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/bug.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/list.h>
25 #include <linux/module.h>
26 #include <linux/vmalloc.h>
28 #include <linux/mman.h>
29 #include <linux/sched.h>
30 #include <linux/kvm.h>
31 #include <linux/kvm_irqfd.h>
32 #include <linux/irqbypass.h>
33 #include <linux/sched/stat.h>
34 #include <trace/events/kvm.h>
35 #include <kvm/arm_pmu.h>
36 #include <kvm/arm_psci.h>
38 #define CREATE_TRACE_POINTS
41 #include <linux/uaccess.h>
42 #include <asm/ptrace.h>
44 #include <asm/tlbflush.h>
45 #include <asm/cacheflush.h>
46 #include <asm/cpufeature.h>
48 #include <asm/kvm_arm.h>
49 #include <asm/kvm_asm.h>
50 #include <asm/kvm_mmu.h>
51 #include <asm/kvm_emulate.h>
52 #include <asm/kvm_coproc.h>
53 #include <asm/sections.h>
56 __asm__(".arch_extension virt");
59 DEFINE_PER_CPU(kvm_cpu_context_t
, kvm_host_cpu_state
);
60 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page
);
62 /* Per-CPU variable containing the currently running vcpu. */
63 static DEFINE_PER_CPU(struct kvm_vcpu
*, kvm_arm_running_vcpu
);
65 /* The VMID used in the VTTBR */
66 static atomic64_t kvm_vmid_gen
= ATOMIC64_INIT(1);
67 static u32 kvm_next_vmid
;
68 static unsigned int kvm_vmid_bits __read_mostly
;
69 static DEFINE_RWLOCK(kvm_vmid_lock
);
71 static bool vgic_present
;
73 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled
);
75 static void kvm_arm_set_running_vcpu(struct kvm_vcpu
*vcpu
)
77 __this_cpu_write(kvm_arm_running_vcpu
, vcpu
);
80 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use
);
83 * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
84 * Must be called from non-preemptible context
86 struct kvm_vcpu
*kvm_arm_get_running_vcpu(void)
88 return __this_cpu_read(kvm_arm_running_vcpu
);
92 * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
94 struct kvm_vcpu
* __percpu
*kvm_get_running_vcpus(void)
96 return &kvm_arm_running_vcpu
;
99 int kvm_arch_vcpu_should_kick(struct kvm_vcpu
*vcpu
)
101 return kvm_vcpu_exiting_guest_mode(vcpu
) == IN_GUEST_MODE
;
104 int kvm_arch_hardware_setup(void)
109 void kvm_arch_check_processor_compat(void *rtn
)
116 * kvm_arch_init_vm - initializes a VM data structure
117 * @kvm: pointer to the KVM struct
119 int kvm_arch_init_vm(struct kvm
*kvm
, unsigned long type
)
123 ret
= kvm_arm_setup_stage2(kvm
, type
);
127 kvm
->arch
.last_vcpu_ran
= alloc_percpu(typeof(*kvm
->arch
.last_vcpu_ran
));
128 if (!kvm
->arch
.last_vcpu_ran
)
131 for_each_possible_cpu(cpu
)
132 *per_cpu_ptr(kvm
->arch
.last_vcpu_ran
, cpu
) = -1;
134 ret
= kvm_alloc_stage2_pgd(kvm
);
138 ret
= create_hyp_mappings(kvm
, kvm
+ 1, PAGE_HYP
);
140 goto out_free_stage2_pgd
;
142 kvm_vgic_early_init(kvm
);
144 /* Mark the initial VMID generation invalid */
145 kvm
->arch
.vmid_gen
= 0;
147 /* The maximum number of VCPUs is limited by the host's GIC model */
148 kvm
->arch
.max_vcpus
= vgic_present
?
149 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS
;
153 kvm_free_stage2_pgd(kvm
);
155 free_percpu(kvm
->arch
.last_vcpu_ran
);
156 kvm
->arch
.last_vcpu_ran
= NULL
;
160 bool kvm_arch_has_vcpu_debugfs(void)
165 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu
*vcpu
)
170 vm_fault_t
kvm_arch_vcpu_fault(struct kvm_vcpu
*vcpu
, struct vm_fault
*vmf
)
172 return VM_FAULT_SIGBUS
;
177 * kvm_arch_destroy_vm - destroy the VM data structure
178 * @kvm: pointer to the KVM struct
180 void kvm_arch_destroy_vm(struct kvm
*kvm
)
184 kvm_vgic_destroy(kvm
);
186 free_percpu(kvm
->arch
.last_vcpu_ran
);
187 kvm
->arch
.last_vcpu_ran
= NULL
;
189 for (i
= 0; i
< KVM_MAX_VCPUS
; ++i
) {
191 kvm_arch_vcpu_free(kvm
->vcpus
[i
]);
192 kvm
->vcpus
[i
] = NULL
;
195 atomic_set(&kvm
->online_vcpus
, 0);
198 int kvm_vm_ioctl_check_extension(struct kvm
*kvm
, long ext
)
202 case KVM_CAP_IRQCHIP
:
205 case KVM_CAP_IOEVENTFD
:
206 case KVM_CAP_DEVICE_CTRL
:
207 case KVM_CAP_USER_MEMORY
:
208 case KVM_CAP_SYNC_MMU
:
209 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
210 case KVM_CAP_ONE_REG
:
211 case KVM_CAP_ARM_PSCI
:
212 case KVM_CAP_ARM_PSCI_0_2
:
213 case KVM_CAP_READONLY_MEM
:
214 case KVM_CAP_MP_STATE
:
