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.h>
20 #include <linux/cpu_pm.h>
21 #include <linux/errno.h>
22 #include <linux/err.h>
23 #include <linux/kvm_host.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>
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>
49 __asm__(".arch_extension virt");
52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
53 static kvm_cpu_context_t __percpu
*kvm_host_cpu_state
;
54 static unsigned long hyp_default_vectors
;
56 /* Per-CPU variable containing the currently running vcpu. */
57 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
59 /* The VMID used in the VTTBR */
60 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
61 static u8 kvm_next_vmid
;
62 static DEFINE_SPINLOCK(kvm_vmid_lock
);
64 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
66 BUG_ON(preemptible());
67 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
71 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
72 * Must be called from non-preemptible context
74 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
76 BUG_ON(preemptible());
77 return __this_cpu_read(kvm_arm_running_vcpu
);
81 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
83 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
85 return &kvm_arm_running_vcpu
;
88 int kvm_arch_hardware_enable(void)
93 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
95 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
98 int kvm_arch_hardware_setup(void)
103 void kvm_arch_check_processor_compat(void *rtn
)
110 * kvm_arch_init_vm - initializes a VM data structure
111 * @kvm: pointer to the KVM struct
113 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
120 ret
= kvm_alloc_stage2_pgd(kvm
);
124 ret
= create_hyp_mappings(kvm
, kvm
+ 1);
126 goto out_free_stage2_pgd
;
130 /* Mark the initial VMID generation invalid */
131 kvm
->arch
.vmid_gen
= 0;
133 /* The maximum number of VCPUs is limited by the host's GIC model */
134 kvm
->arch
.max_vcpus
= kvm_vgic_get_max_vcpus();
138 kvm_free_stage2_pgd(kvm
);
143 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
145 return VM_FAULT_SIGBUS
;
150 * kvm_arch_destroy_vm - destroy the VM data structure
151 * @kvm: pointer to the KVM struct
153 void kvm_arch_destroy_vm(struct kvm
*kvm
)
157 kvm_free_stage2_pgd(kvm
);
159 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
161 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
162 kvm
->vcpus
[i
] = NULL
;
166 kvm_vgic_destroy(kvm
);
169 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
173 case KVM_CAP_IRQCHIP
:
174 case KVM_CAP_IOEVENTFD
:
175 case KVM_CAP_DEVICE_CTRL
:
176 case KVM_CAP_USER_MEMORY
:
177 case KVM_CAP_SYNC_MMU
:
178 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
179 case KVM_CAP_ONE_REG
:
180 case KVM_CAP_ARM_PSCI
:
181 case KVM_CAP_ARM_PSCI_0_2
:
182 case KVM_CAP_READONLY_MEM
:
183 case KVM_CAP_MP_STATE
:
186 case KVM_CAP_COALESCED_MMIO
:
187 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
189 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
192 case KVM_CAP_NR_VCPUS
:
193 r
= num_online_cpus();
195 case KVM_CAP_MAX_VCPUS
:
199 r
= kvm_arch_dev_ioctl_check_extension(ext
);
205 long kvm_arch_dev_ioctl(struct file
*filp
,
206 unsigned int ioctl
, unsigned long arg
)
212 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
215 struct kvm_vcpu
*vcpu
;
217 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
222 if (id
>= kvm
->arch
.max_vcpus
) {
227 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
233 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
237 err
= create_hyp_mappings(vcpu
, vcpu
+ 1);
243 kvm_vcpu_uninit(vcpu
);
245 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
250 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
254 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
256 kvm_mmu_free_memory_caches(vcpu
);
257 kvm_timer_vcpu_terminate(vcpu
);
258 kvm_vgic_vcpu_destroy(vcpu
);
259 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
262 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
264 kvm_arch_vcpu_free(vcpu
);
267 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
269 return kvm_timer_should_fire(vcpu
);
272 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
274 /* Force users to call KVM_ARM_VCPU_INIT */
275 vcpu
->arch
.target
= -1;
276 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
278 /* Set up the timer */
279 kvm_timer_vcpu_init(vcpu
);
281 kvm_arm_reset_debug_ptr(vcpu
);
286 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
289 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
291 kvm_arm_set_running_vcpu(vcpu
);
294 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
297 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
298 * if the vcpu is no longer assigned to a cpu. This is used for the
299 * optimized make_all_cpus_request path.
