addr dev_addr
;
1010 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1012 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1014 case KVM_ARM_PREFERRED_TARGET
: {
1016 struct kvm_vcpu_init init
;
1018 err
= kvm_vcpu_preferred_target(&init
);
1022 if (copy_to_user(argp
, &init
, sizeof(init
)))
1032 static void cpu_init_hyp_mode(void *dummy
)
1034 phys_addr_t boot_pgd_ptr
;
1035 phys_addr_t pgd_ptr
;
1036 unsigned long hyp_stack_ptr
;
1037 unsigned long stack_page
;
1038 unsigned long vector_ptr
;
1040 /* Switch from the HYP stub to our own HYP init vector */
1041 __hyp_set_vectors(kvm_get_idmap_vector());
1043 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
1044 pgd_ptr
= kvm_mmu_get_httbr();
1045 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1046 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1047 vector_ptr
= (unsigned long)kvm_ksym_ref(__kvm_hyp_vector
);
1049 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1050 __cpu_init_stage2();
1052 kvm_arm_init_debug();
1055 static void cpu_hyp_reinit(void)
1057 if (is_kernel_in_hyp_mode()) {
1059 * __cpu_init_stage2() is safe to call even if the PM
1060 * event was cancelled before the CPU was reset.
1062 __cpu_init_stage2();
1064 if (__hyp_get_vectors() == hyp_default_vectors
)
1065 cpu_init_hyp_mode(NULL
);
1069 static void cpu_hyp_reset(void)
1071 phys_addr_t boot_pgd_ptr
;
1072 phys_addr_t phys_idmap_start
;
1074 if (!is_kernel_in_hyp_mode()) {
1075 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
1076 phys_idmap_start
= kvm_get_idmap_start();
1078 __cpu_reset_hyp_mode(boot_pgd_ptr
, phys_idmap_start
);
1082 static void _kvm_arch_hardware_enable(void *discard
)
1084 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1086 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1090 int kvm_arch_hardware_enable(void)
1092 _kvm_arch_hardware_enable(NULL
);
1096 static void _kvm_arch_hardware_disable(void *discard
)
1098 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1100 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1104 void kvm_arch_hardware_disable(void)
1106 _kvm_arch_hardware_disable(NULL
);
1109 #ifdef CONFIG_CPU_PM
1110 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1115 * kvm_arm_hardware_enabled is left with its old value over
1116 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1121 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1123 * don't update kvm_arm_hardware_enabled here
1124 * so that the hardware will be re-enabled
1125 * when we resume. See below.
1131 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1132 /* The hardware was enabled before suspend. */
1142 static struct notifier_block hyp_init_cpu_pm_nb
= {
1143 .notifier_call
= hyp_init_cpu_pm_notifier
,
1146 static void __init
hyp_cpu_pm_init(void)
1148 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1150 static void __init
hyp_cpu_pm_exit(void)
1152 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1155 static inline void hyp_cpu_pm_init(void)
1158 static inline void hyp_cpu_pm_exit(void)
1163 static void teardown_common_resources(void)
1165 free_percpu(kvm_host_cpu_state
);
1168 static int init_common_resources(void)
1170 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1171 if (!kvm_host_cpu_state
) {
1172 kvm_err("Cannot allocate host CPU state\n");
1179 static int init_subsystems(void)
1184 * Enable hardware so that subsystem initialisation can access EL2.
1186 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1189 * Register CPU lower-power notifier
1194 * Init HYP view of VGIC
1196 err
= kvm_vgic_hyp_init();
1199 vgic_present
= true;
1203 vgic_present
= false;
1211 * Init HYP architected timer support
1213 err
= kvm_timer_hyp_init();
1218 kvm_coproc_table_init();
1221 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1226 static void teardown_hyp_mode(void)
1230 if (is_kernel_in_hyp_mode())
1234 for_each_possible_cpu(cpu
)
1235 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1239 static int init_vhe_mode(void)
1241 /* set size of VMID supported by CPU */
1242 kvm_vmid_bits
= kvm_get_vmid_bits();
1243 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1245 kvm_info("VHE mode initialized successfully\n");
1250 * Inits Hyp-mode on all online CPUs
1252 static int init_hyp_mode(void)
1258 * Allocate Hyp PGD and setup Hyp identity mapping
1260 err
= kvm_mmu_init();
1265 * It is probably enough to obtain the default on one
1266 * CPU. It's unlikely to be different on the others.
1268 hyp_default_vectors
= __hyp_get_vectors();
1271 * Allocate stack pages for Hypervisor-mode
1273 for_each_possible_cpu(cpu
) {
1274 unsigned long stack_page
;
1276 stack_page
= __get_free_page(GFP_KERNEL
);
1282 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1286 * Map the Hyp-code called directly from the host
1288 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1289 kvm_ksym_ref(__hyp_text_end
));
1291 kvm_err("Cannot map world-switch code\n");
1295 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1296 kvm_ksym_ref(__end_rodata
));
1298 kvm_err("Cannot map rodata section\n");
1303 * Map the Hyp stack pages
1305 for_each_possible_cpu(cpu
) {
1306 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1307 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
1310 kvm_err("Cannot map hyp stack\n");
1315 for_each_possible_cpu(cpu
) {
1316 kvm_cpu_context_t
*cpu_ctxt
;
1318 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1319 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1322 kvm_err("Cannot map host CPU state: %d\n", err
);
1327 #ifndef CONFIG_HOTPLUG_CPU
1328 free_boot_hyp_pgd();
1331 /* set size of VMID supported by CPU */
1332 kvm_vmid_bits
= kvm_get_vmid_bits();
1333 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1335 kvm_info("Hyp mode initialized successfully\n");
1340 teardown_hyp_mode();
1341 kvm_err("error initializing Hyp mode: %d\n", err
);
1345 static void check_kvm_target_cpu(void *ret
)
1347 *(int *)ret
= kvm_target_cpu();
1350 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1352 struct kvm_vcpu
*vcpu
;
1355 mpidr
&= MPIDR_HWID_BITMASK
;
1356 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1357 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1364 * Initialize Hyp-mode and memory mappings on all CPUs.
1366 int kvm_arch_init(void *opaque
)
1371 if (!is_hyp_mode_available()) {
1372 kvm_err("HYP mode not available\n");
1376 for_each_online_cpu(cpu
) {
1377 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1379 kvm_err("Error, CPU %d not supported!\n", cpu
);
1384 err
= init_common_resources();
1388 if (is_kernel_in_hyp_mode())
1389 err
= init_vhe_mode();
1391 err
= init_hyp_mode();
1395 err
= init_subsystems();
1402 teardown_hyp_mode();
1404 teardown_common_resources();
1408 /* NOP: Compiling as a module not supported */
1409 void kvm_arch_exit(void)
1411 kvm_perf_teardown();
1414 static int arm_init(void)
1416 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1420 module_init(arm_init
);