215 case KVM_CAP_IMMEDIATE_EXIT
:
216 case KVM_CAP_VCPU_EVENTS
:
219 case KVM_CAP_ARM_SET_DEVICE_ADDR
:
222 case KVM_CAP_NR_VCPUS
:
223 r
= num_online_cpus();
225 case KVM_CAP_MAX_VCPUS
:
228 case KVM_CAP_NR_MEMSLOTS
:
229 r
= KVM_USER_MEM_SLOTS
;
231 case KVM_CAP_MSI_DEVID
:
235 r
= kvm
->arch
.vgic
.msis_require_devid
;
237 case KVM_CAP_ARM_USER_IRQ
:
239 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
240 * (bump this number if adding more devices)
245 r
= kvm_arch_vm_ioctl_check_extension(kvm
, ext
);
251 long kvm_arch_dev_ioctl(struct file
*filp
,
252 unsigned int ioctl
, unsigned long arg
)
257 struct kvm
*kvm_arch_alloc_vm(void)
260 return kzalloc(sizeof(struct kvm
), GFP_KERNEL
);
262 return vzalloc(sizeof(struct kvm
));
265 void kvm_arch_free_vm(struct kvm
*kvm
)
273 struct kvm_vcpu
*kvm_arch_vcpu_create(struct kvm
*kvm
, unsigned int id
)
276 struct kvm_vcpu
*vcpu
;
278 if (irqchip_in_kernel(kvm
) && vgic_initialized(kvm
)) {
283 if (id
>= kvm
->arch
.max_vcpus
) {
288 vcpu
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
294 err
= kvm_vcpu_init(vcpu
, kvm
, id
);
298 err
= create_hyp_mappings(vcpu
, vcpu
+ 1, PAGE_HYP
);
304 kvm_vcpu_uninit(vcpu
);
306 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
311 void kvm_arch_vcpu_postcreate(struct kvm_vcpu
*vcpu
)
315 void kvm_arch_vcpu_free(struct kvm_vcpu
*vcpu
)
317 if (vcpu
->arch
.has_run_once
&& unlikely(!irqchip_in_kernel(vcpu
->kvm
)))
318 static_branch_dec(&userspace_irqchip_in_use
);
320 kvm_mmu_free_memory_caches(vcpu
);
321 kvm_timer_vcpu_terminate(vcpu
);
322 kvm_pmu_vcpu_destroy(vcpu
);
323 kvm_vcpu_uninit(vcpu
);
324 kmem_cache_free(kvm_vcpu_cache
, vcpu
);
327 void kvm_arch_vcpu_destroy(struct kvm_vcpu
*vcpu
)
329 kvm_arch_vcpu_free(vcpu
);
332 int kvm_cpu_has_pending_timer(struct kvm_vcpu
*vcpu
)
334 return kvm_timer_is_pending(vcpu
);
337 void kvm_arch_vcpu_blocking(struct kvm_vcpu
*vcpu
)
339 kvm_timer_schedule(vcpu
);
340 kvm_vgic_v4_enable_doorbell(vcpu
);
343 void kvm_arch_vcpu_unblocking(struct kvm_vcpu
*vcpu
)
345 kvm_timer_unschedule(vcpu
);
346 kvm_vgic_v4_disable_doorbell(vcpu
);
349 int kvm_arch_vcpu_init(struct kvm_vcpu
*vcpu
)
351 /* Force users to call KVM_ARM_VCPU_INIT */
352 vcpu
->arch
.target
= -1;
353 bitmap_zero(vcpu
->arch
.features
, KVM_VCPU_MAX_FEATURES
);
355 /* Set up the timer */
356 kvm_timer_vcpu_init(vcpu
);
358 kvm_arm_reset_debug_ptr(vcpu
);
360 return kvm_vgic_vcpu_init(vcpu
);
363 void kvm_arch_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
367 last_ran
= this_cpu_ptr(vcpu
->kvm
->arch
.last_vcpu_ran
);
370 * We might get preempted before the vCPU actually runs, but
371 * over-invalidation doesn't affect correctness.
373 if (*last_ran
!= vcpu
->vcpu_id
) {
374 kvm_call_hyp(__kvm_tlb_flush_local_vmid
, vcpu
);
375 *last_ran
= vcpu
->vcpu_id
;
379 vcpu
->arch
.host_cpu_context
= this_cpu_ptr(&kvm_host_cpu_state
);
381 kvm_arm_set_running_vcpu(vcpu
);
383 kvm_timer_vcpu_load(vcpu
);
384 kvm_vcpu_load_sysregs(vcpu
);
385 kvm_arch_vcpu_load_fp(vcpu
);
387 if (single_task_running())
388 vcpu_clear_wfe_traps(vcpu
);
390 vcpu_set_wfe_traps(vcpu
);
393 void kvm_arch_vcpu_put(struct kvm_vcpu
*vcpu
)
395 kvm_arch_vcpu_put_fp(vcpu
);
396 kvm_vcpu_put_sysregs(vcpu
);
397 kvm_timer_vcpu_put(vcpu
);
402 kvm_arm_set_running_vcpu(NULL
);
405 static void vcpu_power_off(struct kvm_vcpu
*vcpu
)
407 vcpu
->arch
.power_off
= true;
408 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
412 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu
*vcpu
,
413 struct kvm_mp_state
*mp_state
)
415 if (vcpu
->arch
.power_off
)
416 mp_state
->mp_state
= KVM_MP_STATE_STOPPED
;
418 mp_state
->mp_state
= KVM_MP_STATE_RUNNABLE
;
423 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu
*vcpu
,
424 struct kvm_mp_state
*mp_state
)
428 switch (mp_state
->mp_state
) {
429 case KVM_MP_STATE_RUNNABLE
:
430 vcpu
->arch
.power_off
= false;
432 case KVM_MP_STATE_STOPPED
:
433 vcpu_power_off(vcpu
);
443 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
444 * @v: The VCPU pointer
446 * If the guest CPU is not waiting for interrupts or an interrupt line is
447 * asserted, the CPU is by definition runnable.
