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 bool vgic_present
;
66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
68 BUG_ON(preemptible());
69 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
73 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
74 * Must be called from non-preemptible context
76 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
78 BUG_ON(preemptible());
79 return __this_cpu_read(kvm_arm_running_vcpu
);
83 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
85 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
87 return &kvm_arm_running_vcpu
;
90 int kvm_arch_hardware_enable(void)
95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
97 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
100 int kvm_arch_hardware_setup(void)
105 void kvm_arch_check_processor_compat(void *rtn
)
112 * kvm_arch_init_vm - initializes a VM data structure
113 * @kvm: pointer to the KVM struct
115 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
122 ret
= kvm_alloc_stage2_pgd(kvm
);
126 ret
= create_hyp_mappings(kvm
, kvm
+ 1);
128 goto out_free_stage2_pgd
;
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
= kvm_vgic_get_max_vcpus();
140 kvm_free_stage2_pgd(kvm
);
145 int kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
147 return VM_FAULT_SIGBUS
;
152 * kvm_arch_destroy_vm - destroy the VM data structure
153 * @kvm: pointer to the KVM struct
155 void kvm_arch_destroy_vm(struct kvm
*kvm
)
159 kvm_free_stage2_pgd(kvm
);
161 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
163 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
164 kvm
->vcpus
[i
] = NULL
;
168 kvm_vgic_destroy(kvm
);
171 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
175 case KVM_CAP_IRQCHIP
:
178 case KVM_CAP_DEVICE_CTRL
:
179 case KVM_CAP_USER_MEMORY
:
180 case KVM_CAP_SYNC_MMU
:
181 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
182 case KVM_CAP_ONE_REG
:
183 case KVM_CAP_ARM_PSCI
:
184 case KVM_CAP_ARM_PSCI_0_2
:
185 case KVM_CAP_READONLY_MEM
:
188 case KVM_CAP_COALESCED_MMIO
:
189 r
= KVM_COALESCED_MMIO_PAGE_OFFSET
;
191 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
194 case KVM_CAP_NR_VCPUS
:
195 r
= num_online_cpus();
197 case KVM_CAP_MAX_VCPUS
:
201 r
= kvm_arch_dev_ioctl_check_extension(ext
);
207 long kvm_arch_dev_ioctl(struct file
*filp
,
208 unsigned int ioctl
, unsigned long arg
)
214 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
217 struct kvm_vcpu
*vcpu
;
219 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
224 if (id
>= kvm
->arch
.max_vcpus
) {
229 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
235 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
239 err
= create_hyp_mappings(vcpu
, vcpu
+ 1);
245 kvm_vcpu_uninit(vcpu
);
247 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
252 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
256 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
258 kvm_mmu_free_memory_caches(vcpu
);
259 kvm_timer_vcpu_terminate(vcpu
);
260 kvm_vgic_vcpu_destroy(vcpu
);
261 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
264 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
266 kvm_arch_vcpu_free(vcpu
);
269 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
274 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
276 /* Force users to call KVM_ARM_VCPU_INIT */
277 vcpu
->arch
.target
= -1;
278 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
280 /* Set up the timer */
281 kvm_timer_vcpu_init(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_set_guest_debug(struct kvm_vcpu
*vcpu
,
307 struct kvm_guest_debug
*dbg
)
313 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
314 struct kvm_mp_state
*mp_state
)
319 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
320 struct kvm_mp_state
*mp_state
)
326 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
327 * @v: The VCPU pointer
329 * If the guest CPU is not waiting for interrupts or an interrupt line is
330 * asserted, the CPU is by definition runnable.