303 kvm_arm_set_running_vcpu(NULL
);
306 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
307 struct kvm_mp_state
*mp_state
)
309 if (vcpu
->arch
.pause
)
310 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
312 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
317 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
318 struct kvm_mp_state
*mp_state
)
320 switch (mp_state
->mp_state
) {
321 case KVM_MP_STATE_RUNNABLE
:
322 vcpu
->arch
.pause
= false;
324 case KVM_MP_STATE_STOPPED
:
325 vcpu
->arch
.pause
= true;
335 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
336 * @v: The VCPU pointer
338 * If the guest CPU is not waiting for interrupts or an interrupt line is
339 * asserted, the CPU is by definition runnable.
341 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
343 return !!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
);
346 /* Just ensure a guest exit from a particular CPU */
347 static void exit_vm_noop(void *info
)
351 void force_vm_exit(const cpumask_t
*mask
)
353 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
357 * need_new_vmid_gen - check that the VMID is still valid
358 * @kvm: The VM's VMID to checkt
360 * return true if there is a new generation of VMIDs being used
362 * The hardware supports only 256 values with the value zero reserved for the
363 * host, so we check if an assigned value belongs to a previous generation,
364 * which which requires us to assign a new value. If we're the first to use a
365 * VMID for the new generation, we must flush necessary caches and TLBs on all
368 static bool need_new_vmid_gen(struct kvm
*kvm
)
370 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
374 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
375 * @kvm The guest that we are about to run
377 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
378 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
381 static void update_vttbr(struct kvm
*kvm
)
383 phys_addr_t pgd_phys
;
386 if (!need_new_vmid_gen(kvm
))
389 spin_lock(&kvm_vmid_lock
);
392 * We need to re-check the vmid_gen here to ensure that if another vcpu
393 * already allocated a valid vmid for this vm, then this vcpu should
396 if (!need_new_vmid_gen(kvm
)) {
397 spin_unlock(&kvm_vmid_lock
);
401 /* First user of a new VMID generation? */
402 if (unlikely(kvm_next_vmid
== 0)) {
403 atomic64_inc(&kvm_vmid_gen
);
407 * On SMP we know no other CPUs can use this CPU's or each
408 * other's VMID after force_vm_exit returns since the
409 * kvm_vmid_lock blocks them from reentry to the guest.
411 force_vm_exit(cpu_all_mask
);
413 * Now broadcast TLB + ICACHE invalidation over the inner
414 * shareable domain to make sure all data structures are
417 kvm_call_hyp(__kvm_flush_vm_context
);
420 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
421 kvm
->arch
.vmid
= kvm_next_vmid
;
424 /* update vttbr to be used with the new vmid */
425 pgd_phys
= virt_to_phys(kvm_get_hwpgd(kvm
));
426 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
427 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK
;
428 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
430 spin_unlock(&kvm_vmid_lock
);
433 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
435 struct kvm
*kvm
= vcpu
->kvm
;
438 if (likely(vcpu
->arch
.has_run_once
))
441 vcpu
->arch
.has_run_once
= true;
444 * Map the VGIC hardware resources before running a vcpu the first
447 if (unlikely(!vgic_ready(kvm
))) {
448 ret
= kvm_vgic_map_resources(kvm
);
454 * Enable the arch timers only if we have an in-kernel VGIC
455 * and it has been properly initialized, since we cannot handle
456 * interrupts from the virtual timer with a userspace gic.
458 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
459 kvm_timer_enable(kvm
);
464 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
466 return vgic_initialized(kvm
);
469 static void vcpu_pause(struct kvm_vcpu
*vcpu
)
471 wait_queue_head_t
*wq
= kvm_arch_vcpu_wq(vcpu
);
473 wait_event_interruptible(*wq
, !vcpu
->arch
.pause
);
476 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
478 return vcpu
->arch
.target
>= 0;
482 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
483 * @vcpu: The VCPU pointer
484 * @run: The kvm_run structure pointer used for userspace state exchange
486 * This function is called through the VCPU_RUN ioctl called from user space. It
487 * will execute VM code in a loop until the time slice for the process is used
488 * or some emulation is needed from user space in which case the function will
489 * return with return value 0 and with the kvm_run structure filled in with the
490 * required data for the requested emulation.