449 int kvm_arch_vcpu_runnable(struct kvm_vcpu
*v
)
451 bool irq_lines
= *vcpu_hcr(v
) & (HCR_VI
| HCR_VF
);
452 return ((irq_lines
|| kvm_vgic_vcpu_pending_irq(v
))
453 && !v
->arch
.power_off
&& !v
->arch
.pause
);
456 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu
*vcpu
)
458 return vcpu_mode_priv(vcpu
);
461 /* Just ensure a guest exit from a particular CPU */
462 static void exit_vm_noop(void *info
)
466 void force_vm_exit(const cpumask_t
*mask
)
469 smp_call_function_many(mask
, exit_vm_noop
, NULL
, true);
474 * need_new_vmid_gen - check that the VMID is still valid
475 * @kvm: The VM's VMID to check
477 * return true if there is a new generation of VMIDs being used
479 * The hardware supports only 256 values with the value zero reserved for the
480 * host, so we check if an assigned value belongs to a previous generation,
481 * which which requires us to assign a new value. If we're the first to use a
482 * VMID for the new generation, we must flush necessary caches and TLBs on all
485 static bool need_new_vmid_gen(struct kvm
*kvm
)
487 return unlikely(kvm
->arch
.vmid_gen
!= atomic64_read(&kvm_vmid_gen
));
491 * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
492 * @kvm The guest that we are about to run
494 * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
495 * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
498 static void update_vttbr(struct kvm
*kvm
)
500 phys_addr_t pgd_phys
;
501 u64 vmid
, cnp
= kvm_cpu_has_cnp() ? VTTBR_CNP_BIT
: 0;
504 read_lock(&kvm_vmid_lock
);
505 new_gen
= need_new_vmid_gen(kvm
);
506 read_unlock(&kvm_vmid_lock
);
511 write_lock(&kvm_vmid_lock
);
514 * We need to re-check the vmid_gen here to ensure that if another vcpu
515 * already allocated a valid vmid for this vm, then this vcpu should
518 if (!need_new_vmid_gen(kvm
)) {
519 write_unlock(&kvm_vmid_lock
);
523 /* First user of a new VMID generation? */
524 if (unlikely(kvm_next_vmid
== 0)) {
525 atomic64_inc(&kvm_vmid_gen
);
529 * On SMP we know no other CPUs can use this CPU's or each
530 * other's VMID after force_vm_exit returns since the
531 * kvm_vmid_lock blocks them from reentry to the guest.
533 force_vm_exit(cpu_all_mask
);
535 * Now broadcast TLB + ICACHE invalidation over the inner
536 * shareable domain to make sure all data structures are
539 kvm_call_hyp(__kvm_flush_vm_context
);
542 kvm
->arch
.vmid_gen
= atomic64_read(&kvm_vmid_gen
);
543 kvm
->arch
.vmid
= kvm_next_vmid
;
545 kvm_next_vmid
&= (1 << kvm_vmid_bits
) - 1;
547 /* update vttbr to be used with the new vmid */
548 pgd_phys
= virt_to_phys(kvm
->arch
.pgd
);
549 BUG_ON(pgd_phys
& ~kvm_vttbr_baddr_mask(kvm
));
550 vmid
= ((u64
)(kvm
->arch
.vmid
) << VTTBR_VMID_SHIFT
) & VTTBR_VMID_MASK(kvm_vmid_bits
);
551 kvm
->arch
.vttbr
= kvm_phys_to_vttbr(pgd_phys
) | vmid
| cnp
;
553 write_unlock(&kvm_vmid_lock
);
556 static int kvm_vcpu_first_run_init(struct kvm_vcpu
*vcpu
)
558 struct kvm
*kvm
= vcpu
->kvm
;
561 if (likely(vcpu
->arch
.has_run_once
))
564 vcpu
->arch
.has_run_once
= true;
566 if (likely(irqchip_in_kernel(kvm
))) {
568 * Map the VGIC hardware resources before running a vcpu the
569 * first time on this VM.
571 if (unlikely(!vgic_ready(kvm
))) {
572 ret
= kvm_vgic_map_resources(kvm
);
578 * Tell the rest of the code that there are userspace irqchip
581 static_branch_inc(&userspace_irqchip_in_use
);
584 ret
= kvm_timer_enable(vcpu
);
588 ret
= kvm_arm_pmu_v3_enable(vcpu
);
593 bool kvm_arch_intc_initialized(struct kvm
*kvm
)
595 return vgic_initialized(kvm
);
598 void kvm_arm_halt_guest(struct kvm
*kvm
)
601 struct kvm_vcpu
*vcpu
;
603 kvm_for_each_vcpu(i
, vcpu
, kvm
)
604 vcpu
->arch
.pause
= true;
605 kvm_make_all_cpus_request(kvm
, KVM_REQ_SLEEP
);
608 void kvm_arm_resume_guest(struct kvm
*kvm
)
611 struct kvm_vcpu
*vcpu
;
613 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
614 vcpu
->arch
.pause
= false;
615 swake_up_one(kvm_arch_vcpu_wq(vcpu
));
619 static void vcpu_req_sleep(struct kvm_vcpu
*vcpu
)
621 struct swait_queue_head
*wq
= kvm_arch_vcpu_wq(vcpu
);
623 swait_event_interruptible_exclusive(*wq
, ((!vcpu
->arch
.power_off
) &&
624 (!vcpu
->arch
.pause
)));
626 if (vcpu
->arch
.power_off
|| vcpu
->arch
.pause
) {
627 /* Awaken to handle a signal, request we sleep again later. */
628 kvm_make_request(KVM_REQ_SLEEP
, vcpu
);
632 static int kvm_vcpu_initialized(struct kvm_vcpu
*vcpu
)
634 return vcpu
->arch
.target
>= 0;
637 static void check_vcpu_requests(struct kvm_vcpu
*vcpu
)
639 if (kvm_request_pending(vcpu
)) {
640 if (kvm_check_request(KVM_REQ_SLEEP
, vcpu
))
641 vcpu_req_sleep(vcpu
);
644 * Clear IRQ_PENDING requests that were made to guarantee
645 * that a VCPU sees new virtual interrupts.
