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
);
284 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
287 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(kvm_host_cpu_state
);
289 kvm_arm_set_running_vcpu(vcpu
);
292 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
295 * The arch-generic KVM code expects the cpu field of a vcpu to be -1
296 * if the vcpu is no longer assigned to a cpu. This is used for the
297 * optimized make_all_cpus_request path.
301 kvm_arm_set_running_vcpu(NULL
);
304 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
305 struct kvm_mp_state
*mp_state
)
307 if (vcpu
->arch
.pause
)
308 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
310 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
315 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
316 struct kvm_mp_state
*mp_state
)
318 switch (mp_state
->mp_state
) {
319 case KVM_MP_STATE_RUNNABLE
:
320 vcpu
->arch
.pause
= false;
322 case KVM_MP_STATE_STOPPED
:
323 vcpu
->arch
.pause
= true;
333 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
334 * @v: The VCPU pointer
336 * If the guest CPU is not waiting for interrupts or an interrupt line is
337 * asserted, the CPU is by definition runnable.
339 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
341 return !!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
);
344 /* Just ensure a guest exit from a particular CPU */
345 static void exit_vm_noop(void *info
)
349 void force_vm_exit(const cpumask_t
*mask
)
351 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
355 * need_new_vmid_gen - check that the VMID is still valid
356 * @kvm: The VM's VMID to checkt
358 * return true if there is a new generation of VMIDs being used
360 * The hardware supports only 256 values with the value zero reserved for the
361 * host, so we check if an assigned value belongs to a previous generation,
362 * which which requires us to assign a new value. If we're the first to use a
363 * VMID for the new generation, we must flush necessary caches and TLBs on all
366 static bool need_new_vmid_gen(struct kvm
*kvm
)
368 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
372 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
373 * @kvm The guest that we are about to run
375 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
376 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
379 static void update_vttbr(struct kvm
*kvm
)
381 phys_addr_t pgd_phys
;
384 if (!need_new_vmid_gen(kvm
))
387 spin_lock(&kvm_vmid_lock
);
390 * We need to re-check the vmid_gen here to ensure that if another vcpu
391 * already allocated a valid vmid for this vm, then this vcpu should
394 if (!need_new_vmid_gen(kvm
)) {
395 spin_unlock(&kvm_vmid_lock
);
399 /* First user of a new VMID generation? */
400 if (unlikely(kvm_next_vmid
== 0)) {
401 atomic64_inc(&kvm_vmid_gen
);
405 * On SMP we know no other CPUs can use this CPU's or each
406 * other's VMID after force_vm_exit returns since the
407 * kvm_vmid_lock blocks them from reentry to the guest.
409 force_vm_exit(cpu_all_mask
);
411 * Now broadcast TLB + ICACHE invalidation over the inner
412 * shareable domain to make sure all data structures are
415 kvm_call_hyp(__kvm_flush_vm_context
);
418 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
419 kvm
->arch
.vmid
= kvm_next_vmid
;
422 /* update vttbr to be used with the new vmid */
423 pgd_phys
= virt_to_phys(kvm_get_hwpgd(kvm
));
424 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
425 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK
;
426 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
428 spin_unlock(&kvm_vmid_lock
);
431 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
433 struct kvm
*kvm
= vcpu
->kvm
;
436 if (likely(vcpu
->arch
.has_run_once
))
439 vcpu
->arch
.has_run_once
= true;
442 * Map the VGIC hardware resources before running a vcpu the first
445 if (unlikely(!vgic_ready(kvm
))) {
446 ret
= kvm_vgic_map_resources(kvm
);
452 * Enable the arch timers only if we have an in-kernel VGIC
453 * and it has been properly initialized, since we cannot handle
454 * interrupts from the virtual timer with a userspace gic.
456 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
457 kvm_timer_enable(kvm
);
462 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
464 return vgic_initialized(kvm
);
467 static void vcpu_pause(struct kvm_vcpu
*vcpu
)
469 wait_queue_head_t
*wq
= kvm_arch_vcpu_wq(vcpu
);
471 wait_event_interruptible(*wq
, !vcpu
->arch
.pause
);
474 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
476 return vcpu
->arch
.target
>= 0;
480 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
481 * @vcpu: The VCPU pointer
482 * @run: The kvm_run structure pointer used for userspace state exchange
484 * This function is called through the VCPU_RUN ioctl called from user space. It
485 * will execute VM code in a loop until the time slice for the process is used
486 * or some emulation is needed from user space in which case the function will
487 * return with return value 0 and with the kvm_run structure filled in with the
488 * required data for the requested emulation.
