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
:
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
:
185 case KVM_CAP_COALESCED_MMIO
:
186 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
188 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
191 case KVM_CAP_NR_VCPUS
:
192 r
= num_online_cpus();
194 case KVM_CAP_MAX_VCPUS
:
198 r
= kvm_arch_dev_ioctl_check_extension(ext
);
204 long kvm_arch_dev_ioctl(struct file
*filp
,
205 unsigned int ioctl
, unsigned long arg
)
211 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
214 struct kvm_vcpu
*vcpu
;
216 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
221 if (id
>= kvm
->arch
.max_vcpus
) {
226 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
232 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
236 err
= create_hyp_mappings(vcpu
, vcpu
+ 1);
242 kvm_vcpu_uninit(vcpu
);
244 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
249 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
253 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
255 kvm_mmu_free_memory_caches(vcpu
);
256 kvm_timer_vcpu_terminate(vcpu
);
257 kvm_vgic_vcpu_destroy(vcpu
);
258 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
261 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
263 kvm_arch_vcpu_free(vcpu
);
266 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
271 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
273 /* Force users to call KVM_ARM_VCPU_INIT */
274 vcpu
->arch
.target
= -1;
275 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
277 /* Set up the timer */
278 kvm_timer_vcpu_init(vcpu
);
283 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
286 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
288 kvm_arm_set_running_vcpu(vcpu
);
291 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
294 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
295 * if the vcpu is no longer assigned to a cpu. This is used for the
296 * optimized make_all_cpus_request path.
300 kvm_arm_set_running_vcpu(NULL
);
303 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu
*vcpu
,
304 struct kvm_guest_debug
*dbg
)
310 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
311 struct kvm_mp_state
*mp_state
)
316 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
317 struct kvm_mp_state
*mp_state
)
323 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
324 * @v: The VCPU pointer
326 * If the guest CPU is not waiting for interrupts or an interrupt line is
327 * asserted, the CPU is by definition runnable.
329 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
331 return !!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
);
334 /* Just ensure a guest exit from a particular CPU */
335 static void exit_vm_noop(void *info
)
339 void force_vm_exit(const cpumask_t
*mask
)
341 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
345 * need_new_vmid_gen - check that the VMID is still valid
346 * @kvm: The VM's VMID to checkt
348 * return true if there is a new generation of VMIDs being used
350 * The hardware supports only 256 values with the value zero reserved for the
351 * host, so we check if an assigned value belongs to a previous generation,
352 * which which requires us to assign a new value. If we're the first to use a
353 * VMID for the new generation, we must flush necessary caches and TLBs on all
356 static bool need_new_vmid_gen(struct kvm
*kvm
)
358 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
362 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
363 * @kvm The guest that we are about to run
365 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
366 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
369 static void update_vttbr(struct kvm
*kvm
)
371 phys_addr_t pgd_phys
;
374 if (!need_new_vmid_gen(kvm
))
377 spin_lock(&kvm_vmid_lock
);
380 * We need to re-check the vmid_gen here to ensure that if another vcpu
381 * already allocated a valid vmid for this vm, then this vcpu should
384 if (!need_new_vmid_gen(kvm
)) {
385 spin_unlock(&kvm_vmid_lock
);
389 /* First user of a new VMID generation? */
390 if (unlikely(kvm_next_vmid
== 0)) {
391 atomic64_inc(&kvm_vmid_gen
);
395 * On SMP we know no other CPUs can use this CPU's or each
396 * other's VMID after force_vm_exit returns since the
397 * kvm_vmid_lock blocks them from reentry to the guest.
399 force_vm_exit(cpu_all_mask
);
401 * Now broadcast TLB + ICACHE invalidation over the inner
402 * shareable domain to make sure all data structures are
405 kvm_call_hyp(__kvm_flush_vm_context
);
408 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
409 kvm
->arch
.vmid
= kvm_next_vmid
;
412 /* update vttbr to be used with the new vmid */
413 pgd_phys
= virt_to_phys(kvm_get_hwpgd(kvm
));
414 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
415 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK
;
416 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
418 spin_unlock(&kvm_vmid_lock
);
421 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
423 struct kvm
*kvm
= vcpu
->kvm
;
426 if (likely(vcpu
->arch
.has_run_once
))
429 vcpu
->arch
.has_run_once
= true;
432 * Map the VGIC hardware resources before running a vcpu the first
435 if (unlikely(!vgic_ready(kvm
))) {
436 ret
= kvm_vgic_map_resources(kvm
);
442 * Enable the arch timers only if we have an in-kernel VGIC
443 * and it has been properly initialized, since we cannot handle
444 * interrupts from the virtual timer with a userspace gic.