647 kvm_check_request(KVM_REQ_IRQ_PENDING
, vcpu
);
652 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
653 * @vcpu: The VCPU pointer
654 * @run: The kvm_run structure pointer used for userspace state exchange
656 * This function is called through the VCPU_RUN ioctl called from user space. It
657 * will execute VM code in a loop until the time slice for the process is used
658 * or some emulation is needed from user space in which case the function will
659 * return with return value 0 and with the kvm_run structure filled in with the
660 * required data for the requested emulation.
662 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu
*vcpu
, struct kvm_run
*run
)
666 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
669 ret
= kvm_vcpu_first_run_init(vcpu
);
673 if (run
->exit_reason
== KVM_EXIT_MMIO
) {
674 ret
= kvm_handle_mmio_return(vcpu
, vcpu
->run
);
677 if (kvm_arm_handle_step_debug(vcpu
, vcpu
->run
))
681 if (run
->immediate_exit
)
686 kvm_sigset_activate(vcpu
);
689 run
->exit_reason
= KVM_EXIT_UNKNOWN
;
692 * Check conditions before entering the guest
696 update_vttbr(vcpu
->kvm
);
698 check_vcpu_requests(vcpu
);
701 * Preparing the interrupts to be injected also
702 * involves poking the GIC, which must be done in a
703 * non-preemptible context.
707 kvm_pmu_flush_hwstate(vcpu
);
711 kvm_vgic_flush_hwstate(vcpu
);
714 * Exit if we have a signal pending so that we can deliver the
715 * signal to user space.
717 if (signal_pending(current
)) {
719 run
->exit_reason
= KVM_EXIT_INTR
;
723 * If we're using a userspace irqchip, then check if we need
724 * to tell a userspace irqchip about timer or PMU level
725 * changes and if so, exit to userspace (the actual level
726 * state gets updated in kvm_timer_update_run and
727 * kvm_pmu_update_run below).
729 if (static_branch_unlikely(&userspace_irqchip_in_use
)) {
730 if (kvm_timer_should_notify_user(vcpu
) ||
731 kvm_pmu_should_notify_user(vcpu
)) {
733 run
->exit_reason
= KVM_EXIT_INTR
;
738 * Ensure we set mode to IN_GUEST_MODE after we disable
739 * interrupts and before the final VCPU requests check.
740 * See the comment in kvm_vcpu_exiting_guest_mode() and
741 * Documentation/virtual/kvm/vcpu-requests.rst
743 smp_store_mb(vcpu
->mode
, IN_GUEST_MODE
);
745 if (ret
<= 0 || need_new_vmid_gen(vcpu
->kvm
) ||
746 kvm_request_pending(vcpu
)) {
747 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
748 isb(); /* Ensure work in x_flush_hwstate is committed */
749 kvm_pmu_sync_hwstate(vcpu
);
750 if (static_branch_unlikely(&userspace_irqchip_in_use
))
751 kvm_timer_sync_hwstate(vcpu
);
752 kvm_vgic_sync_hwstate(vcpu
);
758 kvm_arm_setup_debug(vcpu
);
760 /**************************************************************
763 trace_kvm_entry(*vcpu_pc(vcpu
));
764 guest_enter_irqoff();
767 kvm_arm_vhe_guest_enter();
768 ret
= kvm_vcpu_run_vhe(vcpu
);
769 kvm_arm_vhe_guest_exit();
771 ret
= kvm_call_hyp(__kvm_vcpu_run_nvhe
, vcpu
);
774 vcpu
->mode
= OUTSIDE_GUEST_MODE
;
778 *************************************************************/
780 kvm_arm_clear_debug(vcpu
);
783 * We must sync the PMU state before the vgic state so
784 * that the vgic can properly sample the updated state of the
787 kvm_pmu_sync_hwstate(vcpu
);
790 * Sync the vgic state before syncing the timer state because
791 * the timer code needs to know if the virtual timer
792 * interrupts are active.
794 kvm_vgic_sync_hwstate(vcpu
);
797 * Sync the timer hardware state before enabling interrupts as
798 * we don't want vtimer interrupts to race with syncing the
799 * timer virtual interrupt state.
801 if (static_branch_unlikely(&userspace_irqchip_in_use
))
802 kvm_timer_sync_hwstate(vcpu
);
804 kvm_arch_vcpu_ctxsync_fp(vcpu
);
807 * We may have taken a host interrupt in HYP mode (ie
808 * while executing the guest). This interrupt is still
809 * pending, as we haven't serviced it yet!