490 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
495 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
498 ret
= kvm_vcpu_first_run_init(vcpu
);
502 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
503 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
508 if (vcpu
->sigset_active
)
509 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
512 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
515 * Check conditions before entering the guest
519 update_vttbr(vcpu
->kvm
);
521 if (vcpu
->arch
.pause
)
524 kvm_vgic_flush_hwstate(vcpu
);
525 kvm_timer_flush_hwstate(vcpu
);
531 * Re-check atomic conditions
533 if (signal_pending(current
)) {
535 run
->exit_reason
= KVM_EXIT_INTR
;
538 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
)) {
541 kvm_timer_sync_hwstate(vcpu
);
542 kvm_vgic_sync_hwstate(vcpu
);
546 /**************************************************************
549 trace_kvm_entry(*vcpu_pc(vcpu
));
551 vcpu
->mode
= IN_GUEST_MODE
;
553 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
555 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
558 *************************************************************/
561 * We may have taken a host interrupt in HYP mode (ie
562 * while executing the guest). This interrupt is still
563 * pending, as we haven't serviced it yet!
565 * We're now back in SVC mode, with interrupts
566 * disabled. Enabling the interrupts now will have
567 * the effect of taking the interrupt again, in SVC
573 * We do local_irq_enable() before calling kvm_guest_exit() so
574 * that if a timer interrupt hits while running the guest we
575 * account that tick as being spent in the guest. We enable
576 * preemption after calling kvm_guest_exit() so that if we get
577 * preempted we make sure ticks after that is not counted as
581 trace_kvm_exit(kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
585 kvm_timer_sync_hwstate(vcpu
);
586 kvm_vgic_sync_hwstate(vcpu
);
588 ret
= handle_exit(vcpu
, run
, ret
);
591 if (vcpu
->sigset_active
)
592 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
596 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
602 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
603 bit_index
= __ffs(HCR_VI
);
604 else /* KVM_ARM_IRQ_CPU_FIQ */
605 bit_index
= __ffs(HCR_VF
);
607 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
609 set
= test_and_set_bit(bit_index
, ptr
);
611 set
= test_and_clear_bit(bit_index
, ptr
);
614 * If we didn't change anything, no need to wake up or kick other CPUs
620 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
621 * trigger a world-switch round on the running physical CPU to set the
622 * virtual IRQ/FIQ fields in the HCR appropriately.
629 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
632 u32 irq
= irq_level
->irq
;
633 unsigned int irq_type
, vcpu_idx
, irq_num
;
634 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
635 struct kvm_vcpu
*vcpu
= NULL
;
636 bool level
= irq_level
->level
;
638 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
639 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
640 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
642 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
645 case KVM_ARM_IRQ_TYPE_CPU
:
646 if (irqchip_in_kernel(kvm
))
649 if (vcpu_idx
>= nrcpus
)
652 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
656 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
659 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
660 case KVM_ARM_IRQ_TYPE_PPI
:
661 if (!irqchip_in_kernel(kvm
))
664 if (vcpu_idx
>= nrcpus
)
667 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
671 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
674 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
675 case KVM_ARM_IRQ_TYPE_SPI
:
676 if (!irqchip_in_kernel(kvm
))
679 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
682 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
688 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
689 const struct kvm_vcpu_init
*init
)
692 int phys_target
= kvm_target_cpu();
694 if (init
->target
!= phys_target
)
698 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
699 * use the same target.
701 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
704 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
705 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
706 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
708 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
712 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
713 * use the same feature set.
715 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
716 test_bit(i
, vcpu
->arch
.features
) != set
)
720 set_bit(i
, vcpu
->arch
.features
);
723 vcpu
->arch
.target
= phys_target
;
725 /* Now we know what it is, we can reset it. */
726 return kvm_reset_vcpu(vcpu
);
730 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
731 struct kvm_vcpu_init
*init
)
735 ret
= kvm_vcpu_set_target(vcpu
, init
);
740 * Ensure a rebooted VM will fault in RAM pages and detect if the
741 * guest MMU is turned off and flush the caches as needed.