332 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
334 return !!v
->arch
.irq_lines
|| kvm_vgic_vcpu_pending_irq(v
);
337 /* Just ensure a guest exit from a particular CPU */
338 static void exit_vm_noop(void *info
)
342 void force_vm_exit(const cpumask_t
*mask
)
344 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
348 * need_new_vmid_gen - check that the VMID is still valid
349 * @kvm: The VM's VMID to checkt
351 * return true if there is a new generation of VMIDs being used
353 * The hardware supports only 256 values with the value zero reserved for the
354 * host, so we check if an assigned value belongs to a previous generation,
355 * which which requires us to assign a new value. If we're the first to use a
356 * VMID for the new generation, we must flush necessary caches and TLBs on all
359 static bool need_new_vmid_gen(struct kvm
*kvm
)
361 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
365 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
366 * @kvm The guest that we are about to run
368 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
369 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
372 static void update_vttbr(struct kvm
*kvm
)
374 phys_addr_t pgd_phys
;
377 if (!need_new_vmid_gen(kvm
))
380 spin_lock(&kvm_vmid_lock
);
383 * We need to re-check the vmid_gen here to ensure that if another vcpu
384 * already allocated a valid vmid for this vm, then this vcpu should
387 if (!need_new_vmid_gen(kvm
)) {
388 spin_unlock(&kvm_vmid_lock
);
392 /* First user of a new VMID generation? */
393 if (unlikely(kvm_next_vmid
== 0)) {
394 atomic64_inc(&kvm_vmid_gen
);
398 * On SMP we know no other CPUs can use this CPU's or each
399 * other's VMID after force_vm_exit returns since the
400 * kvm_vmid_lock blocks them from reentry to the guest.
402 force_vm_exit(cpu_all_mask
);
404 * Now broadcast TLB + ICACHE invalidation over the inner
405 * shareable domain to make sure all data structures are
408 kvm_call_hyp(__kvm_flush_vm_context
);
411 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
412 kvm
->arch
.vmid
= kvm_next_vmid
;
415 /* update vttbr to be used with the new vmid */
416 pgd_phys
= virt_to_phys(kvm_get_hwpgd(kvm
));
417 BUG_ON(pgd_phys
& ~VTTBR_BADDR_MASK
);
418 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK
;
419 kvm
->arch
.vttbr
= pgd_phys
| vmid
;
421 spin_unlock(&kvm_vmid_lock
);
424 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
426 struct kvm
*kvm
= vcpu
->kvm
;
429 if (likely(vcpu
->arch
.has_run_once
))
432 vcpu
->arch
.has_run_once
= true;
435 * Map the VGIC hardware resources before running a vcpu the first
438 if (unlikely(!vgic_ready(kvm
))) {
439 ret
= kvm_vgic_map_resources(kvm
);
445 * Enable the arch timers only if we have an in-kernel VGIC
446 * and it has been properly initialized, since we cannot handle
447 * interrupts from the virtual timer with a userspace gic.
449 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
))
450 kvm_timer_enable(kvm
);
455 static void vcpu_pause(struct kvm_vcpu
*vcpu
)
457 wait_queue_head_t
*wq
= kvm_arch_vcpu_wq(vcpu
);
459 wait_event_interruptible(*wq
, !vcpu
->arch
.pause
);
462 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
464 return vcpu
->arch
.target
>= 0;
468 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
469 * @vcpu: The VCPU pointer
470 * @run: The kvm_run structure pointer used for userspace state exchange
472 * This function is called through the VCPU_RUN ioctl called from user space. It
473 * will execute VM code in a loop until the time slice for the process is used
474 * or some emulation is needed from user space in which case the function will
475 * return with return value 0 and with the kvm_run structure filled in with the
476 * required data for the requested emulation.
478 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
483 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
486 ret
= kvm_vcpu_first_run_init(vcpu
);
490 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
491 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
496 if (vcpu
->sigset_active
)
497 sigprocmask(SIG_SETMASK
, &vcpu
->sigset
, &sigsaved
);
500 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
503 * Check conditions before entering the guest
507 update_vttbr(vcpu
->kvm
);
509 if (vcpu
->arch
.pause
)
512 kvm_vgic_flush_hwstate(vcpu
);
513 kvm_timer_flush_hwstate(vcpu
);
518 * Re-check atomic conditions
520 if (signal_pending(current
)) {
522 run
->exit_reason
= KVM_EXIT_INTR
;
525 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
)) {
527 kvm_timer_sync_hwstate(vcpu
);
528 kvm_vgic_sync_hwstate(vcpu
);
532 /**************************************************************
535 trace_kvm_entry(*vcpu_pc(vcpu
));
537 vcpu
->mode
= IN_GUEST_MODE
;
539 ret
= kvm_call_hyp(__kvm_vcpu_run
, vcpu
);
541 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
543 trace_kvm_exit(*vcpu_pc(vcpu
));
545 * We may have taken a host interrupt in HYP mode (ie
546 * while executing the guest). This interrupt is still
547 * pending, as we haven't serviced it yet!