811 * We're now back in SVC mode, with interrupts
812 * disabled. Enabling the interrupts now will have
813 * the effect of taking the interrupt again, in SVC
819 * We do local_irq_enable() before calling guest_exit() so
820 * that if a timer interrupt hits while running the guest we
821 * account that tick as being spent in the guest. We enable
822 * preemption after calling guest_exit() so that if we get
823 * preempted we make sure ticks after that is not counted as
827 trace_kvm_exit(ret
, kvm_vcpu_trap_get_class(vcpu
), *vcpu_pc(vcpu
));
829 /* Exit types that need handling before we can be preempted */
830 handle_exit_early(vcpu
, run
, ret
);
834 ret
= handle_exit(vcpu
, run
, ret
);
837 /* Tell userspace about in-kernel device output levels */
838 if (unlikely(!irqchip_in_kernel(vcpu
->kvm
))) {
839 kvm_timer_update_run(vcpu
);
840 kvm_pmu_update_run(vcpu
);
843 kvm_sigset_deactivate(vcpu
);
849 static int vcpu_interrupt_line(struct kvm_vcpu
*vcpu
, int number
, bool level
)
855 if (number
== KVM_ARM_IRQ_CPU_IRQ
)
856 bit_index
= __ffs(HCR_VI
);
857 else /* KVM_ARM_IRQ_CPU_FIQ */
858 bit_index
= __ffs(HCR_VF
);
860 hcr
= vcpu_hcr(vcpu
);
862 set
= test_and_set_bit(bit_index
, hcr
);
864 set
= test_and_clear_bit(bit_index
, hcr
);
867 * If we didn't change anything, no need to wake up or kick other CPUs
873 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
874 * trigger a world-switch round on the running physical CPU to set the
875 * virtual IRQ/FIQ fields in the HCR appropriately.
877 kvm_make_request(KVM_REQ_IRQ_PENDING
, vcpu
);
883 int kvm_vm_ioctl_irq_line(struct kvm
*kvm
, struct kvm_irq_level
*irq_level
,
886 u32 irq
= irq_level
->irq
;
887 unsigned int irq_type
, vcpu_idx
, irq_num
;
888 int nrcpus
= atomic_read(&kvm
->online_vcpus
);
889 struct kvm_vcpu
*vcpu
= NULL
;
890 bool level
= irq_level
->level
;
892 irq_type
= (irq
>> KVM_ARM_IRQ_TYPE_SHIFT
) & KVM_ARM_IRQ_TYPE_MASK
;
893 vcpu_idx
= (irq
>> KVM_ARM_IRQ_VCPU_SHIFT
) & KVM_ARM_IRQ_VCPU_MASK
;
894 irq_num
= (irq
>> KVM_ARM_IRQ_NUM_SHIFT
) & KVM_ARM_IRQ_NUM_MASK
;
896 trace_kvm_irq_line(irq_type
, vcpu_idx
, irq_num
, irq_level
->level
);
899 case KVM_ARM_IRQ_TYPE_CPU
:
900 if (irqchip_in_kernel(kvm
))
903 if (vcpu_idx
>= nrcpus
)
906 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
910 if (irq_num
> KVM_ARM_IRQ_CPU_FIQ
)
913 return vcpu_interrupt_line(vcpu
, irq_num
, level
);
914 case KVM_ARM_IRQ_TYPE_PPI
:
915 if (!irqchip_in_kernel(kvm
))
918 if (vcpu_idx
>= nrcpus
)
921 vcpu
= kvm_get_vcpu(kvm
, vcpu_idx
);
925 if (irq_num
< VGIC_NR_SGIS
|| irq_num
>= VGIC_NR_PRIVATE_IRQS
)
928 return kvm_vgic_inject_irq(kvm
, vcpu
->vcpu_id
, irq_num
, level
, NULL
);
929 case KVM_ARM_IRQ_TYPE_SPI
:
930 if (!irqchip_in_kernel(kvm
))
933 if (irq_num
< VGIC_NR_PRIVATE_IRQS
)
936 return kvm_vgic_inject_irq(kvm
, 0, irq_num
, level
, NULL
);
942 static int kvm_vcpu_set_target(struct kvm_vcpu
*vcpu
,
943 const struct kvm_vcpu_init
*init
)
946 int phys_target
= kvm_target_cpu();
948 if (init
->target
!= phys_target
)
952 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
953 * use the same target.
955 if (vcpu
->arch
.target
!= -1 && vcpu
->arch
.target
!= init
->target
)
958 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
959 for (i
= 0; i
< sizeof(init
->features
) * 8; i
++) {
960 bool set
= (init
->features
[i
/ 32] & (1 << (i
% 32)));
962 if (set
&& i
>= KVM_VCPU_MAX_FEATURES
)
966 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
967 * use the same feature set.
969 if (vcpu
->arch
.target
!= -1 && i
< KVM_VCPU_MAX_FEATURES
&&
970 test_bit(i
, vcpu
->arch
.features
) != set
)
974 set_bit(i
, vcpu
->arch
.features
);
977 vcpu
->arch
.target
= phys_target
;
979 /* Now we know what it is, we can reset it. */
980 return kvm_reset_vcpu(vcpu
);
984 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu
*vcpu
,
985 struct kvm_vcpu_init
*init
)
989 ret
= kvm_vcpu_set_target(vcpu
, init
);
994 * Ensure a rebooted VM will fault in RAM pages and detect if the
995 * guest MMU is turned off and flush the caches as needed.
997 if (vcpu
->arch
.has_run_once
)
998 stage2_unmap_vm(vcpu
->kvm
);
1000 vcpu_reset_hcr(vcpu
);
1003 * Handle the "start in power-off" case.