743 if (vcpu
->arch
.has_run_once
)
744 stage2_unmap_vm(vcpu
->kvm
);
746 vcpu_reset_hcr(vcpu
);
749 * Handle the "start in power-off" case by marking the VCPU as paused.
751 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
752 vcpu
->arch
.pause
= true;
754 vcpu
->arch
.pause
= false;
759 long kvm_arch_vcpu_ioctl(struct file
*filp
,
760 unsigned int ioctl
, unsigned long arg
)
762 struct kvm_vcpu
*vcpu
= filp
->private_data
;
763 void __user
*argp
= (void __user
*)arg
;
766 case KVM_ARM_VCPU_INIT
: {
767 struct kvm_vcpu_init init
;
769 if (copy_from_user(&init
, argp
, sizeof(init
)))
772 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
774 case KVM_SET_ONE_REG
:
775 case KVM_GET_ONE_REG
: {
776 struct kvm_one_reg reg
;
778 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
781 if (copy_from_user(®
, argp
, sizeof(reg
)))
783 if (ioctl
== KVM_SET_ONE_REG
)
784 return kvm_arm_set_reg(vcpu
, ®
);
786 return kvm_arm_get_reg(vcpu
, ®
);
788 case KVM_GET_REG_LIST
: {
789 struct kvm_reg_list __user
*user_list
= argp
;
790 struct kvm_reg_list reg_list
;
793 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
796 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
799 reg_list
.n
= kvm_arm_num_regs(vcpu
);
800 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
804 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
812 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
814 * @log: slot id and address to which we copy the log
816 * Steps 1-4 below provide general overview of dirty page logging. See
817 * kvm_get_dirty_log_protect() function description for additional details.
819 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
820 * always flush the TLB (step 4) even if previous step failed and the dirty
821 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
822 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
823 * writes will be marked dirty for next log read.
825 * 1. Take a snapshot of the bit and clear it if needed.
826 * 2. Write protect the corresponding page.
827 * 3. Copy the snapshot to the userspace.
828 * 4. Flush TLB's if needed.
830 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
832 bool is_dirty
= false;
835 mutex_lock(&kvm
->slots_lock
);
837 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
840 kvm_flush_remote_tlbs(kvm
);
842 mutex_unlock(&kvm
->slots_lock
);
846 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
847 struct kvm_arm_device_addr
*dev_addr
)
849 unsigned long dev_id
, type
;
851 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
852 KVM_ARM_DEVICE_ID_SHIFT
;
853 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
854 KVM_ARM_DEVICE_TYPE_SHIFT
;
857 case KVM_ARM_DEVICE_VGIC_V2
:
858 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
864 long kvm_arch_vm_ioctl(struct file
*filp
,
865 unsigned int ioctl
, unsigned long arg
)
867 struct kvm
*kvm
= filp
->private_data
;
868 void __user
*argp
= (void __user
*)arg
;
871 case KVM_CREATE_IRQCHIP
: {
872 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
874 case KVM_ARM_SET_DEVICE_ADDR
: {
875 struct kvm_arm_device_addr dev_addr
;
877 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
879 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
881 case KVM_ARM_PREFERRED_TARGET
: {
883 struct kvm_vcpu_init init
;
885 err
= kvm_vcpu_preferred_target(&init
);
889 if (copy_to_user(argp
, &init
, sizeof(init
)))
899 static void cpu_init_hyp_mode(void *dummy
)
901 phys_addr_t boot_pgd_ptr
;
903 unsigned long hyp_stack_ptr
;
904 unsigned long stack_page
;
905 unsigned long vector_ptr
;
907 /* Switch from the HYP stub to our own HYP init vector */
908 __hyp_set_vectors(kvm_get_idmap_vector());
910 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
911 pgd_ptr
= kvm_mmu_get_httbr();
912 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
913 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
914 vector_ptr
= (unsigned long)__kvm_hyp_vector
;
916 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
919 static int hyp_init_cpu_notify(struct notifier_block
*self
,
920 unsigned long action
, void *cpu
)
924 case CPU_STARTING_FROZEN
:
925 if (__hyp_get_vectors() == hyp_default_vectors
)
926 cpu_init_hyp_mode(NULL
);
933 static struct notifier_block hyp_init_cpu_nb
= {
934 .notifier_call
= hyp_init_cpu_notify
,
938 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
942 if (cmd
== CPU_PM_EXIT
&&
943 __hyp_get_vectors() == hyp_default_vectors
) {
944 cpu_init_hyp_mode(NULL
);
951 static struct notifier_block hyp_init_cpu_pm_nb
= {
952 .notifier_call
= hyp_init_cpu_pm_notifier
,
955 static void __init
hyp_cpu_pm_init(void)
957 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
960 static inline void hyp_cpu_pm_init(void)
966 * Inits Hyp-mode on all online CPUs
968 static int init_hyp_mode(void)
974 * Allocate Hyp PGD and setup Hyp identity mapping
976 err
= kvm_mmu_init();
981 * It is probably enough to obtain the default on one
982 * CPU. It's unlikely to be different on the others.