446 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
447 kvm_timer_enable(kvm
);
452 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
454 return vgic_initialized(kvm
);
457 static void vcpu_pause(struct kvm_vcpu
*vcpu
)
459 wait_queue_head_t
*wq
= kvm_arch_vcpu_wq(vcpu
);
461 wait_event_interruptible(*wq
, !vcpu
->arch
.pause
);
464 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
466 return vcpu
->arch
.target
>= 0;
470 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
471 * @vcpu: The VCPU pointer
472 * @run: The kvm_run structure pointer used for userspace state exchange
474 * This function is called through the VCPU_RUN ioctl called from user space. It
475 * will execute VM code in a loop until the time slice for the process is used
476 * or some emulation is needed from user space in which case the function will
477 * return with return value 0 and with the kvm_run structure filled in with the
478 * required data for the requested emulation.
480 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
485 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
488 ret
= kvm_vcpu_first_run_init(vcpu
);
492 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
493 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
498 if (vcpu
->sigset_active
)
499 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
502 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
505 * Check conditions before entering the guest
509 update_vttbr(vcpu
->kvm
);
511 if (vcpu
->arch
.pause
)
514 kvm_vgic_flush_hwstate(vcpu
);
515 kvm_timer_flush_hwstate(vcpu
);
520 * Re-check atomic conditions
522 if (signal_pending(current
)) {
524 run
->exit_reason
= KVM_EXIT_INTR
;
527 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
)) {
529 kvm_timer_sync_hwstate(vcpu
);
530 kvm_vgic_sync_hwstate(vcpu
);
534 /**************************************************************
537 trace_kvm_entry(*vcpu_pc(vcpu
));
539 vcpu
->mode
= IN_GUEST_MODE
;
541 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
543 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
545 trace_kvm_exit(kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
547 * We may have taken a host interrupt in HYP mode (ie
548 * while executing the guest). This interrupt is still
549 * pending, as we haven't serviced it yet!
551 * We're now back in SVC mode, with interrupts
552 * disabled. Enabling the interrupts now will have
553 * the effect of taking the interrupt again, in SVC
560 *************************************************************/
562 kvm_timer_sync_hwstate(vcpu
);
563 kvm_vgic_sync_hwstate(vcpu
);
565 ret
= handle_exit(vcpu
, run
, ret
);
568 if (vcpu
->sigset_active
)
569 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
573 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
579 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
580 bit_index
= __ffs(HCR_VI
);
581 else /* KVM_ARM_IRQ_CPU_FIQ */
582 bit_index
= __ffs(HCR_VF
);
584 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
586 set
= test_and_set_bit(bit_index
, ptr
);
588 set
= test_and_clear_bit(bit_index
, ptr
);
591 * If we didn't change anything, no need to wake up or kick other CPUs
597 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
598 * trigger a world-switch round on the running physical CPU to set the
599 * virtual IRQ/FIQ fields in the HCR appropriately.
606 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
609 u32 irq
= irq_level
->irq
;
610 unsigned int irq_type
, vcpu_idx
, irq_num
;
611 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
612 struct kvm_vcpu
*vcpu
= NULL
;
613 bool level
= irq_level
->level
;
615 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
616 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
617 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
619 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
622 case KVM_ARM_IRQ_TYPE_CPU
:
623 if (irqchip_in_kernel(kvm
))
626 if (vcpu_idx
>= nrcpus
)
629 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
633 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
636 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
637 case KVM_ARM_IRQ_TYPE_PPI
:
638 if (!irqchip_in_kernel(kvm
))
641 if (vcpu_idx
>= nrcpus
)
644 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
648 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
651 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
652 case KVM_ARM_IRQ_TYPE_SPI
:
653 if (!irqchip_in_kernel(kvm
))
656 if (irq_num
< VGIC_NR_PRIVATE_IRQS
||
657 irq_num
> KVM_ARM_IRQ_GIC_MAX
)
660 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
666 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
667 const struct kvm_vcpu_init
*init
)
670 int phys_target
= kvm_target_cpu();
672 if (init
->target
!= phys_target
)
676 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
677 * use the same target.
679 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
682 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
683 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
684 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
686 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
690 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
691 * use the same feature set.