492 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
497 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
500 ret
= kvm_vcpu_first_run_init(vcpu
);
504 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
505 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
510 if (vcpu
->sigset_active
)
511 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
514 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
517 * Check conditions before entering the guest
521 update_vttbr(vcpu
->kvm
);
523 if (vcpu
->arch
.pause
)
527 * Disarming the background timer must be done in a
528 * preemptible context, as this call may sleep.
530 kvm_timer_flush_hwstate(vcpu
);
533 * Preparing the interrupts to be injected also
534 * involves poking the GIC, which must be done in a
535 * non-preemptible context.
538 kvm_vgic_flush_hwstate(vcpu
);
543 * Re-check atomic conditions
545 if (signal_pending(current
)) {
547 run
->exit_reason
= KVM_EXIT_INTR
;
550 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
)) {
552 kvm_vgic_sync_hwstate(vcpu
);
554 kvm_timer_sync_hwstate(vcpu
);
558 kvm_arm_setup_debug(vcpu
);
560 /**************************************************************
563 trace_kvm_entry(*vcpu_pc(vcpu
));
565 vcpu
->mode
= IN_GUEST_MODE
;
567 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
569 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
572 *************************************************************/
574 kvm_arm_clear_debug(vcpu
);
577 * We may have taken a host interrupt in HYP mode (ie
578 * while executing the guest). This interrupt is still
579 * pending, as we haven't serviced it yet!
581 * We're now back in SVC mode, with interrupts
582 * disabled. Enabling the interrupts now will have
583 * the effect of taking the interrupt again, in SVC
589 * We do local_irq_enable() before calling kvm_guest_exit() so
590 * that if a timer interrupt hits while running the guest we
591 * account that tick as being spent in the guest. We enable
592 * preemption after calling kvm_guest_exit() so that if we get
593 * preempted we make sure ticks after that is not counted as
597 trace_kvm_exit(kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
599 kvm_vgic_sync_hwstate(vcpu
);
603 kvm_timer_sync_hwstate(vcpu
);
605 ret
= handle_exit(vcpu
, run
, ret
);
608 if (vcpu
->sigset_active
)
609 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
613 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
619 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
620 bit_index
= __ffs(HCR_VI
);
621 else /* KVM_ARM_IRQ_CPU_FIQ */
622 bit_index
= __ffs(HCR_VF
);
624 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
626 set
= test_and_set_bit(bit_index
, ptr
);
628 set
= test_and_clear_bit(bit_index
, ptr
);
631 * If we didn't change anything, no need to wake up or kick other CPUs
637 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
638 * trigger a world-switch round on the running physical CPU to set the
639 * virtual IRQ/FIQ fields in the HCR appropriately.
646 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
649 u32 irq
= irq_level
->irq
;
650 unsigned int irq_type
, vcpu_idx
, irq_num
;
651 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
652 struct kvm_vcpu
*vcpu
= NULL
;
653 bool level
= irq_level
->level
;
655 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
656 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
657 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
659 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
662 case KVM_ARM_IRQ_TYPE_CPU
:
663 if (irqchip_in_kernel(kvm
))
666 if (vcpu_idx
>= nrcpus
)
669 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
673 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
676 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
677 case KVM_ARM_IRQ_TYPE_PPI
:
678 if (!irqchip_in_kernel(kvm
))
681 if (vcpu_idx
>= nrcpus
)
684 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
688 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
691 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
692 case KVM_ARM_IRQ_TYPE_SPI
:
693 if (!irqchip_in_kernel(kvm
))
696 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
699 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
705 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
706 const struct kvm_vcpu_init
*init
)
709 int phys_target
= kvm_target_cpu();
711 if (init
->target
!= phys_target
)
715 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
716 * use the same target.
718 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
721 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
722 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
723 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
725 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
729 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
730 * use the same feature set.
732 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
733 test_bit(i
, vcpu
->arch
.features
) != set
)
737 set_bit(i
, vcpu
->arch
.features
);
740 vcpu
->arch
.target
= phys_target
;
742 /* Now we know what it is, we can reset it. */
743 return kvm_reset_vcpu(vcpu
);
747 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
748 struct kvm_vcpu_init
*init
)
752 ret
= kvm_vcpu_set_target(vcpu
, init
);
757 * Ensure a rebooted VM will fault in RAM pages and detect if the
758 * guest MMU is turned off and flush the caches as needed.