549 * We're now back in SVC mode, with interrupts
550 * disabled. Enabling the interrupts now will have
551 * the effect of taking the interrupt again, in SVC
558 *************************************************************/
560 kvm_timer_sync_hwstate(vcpu
);
561 kvm_vgic_sync_hwstate(vcpu
);
563 ret
= handle_exit(vcpu
, run
, ret
);
566 if (vcpu
->sigset_active
)
567 sigprocmask(SIG_SETMASK
, &sigsaved
, NULL
);
571 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
577 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
578 bit_index
= __ffs(HCR_VI
);
579 else /* KVM_ARM_IRQ_CPU_FIQ */
580 bit_index
= __ffs(HCR_VF
);
582 ptr
= (unsigned long *)&vcpu
->arch
.irq_lines
;
584 set
= test_and_set_bit(bit_index
, ptr
);
586 set
= test_and_clear_bit(bit_index
, ptr
);
589 * If we didn't change anything, no need to wake up or kick other CPUs
595 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
596 * trigger a world-switch round on the running physical CPU to set the
597 * virtual IRQ/FIQ fields in the HCR appropriately.
604 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
607 u32 irq
= irq_level
->irq
;
608 unsigned int irq_type
, vcpu_idx
, irq_num
;
609 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
610 struct kvm_vcpu
*vcpu
= NULL
;
611 bool level
= irq_level
->level
;
613 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
614 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
615 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
617 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
620 case KVM_ARM_IRQ_TYPE_CPU
:
621 if (irqchip_in_kernel(kvm
))
624 if (vcpu_idx
>= nrcpus
)
627 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
631 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
634 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
635 case KVM_ARM_IRQ_TYPE_PPI
:
636 if (!irqchip_in_kernel(kvm
))
639 if (vcpu_idx
>= nrcpus
)
642 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
646 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
649 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
);
650 case KVM_ARM_IRQ_TYPE_SPI
:
651 if (!irqchip_in_kernel(kvm
))
654 if (irq_num
< VGIC_NR_PRIVATE_IRQS
||
655 irq_num
> KVM_ARM_IRQ_GIC_MAX
)
658 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
);
664 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
665 const struct kvm_vcpu_init
*init
)
668 int phys_target
= kvm_target_cpu();
670 if (init
->target
!= phys_target
)
674 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
675 * use the same target.
677 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
680 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
681 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
682 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
684 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
688 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
689 * use the same feature set.
691 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
692 test_bit(i
, vcpu
->arch
.features
) != set
)
696 set_bit(i
, vcpu
->arch
.features
);
699 vcpu
->arch
.target
= phys_target
;
701 /* Now we know what it is, we can reset it. */
702 return kvm_reset_vcpu(vcpu
);
706 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
707 struct kvm_vcpu_init
*init
)
711 ret
= kvm_vcpu_set_target(vcpu
, init
);
716 * Ensure a rebooted VM will fault in RAM pages and detect if the
717 * guest MMU is turned off and flush the caches as needed.
719 if (vcpu
->arch
.has_run_once
)
720 stage2_unmap_vm(vcpu
->kvm
);
722 vcpu_reset_hcr(vcpu
);
725 * Handle the "start in power-off" case by marking the VCPU as paused.