693 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
694 test_bit(i
, vcpu
->arch
.features
) != set
)
698 set_bit(i
, vcpu
->arch
.features
);
701 vcpu
->arch
.target
= phys_target
;
703 /* Now we know what it is, we can reset it. */
704 return kvm_reset_vcpu(vcpu
);
708 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
709 struct kvm_vcpu_init
*init
)
713 ret
= kvm_vcpu_set_target(vcpu
, init
);
718 * Ensure a rebooted VM will fault in RAM pages and detect if the
719 * guest MMU is turned off and flush the caches as needed.
721 if (vcpu
->arch
.has_run_once
)
722 stage2_unmap_vm(vcpu
->kvm
);
724 vcpu_reset_hcr(vcpu
);
727 * Handle the "start in power-off" case by marking the VCPU as paused.
729 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
730 vcpu
->arch
.pause
= true;
732 vcpu
->arch
.pause
= false;
737 long kvm_arch_vcpu_ioctl(struct file
*filp
,
738 unsigned int ioctl
, unsigned long arg
)
740 struct kvm_vcpu
*vcpu
= filp
->private_data
;
741 void __user
*argp
= (void __user
*)arg
;
744 case KVM_ARM_VCPU_INIT
: {
745 struct kvm_vcpu_init init
;
747 if (copy_from_user(&init
, argp
, sizeof(init
)))
750 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
752 case KVM_SET_ONE_REG
:
753 case KVM_GET_ONE_REG
: {
754 struct kvm_one_reg reg
;
756 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
759 if (copy_from_user(®
, argp
, sizeof(reg
)))
761 if (ioctl
== KVM_SET_ONE_REG
)
762 return kvm_arm_set_reg(vcpu
, ®
);
764 return kvm_arm_get_reg(vcpu
, ®
);
766 case KVM_GET_REG_LIST
: {
767 struct kvm_reg_list __user
*user_list
= argp
;
768 struct kvm_reg_list reg_list
;
771 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
774 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
777 reg_list
.n
= kvm_arm_num_regs(vcpu
);
778 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
782 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
790 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
792 * @log: slot id and address to which we copy the log
794 * Steps 1-4 below provide general overview of dirty page logging. See
795 * kvm_get_dirty_log_protect() function description for additional details.
797 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
798 * always flush the TLB (step 4) even if previous step failed and the dirty
799 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
800 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
801 * writes will be marked dirty for next log read.
803 * 1. Take a snapshot of the bit and clear it if needed.
804 * 2. Write protect the corresponding page.
805 * 3. Copy the snapshot to the userspace.
806 * 4. Flush TLB's if needed.
808 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
810 bool is_dirty
= false;
813 mutex_lock(&kvm
->slots_lock
);
815 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
818 kvm_flush_remote_tlbs(kvm
);
820 mutex_unlock(&kvm
->slots_lock
);
824 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
825 struct kvm_arm_device_addr
*dev_addr
)
827 unsigned long dev_id
, type
;
829 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
830 KVM_ARM_DEVICE_ID_SHIFT
;
831 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
832 KVM_ARM_DEVICE_TYPE_SHIFT
;
835 case KVM_ARM_DEVICE_VGIC_V2
:
836 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
842 long kvm_arch_vm_ioctl(struct file
*filp
,
843 unsigned int ioctl
, unsigned long arg
)
845 struct kvm
*kvm
= filp
->private_data
;
846 void __user
*argp
= (void __user
*)arg
;
849 case KVM_CREATE_IRQCHIP
: {
850 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
852 case KVM_ARM_SET_DEVICE_ADDR
: {
853 struct kvm_arm_device_addr dev_addr
;
855 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
857 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
859 case KVM_ARM_PREFERRED_TARGET
: {
861 struct kvm_vcpu_init init
;
863 err
= kvm_vcpu_preferred_target(&init
);
867 if (copy_to_user(argp
, &init
, sizeof(init
)))
877 static void cpu_init_hyp_mode(void *dummy
)
879 phys_addr_t boot_pgd_ptr
;
881 unsigned long hyp_stack_ptr
;
882 unsigned long stack_page
;
883 unsigned long vector_ptr
;
885 /* Switch from the HYP stub to our own HYP init vector */
886 __hyp_set_vectors(kvm_get_idmap_vector());
888 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
889 pgd_ptr
= kvm_mmu_get_httbr();
890 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
891 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
892 vector_ptr
= (unsigned long)__kvm_hyp_vector
;
894 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
897 static int hyp_init_cpu_notify(struct notifier_block
*self
,
898 unsigned long action
, void *cpu
)
902 case CPU_STARTING_FROZEN
:
903 if (__hyp_get_vectors() == hyp_default_vectors
)
904 cpu_init_hyp_mode(NULL
);
911 static struct notifier_block hyp_init_cpu_nb
= {
912 .notifier_call
= hyp_init_cpu_notify
,
916 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
920 if (cmd
== CPU_PM_EXIT
&&
921 __hyp_get_vectors() == hyp_default_vectors
) {
922 cpu_init_hyp_mode(NULL
);
929 static struct notifier_block hyp_init_cpu_pm_nb
= {
930 .notifier_call
= hyp_init_cpu_pm_notifier
,
933 static void __init
hyp_cpu_pm_init(void)
935 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
938 static inline void hyp_cpu_pm_init(void)
944 * Inits Hyp-mode on all online CPUs
946 static int init_hyp_mode(void)
952 * Allocate Hyp PGD and setup Hyp identity mapping
954 err
= kvm_mmu_init();
959 * It is probably enough to obtain the default on one
960 * CPU. It's unlikely to be different on the others.