1005 if (test_bit(KVM_ARM_VCPU_POWER_OFF
, vcpu
->arch
.features
))
1006 vcpu_power_off(vcpu
);
1008 vcpu
->arch
.power_off
= false;
1013 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu
*vcpu
,
1014 struct kvm_device_attr
*attr
)
1018 switch (attr
->group
) {
1020 ret
= kvm_arm_vcpu_arch_set_attr(vcpu
, attr
);
1027 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu
*vcpu
,
1028 struct kvm_device_attr
*attr
)
1032 switch (attr
->group
) {
1034 ret
= kvm_arm_vcpu_arch_get_attr(vcpu
, attr
);
1041 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu
*vcpu
,
1042 struct kvm_device_attr
*attr
)
1046 switch (attr
->group
) {
1048 ret
= kvm_arm_vcpu_arch_has_attr(vcpu
, attr
);
1055 static int kvm_arm_vcpu_get_events(struct kvm_vcpu
*vcpu
,
1056 struct kvm_vcpu_events
*events
)
1058 memset(events
, 0, sizeof(*events
));
1060 return __kvm_arm_vcpu_get_events(vcpu
, events
);
1063 static int kvm_arm_vcpu_set_events(struct kvm_vcpu
*vcpu
,
1064 struct kvm_vcpu_events
*events
)
1068 /* check whether the reserved field is zero */
1069 for (i
= 0; i
< ARRAY_SIZE(events
->reserved
); i
++)
1070 if (events
->reserved
[i
])
1073 /* check whether the pad field is zero */
1074 for (i
= 0; i
< ARRAY_SIZE(events
->exception
.pad
); i
++)
1075 if (events
->exception
.pad
[i
])
1078 return __kvm_arm_vcpu_set_events(vcpu
, events
);
1081 long kvm_arch_vcpu_ioctl(struct file
*filp
,
1082 unsigned int ioctl
, unsigned long arg
)
1084 struct kvm_vcpu
*vcpu
= filp
->private_data
;
1085 void __user
*argp
= (void __user
*)arg
;
1086 struct kvm_device_attr attr
;
1090 case KVM_ARM_VCPU_INIT
: {
1091 struct kvm_vcpu_init init
;
1094 if (copy_from_user(&init
, argp
, sizeof(init
)))
1097 r
= kvm_arch_vcpu_ioctl_vcpu_init(vcpu
, &init
);
1100 case KVM_SET_ONE_REG
:
1101 case KVM_GET_ONE_REG
: {
1102 struct kvm_one_reg reg
;
1105 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1109 if (copy_from_user(®
, argp
, sizeof(reg
)))
1112 if (ioctl
== KVM_SET_ONE_REG
)
1113 r
= kvm_arm_set_reg(vcpu
, ®
);
1115 r
= kvm_arm_get_reg(vcpu
, ®
);
1118 case KVM_GET_REG_LIST
: {
1119 struct kvm_reg_list __user
*user_list
= argp
;
1120 struct kvm_reg_list reg_list
;
1124 if (unlikely(!kvm_vcpu_initialized(vcpu
)))
1128 if (copy_from_user(®_list
, user_list
, sizeof(reg_list
)))
1131 reg_list
.n
= kvm_arm_num_regs(vcpu
);
1132 if (copy_to_user(user_list
, ®_list
, sizeof(reg_list
)))
1137 r
= kvm_arm_copy_reg_indices(vcpu
, user_list
->reg
);
1140 case KVM_SET_DEVICE_ATTR
: {
1142 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1144 r
= kvm_arm_vcpu_set_attr(vcpu
, &attr
);
1147 case KVM_GET_DEVICE_ATTR
: {
1149 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1151 r
= kvm_arm_vcpu_get_attr(vcpu
, &attr
);
1154 case KVM_HAS_DEVICE_ATTR
: {
1156 if (copy_from_user(&attr
, argp
, sizeof(attr
)))
1158 r
= kvm_arm_vcpu_has_attr(vcpu
, &attr
);
1161 case KVM_GET_VCPU_EVENTS
: {
1162 struct kvm_vcpu_events events
;
1164 if (kvm_arm_vcpu_get_events(vcpu
, &events
))
1167 if (copy_to_user(argp
, &events
, sizeof(events
)))
1172 case KVM_SET_VCPU_EVENTS
: {
1173 struct kvm_vcpu_events events
;
1175 if (copy_from_user(&events
, argp
, sizeof(events
)))
1178 return kvm_arm_vcpu_set_events(vcpu
, &events
);
1188 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1189 * @kvm: kvm instance
1190 * @log: slot id and address to which we copy the log
1192 * Steps 1-4 below provide general overview of dirty page logging. See
1193 * kvm_get_dirty_log_protect() function description for additional details.
1195 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1196 * always flush the TLB (step 4) even if previous step failed and the dirty
1197 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1198 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1199 * writes will be marked dirty for next log read.
1201 * 1. Take a snapshot of the bit and clear it if needed.
1202 * 2. Write protect the corresponding page.
1203 * 3. Copy the snapshot to the userspace.
1204 * 4. Flush TLB's if needed.