760 if (vcpu
->arch
.has_run_once
)
761 stage2_unmap_vm(vcpu
->kvm
);
763 vcpu_reset_hcr(vcpu
);
766 * Handle the "start in power-off" case by marking the VCPU as paused.
768 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
769 vcpu
->arch
.pause
= true;
771 vcpu
->arch
.pause
= false;
776 long kvm_arch_vcpu_ioctl(struct file
*filp
,
777 unsigned int ioctl
, unsigned long arg
)
779 struct kvm_vcpu
*vcpu
= filp
->private_data
;
780 void __user
*argp
= (void __user
*)arg
;
783 case KVM_ARM_VCPU_INIT
: {
784 struct kvm_vcpu_init init
;
786 if (copy_from_user(&init
, argp
, sizeof(init
)))
789 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
791 case KVM_SET_ONE_REG
:
792 case KVM_GET_ONE_REG
: {
793 struct kvm_one_reg reg
;
795 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
798 if (copy_from_user(®
, argp
, sizeof(reg
)))
800 if (ioctl
== KVM_SET_ONE_REG
)
801 return kvm_arm_set_reg(vcpu
, ®
);
803 return kvm_arm_get_reg(vcpu
, ®
);
805 case KVM_GET_REG_LIST
: {
806 struct kvm_reg_list __user
*user_list
= argp
;
807 struct kvm_reg_list reg_list
;
810 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
813 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
816 reg_list
.n
= kvm_arm_num_regs(vcpu
);
817 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
821 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
829 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
831 * @log: slot id and address to which we copy the log
833 * Steps 1-4 below provide general overview of dirty page logging. See
834 * kvm_get_dirty_log_protect() function description for additional details.
836 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
837 * always flush the TLB (step 4) even if previous step failed and the dirty
838 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
839 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
840 * writes will be marked dirty for next log read.
842 * 1. Take a snapshot of the bit and clear it if needed.
843 * 2. Write protect the corresponding page.
844 * 3. Copy the snapshot to the userspace.
845 * 4. Flush TLB's if needed.
847 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
849 bool is_dirty
= false;
852 mutex_lock(&kvm
->slots_lock
);
854 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
857 kvm_flush_remote_tlbs(kvm
);
859 mutex_unlock(&kvm
->slots_lock
);
863 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
864 struct kvm_arm_device_addr
*dev_addr
)
866 unsigned long dev_id
, type
;
868 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
869 KVM_ARM_DEVICE_ID_SHIFT
;
870 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
871 KVM_ARM_DEVICE_TYPE_SHIFT
;
874 case KVM_ARM_DEVICE_VGIC_V2
:
875 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
881 long kvm_arch_vm_ioctl(struct file
*filp
,
882 unsigned int ioctl
, unsigned long arg
)
884 struct kvm
*kvm
= filp
->private_data
;
885 void __user
*argp
= (void __user
*)arg
;
888 case KVM_CREATE_IRQCHIP
: {
889 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
891 case KVM_ARM_SET_DEVICE_ADDR
: {
892 struct kvm_arm_device_addr dev_addr
;
894 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
896 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
898 case KVM_ARM_PREFERRED_TARGET
: {
900 struct kvm_vcpu_init init
;
902 err
= kvm_vcpu_preferred_target(&init
);
906 if (copy_to_user(argp
, &init
, sizeof(init
)))
916 static void cpu_init_hyp_mode(void *dummy
)
918 phys_addr_t boot_pgd_ptr
;
920 unsigned long hyp_stack_ptr
;
921 unsigned long stack_page
;
922 unsigned long vector_ptr
;
924 /* Switch from the HYP stub to our own HYP init vector */
925 __hyp_set_vectors(kvm_get_idmap_vector());
927 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
928 pgd_ptr
= kvm_mmu_get_httbr();
929 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
930 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
931 vector_ptr
= (unsigned long)__kvm_hyp_vector
;
933 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
935 kvm_arm_init_debug();
938 static int hyp_init_cpu_notify(struct notifier_block
*self
,
939 unsigned long action
, void *cpu
)
943 case CPU_STARTING_FROZEN
:
944 if (__hyp_get_vectors() == hyp_default_vectors
)
945 cpu_init_hyp_mode(NULL
);
952 static struct notifier_block hyp_init_cpu_nb
= {
953 .notifier_call
= hyp_init_cpu_notify
,
957 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
961 if (cmd
== CPU_PM_EXIT
&&
962 __hyp_get_vectors() == hyp_default_vectors
) {
963 cpu_init_hyp_mode(NULL
);
970 static struct notifier_block hyp_init_cpu_pm_nb
= {
971 .notifier_call
= hyp_init_cpu_pm_notifier
,
974 static void __init
hyp_cpu_pm_init(void)
976 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
979 static inline void hyp_cpu_pm_init(void)
985 * Inits Hyp-mode on all online CPUs
987 static int init_hyp_mode(void)
993 * Allocate Hyp PGD and setup Hyp identity mapping
995 err
= kvm_mmu_init();
1000 * It is probably enough to obtain the default on one
1001 * CPU. It's unlikely to be different on the others.