727 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
728 vcpu
->arch
.pause
= true;
730 vcpu
->arch
.pause
= false;
735 long kvm_arch_vcpu_ioctl(struct file
*filp
,
736 unsigned int ioctl
, unsigned long arg
)
738 struct kvm_vcpu
*vcpu
= filp
->private_data
;
739 void __user
*argp
= (void __user
*)arg
;
742 case KVM_ARM_VCPU_INIT
: {
743 struct kvm_vcpu_init init
;
745 if (copy_from_user(&init
, argp
, sizeof(init
)))
748 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
750 case KVM_SET_ONE_REG
:
751 case KVM_GET_ONE_REG
: {
752 struct kvm_one_reg reg
;
754 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
757 if (copy_from_user(®
, argp
, sizeof(reg
)))
759 if (ioctl
== KVM_SET_ONE_REG
)
760 return kvm_arm_set_reg(vcpu
, ®
);
762 return kvm_arm_get_reg(vcpu
, ®
);
764 case KVM_GET_REG_LIST
: {
765 struct kvm_reg_list __user
*user_list
= argp
;
766 struct kvm_reg_list reg_list
;
769 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
772 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
775 reg_list
.n
= kvm_arm_num_regs(vcpu
);
776 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
780 return kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
788 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
790 * @log: slot id and address to which we copy the log
792 * Steps 1-4 below provide general overview of dirty page logging. See
793 * kvm_get_dirty_log_protect() function description for additional details.
795 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
796 * always flush the TLB (step 4) even if previous step failed and the dirty
797 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
798 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
799 * writes will be marked dirty for next log read.
801 * 1. Take a snapshot of the bit and clear it if needed.
802 * 2. Write protect the corresponding page.
803 * 3. Copy the snapshot to the userspace.
804 * 4. Flush TLB's if needed.
806 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
808 bool is_dirty
= false;
811 mutex_lock(&kvm
->slots_lock
);
813 r
= kvm_get_dirty_log_protect(kvm
, log
, &is_dirty
);
816 kvm_flush_remote_tlbs(kvm
);
818 mutex_unlock(&kvm
->slots_lock
);
822 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
823 struct kvm_arm_device_addr
*dev_addr
)
825 unsigned long dev_id
, type
;
827 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
828 KVM_ARM_DEVICE_ID_SHIFT
;
829 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
830 KVM_ARM_DEVICE_TYPE_SHIFT
;
833 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
: {
851 return kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
855 case KVM_ARM_SET_DEVICE_ADDR
: {
856 struct kvm_arm_device_addr dev_addr
;
858 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
860 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
862 case KVM_ARM_PREFERRED_TARGET
: {
864 struct kvm_vcpu_init init
;
866 err
= kvm_vcpu_preferred_target(&init
);
870 if (copy_to_user(argp
, &init
, sizeof(init
)))
880 static void cpu_init_hyp_mode(void *dummy
)
882 phys_addr_t boot_pgd_ptr
;
884 unsigned long hyp_stack_ptr
;
885 unsigned long stack_page
;
886 unsigned long vector_ptr
;
888 /* Switch from the HYP stub to our own HYP init vector */
889 __hyp_set_vectors(kvm_get_idmap_vector());
891 boot_pgd_ptr
= kvm_mmu_get_boot_httbr();
892 pgd_ptr
= kvm_mmu_get_httbr();
893 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
894 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
895 vector_ptr
= (unsigned long)__kvm_hyp_vector
;
897 __cpu_init_hyp_mode(boot_pgd_ptr
, pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
900 static int hyp_init_cpu_notify(struct notifier_block
*self
,
901 unsigned long action
, void *cpu
)
905 case CPU_STARTING_FROZEN
:
906 if (__hyp_get_vectors() == hyp_default_vectors
)
907 cpu_init_hyp_mode(NULL
);
914 static struct notifier_block hyp_init_cpu_nb
= {
915 .notifier_call
= hyp_init_cpu_notify
,
919 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
923 if (cmd
== CPU_PM_EXIT
&&
924 __hyp_get_vectors() == hyp_default_vectors
) {
925 cpu_init_hyp_mode(NULL
);
932 static struct notifier_block hyp_init_cpu_pm_nb
= {
933 .notifier_call
= hyp_init_cpu_pm_notifier
,
936 static void __init
hyp_cpu_pm_init(void)
938 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
941 static inline void hyp_cpu_pm_init(void)
947 * Inits Hyp-mode on all online CPUs
949 static int init_hyp_mode(void)
955 * Allocate Hyp PGD and setup Hyp identity mapping
957 err
= kvm_mmu_init();
962 * It is probably enough to obtain the default on one
963 * CPU. It's unlikely to be different on the others.