1206 int kvm_vm_ioctl_get_dirty_log(struct kvm
*kvm
, struct kvm_dirty_log
*log
)
1211 mutex_lock(&kvm
->slots_lock
);
1213 r
= kvm_get_dirty_log_protect(kvm
, log
, &flush
);
1216 kvm_flush_remote_tlbs(kvm
);
1218 mutex_unlock(&kvm
->slots_lock
);
1222 int kvm_vm_ioctl_clear_dirty_log(struct kvm
*kvm
, struct kvm_clear_dirty_log
*log
)
1227 mutex_lock(&kvm
->slots_lock
);
1229 r
= kvm_clear_dirty_log_protect(kvm
, log
, &flush
);
1232 kvm_flush_remote_tlbs(kvm
);
1234 mutex_unlock(&kvm
->slots_lock
);
1238 static int kvm_vm_ioctl_set_device_addr(struct kvm
*kvm
,
1239 struct kvm_arm_device_addr
*dev_addr
)
1241 unsigned long dev_id
, type
;
1243 dev_id
= (dev_addr
->id
& KVM_ARM_DEVICE_ID_MASK
) >>
1244 KVM_ARM_DEVICE_ID_SHIFT
;
1245 type
= (dev_addr
->id
& KVM_ARM_DEVICE_TYPE_MASK
) >>
1246 KVM_ARM_DEVICE_TYPE_SHIFT
;
1249 case KVM_ARM_DEVICE_VGIC_V2
:
1252 return kvm_vgic_addr(kvm
, type
, &dev_addr
->addr
, true);
1258 long kvm_arch_vm_ioctl(struct file
*filp
,
1259 unsigned int ioctl
, unsigned long arg
)
1261 struct kvm
*kvm
= filp
->private_data
;
1262 void __user
*argp
= (void __user
*)arg
;
1265 case KVM_CREATE_IRQCHIP
: {
1269 mutex_lock(&kvm
->lock
);
1270 ret
= kvm_vgic_create(kvm
, KVM_DEV_TYPE_ARM_VGIC_V2
);
1271 mutex_unlock(&kvm
->lock
);
1274 case KVM_ARM_SET_DEVICE_ADDR
: {
1275 struct kvm_arm_device_addr dev_addr
;
1277 if (copy_from_user(&dev_addr
, argp
, sizeof(dev_addr
)))
1279 return kvm_vm_ioctl_set_device_addr(kvm
, &dev_addr
);
1281 case KVM_ARM_PREFERRED_TARGET
: {
1283 struct kvm_vcpu_init init
;
1285 err
= kvm_vcpu_preferred_target(&init
);
1289 if (copy_to_user(argp
, &init
, sizeof(init
)))
1299 static void cpu_init_hyp_mode(void *dummy
)
1301 phys_addr_t pgd_ptr
;
1302 unsigned long hyp_stack_ptr
;
1303 unsigned long stack_page
;
1304 unsigned long vector_ptr
;
1306 /* Switch from the HYP stub to our own HYP init vector */
1307 __hyp_set_vectors(kvm_get_idmap_vector());
1309 pgd_ptr
= kvm_mmu_get_httbr();
1310 stack_page
= __this_cpu_read(kvm_arm_hyp_stack_page
);
1311 hyp_stack_ptr
= stack_page
+ PAGE_SIZE
;
1312 vector_ptr
= (unsigned long)kvm_get_hyp_vector();
1314 __cpu_init_hyp_mode(pgd_ptr
, hyp_stack_ptr
, vector_ptr
);
1315 __cpu_init_stage2();
1318 static void cpu_hyp_reset(void)
1320 if (!is_kernel_in_hyp_mode())
1321 __hyp_reset_vectors();
1324 static void cpu_hyp_reinit(void)
1328 if (is_kernel_in_hyp_mode())
1329 kvm_timer_init_vhe();
1331 cpu_init_hyp_mode(NULL
);
1333 kvm_arm_init_debug();
1336 kvm_vgic_init_cpu_hardware();
1339 static void _kvm_arch_hardware_enable(void *discard
)
1341 if (!__this_cpu_read(kvm_arm_hardware_enabled
)) {
1343 __this_cpu_write(kvm_arm_hardware_enabled
, 1);
1347 int kvm_arch_hardware_enable(void)
1349 _kvm_arch_hardware_enable(NULL
);
1353 static void _kvm_arch_hardware_disable(void *discard
)
1355 if (__this_cpu_read(kvm_arm_hardware_enabled
)) {
1357 __this_cpu_write(kvm_arm_hardware_enabled
, 0);
1361 void kvm_arch_hardware_disable(void)
1363 _kvm_arch_hardware_disable(NULL
);
1366 #ifdef CONFIG_CPU_PM
1367 static int hyp_init_cpu_pm_notifier(struct notifier_block
*self
,
1372 * kvm_arm_hardware_enabled is left with its old value over
1373 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1378 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1380 * don't update kvm_arm_hardware_enabled here
1381 * so that the hardware will be re-enabled
1382 * when we resume. See below.
1387 case CPU_PM_ENTER_FAILED
:
1389 if (__this_cpu_read(kvm_arm_hardware_enabled
))
1390 /* The hardware was enabled before suspend. */
1400 static struct notifier_block hyp_init_cpu_pm_nb
= {
1401 .notifier_call
= hyp_init_cpu_pm_notifier
,
1404 static void __init
hyp_cpu_pm_init(void)
1406 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb
);
1408 static void __init
hyp_cpu_pm_exit(void)
1410 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb
);
1413 static inline void hyp_cpu_pm_init(void)
1416 static inline void hyp_cpu_pm_exit(void)
1421 static int init_common_resources(void)
1423 /* set size of VMID supported by CPU */
1424 kvm_vmid_bits
= kvm_get_vmid_bits();
1425 kvm_info("%d-bit VMID\n", kvm_vmid_bits
);
1427 kvm_set_ipa_limit();
1432 static int init_subsystems(void)
1437 * Enable hardware so that subsystem initialisation can access EL2.