1003 hyp_default_vectors
= __hyp_get_vectors();
1006 * Allocate stack pages for Hypervisor-mode
1008 for_each_possible_cpu(cpu
) {
1009 unsigned long stack_page
;
1011 stack_page
= __get_free_page(GFP_KERNEL
);
1014 goto out_free_stack_pages
;
1017 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1021 * Map the Hyp-code called directly from the host
1023 err
= create_hyp_mappings(__kvm_hyp_code_start
, __kvm_hyp_code_end
);
1025 kvm_err("Cannot map world-switch code\n");
1026 goto out_free_mappings
;
1030 * Map the Hyp stack pages
1032 for_each_possible_cpu(cpu
) {
1033 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1034 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
1037 kvm_err("Cannot map hyp stack\n");
1038 goto out_free_mappings
;
1043 * Map the host CPU structures
1045 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1046 if (!kvm_host_cpu_state
) {
1048 kvm_err("Cannot allocate host CPU state\n");
1049 goto out_free_mappings
;
1052 for_each_possible_cpu(cpu
) {
1053 kvm_cpu_context_t
*cpu_ctxt
;
1055 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1056 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1059 kvm_err("Cannot map host CPU state: %d\n", err
);
1060 goto out_free_context
;
1065 * Execute the init code on each CPU.
1067 on_each_cpu(cpu_init_hyp_mode
, NULL
, 1);
1070 * Init HYP view of VGIC
1072 err
= kvm_vgic_hyp_init();
1074 goto out_free_context
;
1077 * Init HYP architected timer support
1079 err
= kvm_timer_hyp_init();
1081 goto out_free_mappings
;
1083 #ifndef CONFIG_HOTPLUG_CPU
1084 free_boot_hyp_pgd();
1089 kvm_info("Hyp mode initialized successfully\n");
1093 free_percpu(kvm_host_cpu_state
);
1096 out_free_stack_pages
:
1097 for_each_possible_cpu(cpu
)
1098 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1100 kvm_err("error initializing Hyp mode: %d\n", err
);
1104 static void check_kvm_target_cpu(void *ret
)
1106 *(int *)ret
= kvm_target_cpu();
1109 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1111 struct kvm_vcpu
*vcpu
;
1114 mpidr
&= MPIDR_HWID_BITMASK
;
1115 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1116 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1123 * Initialize Hyp-mode and memory mappings on all CPUs.
1125 int kvm_arch_init(void *opaque
)
1130 if (!is_hyp_mode_available()) {
1131 kvm_err("HYP mode not available\n");
1135 for_each_online_cpu(cpu
) {
1136 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1138 kvm_err("Error, CPU %d not supported!\n", cpu
);
1143 cpu_notifier_register_begin();
1145 err
= init_hyp_mode();
1149 err
= __register_cpu_notifier(&hyp_init_cpu_nb
);
1151 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err
);
1155 cpu_notifier_register_done();
1159 kvm_coproc_table_init();
1162 cpu_notifier_register_done();
1166 /* NOP: Compiling as a module not supported */
1167 void kvm_arch_exit(void)
1169 kvm_perf_teardown();
1172 static int arm_init(void)
1174 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1178 module_init(arm_init
);