962 hyp_default_vectors
= __hyp_get_vectors();
965 * Allocate stack pages for Hypervisor-mode
967 for_each_possible_cpu(cpu
) {
968 unsigned long stack_page
;
970 stack_page
= __get_free_page(GFP_KERNEL
);
973 goto out_free_stack_pages
;
976 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
980 * Map the Hyp-code called directly from the host
982 err
= create_hyp_mappings(__kvm_hyp_code_start
, __kvm_hyp_code_end
);
984 kvm_err("Cannot map world-switch code\n");
985 goto out_free_mappings
;
989 * Map the Hyp stack pages
991 for_each_possible_cpu(cpu
) {
992 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
993 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
996 kvm_err("Cannot map hyp stack\n");
997 goto out_free_mappings
;
1002 * Map the host CPU structures
1004 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1005 if (!kvm_host_cpu_state
) {
1007 kvm_err("Cannot allocate host CPU state\n");
1008 goto out_free_mappings
;
1011 for_each_possible_cpu(cpu
) {
1012 kvm_cpu_context_t
*cpu_ctxt
;
1014 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1015 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1018 kvm_err("Cannot map host CPU state: %d\n", err
);
1019 goto out_free_context
;
1024 * Execute the init code on each CPU.
1026 on_each_cpu(cpu_init_hyp_mode
, NULL
, 1);
1029 * Init HYP view of VGIC
1031 err
= kvm_vgic_hyp_init();
1033 goto out_free_context
;
1036 * Init HYP architected timer support
1038 err
= kvm_timer_hyp_init();
1040 goto out_free_mappings
;
1042 #ifndef CONFIG_HOTPLUG_CPU
1043 free_boot_hyp_pgd();
1048 kvm_info("Hyp mode initialized successfully\n");
1052 free_percpu(kvm_host_cpu_state
);
1055 out_free_stack_pages
:
1056 for_each_possible_cpu(cpu
)
1057 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1059 kvm_err("error initializing Hyp mode: %d\n", err
);
1063 static void check_kvm_target_cpu(void *ret
)
1065 *(int *)ret
= kvm_target_cpu();
1068 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1070 struct kvm_vcpu
*vcpu
;
1073 mpidr
&= MPIDR_HWID_BITMASK
;
1074 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1075 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1082 * Initialize Hyp-mode and memory mappings on all CPUs.
1084 int kvm_arch_init(void *opaque
)
1089 if (!is_hyp_mode_available()) {
1090 kvm_err("HYP mode not available\n");
1094 for_each_online_cpu(cpu
) {
1095 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1097 kvm_err("Error, CPU %d not supported!\n", cpu
);
1102 cpu_notifier_register_begin();
1104 err
= init_hyp_mode();
1108 err
= __register_cpu_notifier(&hyp_init_cpu_nb
);
1110 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err
);
1114 cpu_notifier_register_done();
1118 kvm_coproc_table_init();
1121 cpu_notifier_register_done();
1125 /* NOP: Compiling as a module not supported */
1126 void kvm_arch_exit(void)
1128 kvm_perf_teardown();
1131 static int arm_init(void)
1133 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1137 module_init(arm_init
);