1439 on_each_cpu(_kvm_arch_hardware_enable
, NULL
, 1);
1442 * Register CPU lower-power notifier
1447 * Init HYP view of VGIC
1449 err
= kvm_vgic_hyp_init();
1452 vgic_present
= true;
1456 vgic_present
= false;
1464 * Init HYP architected timer support
1466 err
= kvm_timer_hyp_init(vgic_present
);
1471 kvm_coproc_table_init();
1474 on_each_cpu(_kvm_arch_hardware_disable
, NULL
, 1);
1479 static void teardown_hyp_mode(void)
1484 for_each_possible_cpu(cpu
)
1485 free_page(per_cpu(kvm_arm_hyp_stack_page
, cpu
));
1490 * Inits Hyp-mode on all online CPUs
1492 static int init_hyp_mode(void)
1498 * Allocate Hyp PGD and setup Hyp identity mapping
1500 err
= kvm_mmu_init();
1505 * Allocate stack pages for Hypervisor-mode
1507 for_each_possible_cpu(cpu
) {
1508 unsigned long stack_page
;
1510 stack_page
= __get_free_page(GFP_KERNEL
);
1516 per_cpu(kvm_arm_hyp_stack_page
, cpu
) = stack_page
;
1520 * Map the Hyp-code called directly from the host
1522 err
= create_hyp_mappings(kvm_ksym_ref(__hyp_text_start
),
1523 kvm_ksym_ref(__hyp_text_end
), PAGE_HYP_EXEC
);
1525 kvm_err("Cannot map world-switch code\n");
1529 err
= create_hyp_mappings(kvm_ksym_ref(__start_rodata
),
1530 kvm_ksym_ref(__end_rodata
), PAGE_HYP_RO
);
1532 kvm_err("Cannot map rodata section\n");
1536 err
= create_hyp_mappings(kvm_ksym_ref(__bss_start
),
1537 kvm_ksym_ref(__bss_stop
), PAGE_HYP_RO
);
1539 kvm_err("Cannot map bss section\n");
1543 err
= kvm_map_vectors();
1545 kvm_err("Cannot map vectors\n");
1550 * Map the Hyp stack pages
1552 for_each_possible_cpu(cpu
) {
1553 char *stack_page
= (char *)per_cpu(kvm_arm_hyp_stack_page
, cpu
);
1554 err
= create_hyp_mappings(stack_page
, stack_page
+ PAGE_SIZE
,
1558 kvm_err("Cannot map hyp stack\n");
1563 for_each_possible_cpu(cpu
) {
1564 kvm_cpu_context_t
*cpu_ctxt
;
1566 cpu_ctxt
= per_cpu_ptr(&kvm_host_cpu_state
, cpu
);
1567 err
= create_hyp_mappings(cpu_ctxt
, cpu_ctxt
+ 1, PAGE_HYP
);
1570 kvm_err("Cannot map host CPU state: %d\n", err
);
1575 err
= hyp_map_aux_data();
1577 kvm_err("Cannot map host auxilary data: %d\n", err
);
1582 teardown_hyp_mode();
1583 kvm_err("error initializing Hyp mode: %d\n", err
);
1587 static void check_kvm_target_cpu(void *ret
)
1589 *(int *)ret
= kvm_target_cpu();
1592 struct kvm_vcpu
*kvm_mpidr_to_vcpu(struct kvm
*kvm
, unsigned long mpidr
)
1594 struct kvm_vcpu
*vcpu
;
1597 mpidr
&= MPIDR_HWID_BITMASK
;
1598 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
1599 if (mpidr
== kvm_vcpu_get_mpidr_aff(vcpu
))
1605 bool kvm_arch_has_irq_bypass(void)
1610 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer
*cons
,
1611 struct irq_bypass_producer
*prod
)
1613 struct kvm_kernel_irqfd
*irqfd
=
1614 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1616 return kvm_vgic_v4_set_forwarding(irqfd
->kvm
, prod
->irq
,
1619 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer
*cons
,
1620 struct irq_bypass_producer
*prod
)
1622 struct kvm_kernel_irqfd
*irqfd
=
1623 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1625 kvm_vgic_v4_unset_forwarding(irqfd
->kvm
, prod
->irq
,
1629 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer
*cons
)
1631 struct kvm_kernel_irqfd
*irqfd
=
1632 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1634 kvm_arm_halt_guest(irqfd
->kvm
);
1637 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer
*cons
)
1639 struct kvm_kernel_irqfd
*irqfd
=
1640 container_of(cons
, struct kvm_kernel_irqfd
, consumer
);
1642 kvm_arm_resume_guest(irqfd
->kvm
);
1646 * Initialize Hyp-mode and memory mappings on all CPUs.
1648 int kvm_arch_init(void *opaque
)
1654 if (!is_hyp_mode_available()) {
1655 kvm_info("HYP mode not available\n");
1659 if (!kvm_arch_check_sve_has_vhe()) {
1660 kvm_pr_unimpl("SVE system without VHE unsupported. Broken cpu?");
1664 for_each_online_cpu(cpu
) {
1665 smp_call_function_single(cpu
, check_kvm_target_cpu
, &ret
, 1);
1667 kvm_err("Error, CPU %d not supported!\n", cpu
);
1672 err
= init_common_resources();
1676 in_hyp_mode
= is_kernel_in_hyp_mode();
1679 err
= init_hyp_mode();
1684 err
= init_subsystems();
1689 kvm_info("VHE mode initialized successfully\n");
1691 kvm_info("Hyp mode initialized successfully\n");
1697 teardown_hyp_mode();
1702 /* NOP: Compiling as a module not supported */
1703 void kvm_arch_exit(void)
1705 kvm_perf_teardown();
1708 static int arm_init(void)
1710 int rc
= kvm_init(NULL
, sizeof(struct kvm_vcpu
), 0, THIS_MODULE
);
1714 module_init(arm_init
);