965 hyp_default_vectors
= __hyp_get_vectors();
968 * Allocate stack pages for Hypervisor-mode
970 for_each_possible_cpu(cpu
) {
971 unsigned long stack_page
;
973 stack_page
= __get_free_page(GFP_KERNEL
);
976 goto out_free_stack_pages
;
979 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
983 * Map the Hyp-code called directly from the host
985 err
= create_hyp_mappings(__kvm_hyp_code_start
, __kvm_hyp_code_end
);
987 kvm_err("Cannot map world-switch code\n");
988 goto out_free_mappings
;
992 * Map the Hyp stack pages
994 for_each_possible_cpu(cpu
) {
995 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
996 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
);
999 kvm_err("Cannot map hyp stack\n");
1000 goto out_free_mappings
;
1005 * Map the host CPU structures
1007 kvm_host_cpu_state
= alloc_percpu(kvm_cpu_context_t
);
1008 if (!kvm_host_cpu_state
) {
1010 kvm_err("Cannot allocate host CPU state\n");
1011 goto out_free_mappings
;
1014 for_each_possible_cpu(cpu
) {
1015 kvm_cpu_context_t
*cpu_ctxt
;
1017 cpu_ctxt
= per_cpu_ptr(kvm_host_cpu_state
, cpu
);
1018 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1);
1021 kvm_err("Cannot map host CPU state: %d\n", err
);
1022 goto out_free_context
;
1027 * Execute the init code on each CPU.
1029 on_each_cpu(cpu_init_hyp_mode
, NULL
, 1);
1032 * Init HYP view of VGIC
1034 err
= kvm_vgic_hyp_init();
1036 goto out_free_context
;
1038 #ifdef CONFIG_KVM_ARM_VGIC
1039 vgic_present
= true;
1043 * Init HYP architected timer support
1045 err
= kvm_timer_hyp_init();
1047 goto out_free_mappings
;
1049 #ifndef CONFIG_HOTPLUG_CPU
1050 free_boot_hyp_pgd();
1055 kvm_info("Hyp mode initialized successfully\n");
1059 free_percpu(kvm_host_cpu_state
);
1062 out_free_stack_pages
:
1063 for_each_possible_cpu(cpu
)
1064 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1066 kvm_err("error initializing Hyp mode: %d\n", err
);
1070 static void check_kvm_target_cpu(void *ret
)
1072 *(int *)ret
= kvm_target_cpu();
1075 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1077 struct kvm_vcpu
*vcpu
;
1080 mpidr
&= MPIDR_HWID_BITMASK
;
1081 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1082 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1089 * Initialize Hyp-mode and memory mappings on all CPUs.
1091 int kvm_arch_init(void *opaque
)
1096 if (!is_hyp_mode_available()) {
1097 kvm_err("HYP mode not available\n");
1101 for_each_online_cpu(cpu
) {
1102 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1104 kvm_err("Error, CPU %d not supported!\n", cpu
);
1109 cpu_notifier_register_begin();
1111 err
= init_hyp_mode();
1115 err
= __register_cpu_notifier(&hyp_init_cpu_nb
);
1117 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err
);
1121 cpu_notifier_register_done();
1125 kvm_coproc_table_init();
1128 cpu_notifier_register_done();
1132 /* NOP: Compiling as a module not supported */
1133 void kvm_arch_exit(void)
1135 kvm_perf_teardown();
1138 static int arm_init(void)
1140 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1144 module_init(arm_init
);