984 hyp_default_vectors
= __hyp_get_vectors();
987 * Allocate stack pages for Hypervisor-mode
989 for_each_possible_cpu(cpu
) {
990 unsigned long stack_page
;
992 stack_page
= __get_free_page(GFP_KERNEL
);
995 goto out_free_stack_pages
;
998 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1002 * Map the Hyp-code called directly from the host
1004 err
= create_hyp_mappings(__kvm_hyp_code_start
, __kvm_hyp_code_end
);
1006 kvm_err("Cannot map world-switch code\n");
1007 goto out_free_mappings
;
1011 * Map the Hyp stack pages
1013 for_each_possible_cpu(cpu
) {
1014 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1015 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
1018 kvm_err("Cannot map hyp stack\n");
1019 goto out_free_mappings
;
1024 * Map the host CPU structures
1026 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1027 if (!kvm_host_cpu_state
) {
1029 kvm_err("Cannot allocate host CPU state\n");
1030 goto out_free_mappings
;
1033 for_each_possible_cpu(cpu
) {
1034 kvm_cpu_context_t
*cpu_ctxt
;
1036 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1037 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1040 kvm_err("Cannot map host CPU state: %d\n", err
);
1041 goto out_free_context
;
1046 * Execute the init code on each CPU.
1048 on_each_cpu(cpu_init_hyp_mode
, NULL
, 1);
1051 * Init HYP view of VGIC
1053 err
= kvm_vgic_hyp_init();
1055 goto out_free_context
;
1058 * Init HYP architected timer support
1060 err
= kvm_timer_hyp_init();
1062 goto out_free_mappings
;
1064 #ifndef CONFIG_HOTPLUG_CPU
1065 free_boot_hyp_pgd();
1070 kvm_info("Hyp mode initialized successfully\n");
1074 free_percpu(kvm_host_cpu_state
);
1077 out_free_stack_pages
:
1078 for_each_possible_cpu(cpu
)
1079 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1081 kvm_err("error initializing Hyp mode: %d\n", err
);
1085 static void check_kvm_target_cpu(void *ret
)
1087 *(int *)ret
= kvm_target_cpu();
1090 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1092 struct kvm_vcpu
*vcpu
;
1095 mpidr
&= MPIDR_HWID_BITMASK
;
1096 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1097 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1104 * Initialize Hyp-mode and memory mappings on all CPUs.
1106 int kvm_arch_init(void *opaque
)
1111 if (!is_hyp_mode_available()) {
1112 kvm_err("HYP mode not available\n");
1116 for_each_online_cpu(cpu
) {
1117 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1119 kvm_err("Error, CPU %d not supported!\n", cpu
);
1124 cpu_notifier_register_begin();
1126 err
= init_hyp_mode();
1130 err
= __register_cpu_notifier(&hyp_init_cpu_nb
);
1132 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err
);
1136 cpu_notifier_register_done();
1140 kvm_coproc_table_init();
1143 cpu_notifier_register_done();
1147 /* NOP: Compiling as a module not supported */
1148 void kvm_arch_exit(void)
1150 kvm_perf_teardown();
1153 static int arm_init(void)
1155 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1159 module_init(arm_init
);