Re: [syzbot ci] Re: KVM: x86: Unconditionally recompute CR8 intercept on PPR update

On Thu, Jun 11, 2026, syzbot ci wrote:
> syzbot ci has tested the following series
> 
> [v2] KVM: x86: Unconditionally recompute CR8 intercept on PPR update
> https://lore.kernel.org/all/[email protected]
> * [PATCH v2] KVM: x86: Unconditionally recompute CR8 intercept on PPR update
> 
> and found the following issue:
> WARNING in vmx_update_cr8_intercept

...

> ------------[ cut here ]------------
> debug_locks && !(lock_is_held(&(&vcpu->mutex)->dep_map) || !refcount_read(&vcpu->kvm->users_count))
> WARNING: arch/x86/kvm/vmx/nested.h:61 at get_vmcs12 arch/x86/kvm/vmx/nested.h:60 [inline], CPU#0: syz.2.19/5879
> WARNING: arch/x86/kvm/vmx/nested.h:61 at vmx_update_cr8_intercept+0x3de/0x4e0 arch/x86/kvm/vmx/vmx.c:6879, CPU#0: syz.2.19/5879
> Modules linked in:
> CPU: 0 UID: 0 PID: 5879 Comm: syz.2.19 Not tainted syzkaller #0 PREEMPT(full) 
> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.2-debian-1.16.2-1 04/01/2014
> RIP: 0010:get_vmcs12 arch/x86/kvm/vmx/nested.h:60 [inline]
> RIP: 0010:vmx_update_cr8_intercept+0x3de/0x4e0 arch/x86/kvm/vmx/vmx.c:6879
>  apic_update_ppr arch/x86/kvm/lapic.c:984 [inline]
>  kvm_lapic_reset+0x1c24/0x2980 arch/x86/kvm/lapic.c:3023
>  kvm_vcpu_reset+0x44c/0x1bf0 arch/x86/kvm/x86.c:12986
>  kvm_arch_vcpu_create+0x746/0x8b0 arch/x86/kvm/x86.c:12847
>  kvm_vm_ioctl_create_vcpu+0x428/0x930 virt/kvm/kvm_main.c:4201
>  kvm_vm_ioctl+0x893/0xd50 virt/kvm/kvm_main.c:5159
>  vfs_ioctl fs/ioctl.c:51 [inline]
>  __do_sys_ioctl fs/ioctl.c:597 [inline]
>  __se_sys_ioctl+0xfc/0x170 fs/ioctl.c:583
>  do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
>  do_syscall_64+0x174/0x580 arch/x86/entry/syscall_64.c:94
>  entry_SYSCALL_64_after_hwframe+0x77/0x7f

This is "fine", the assertion just wants to make sure KVM isn't access vmcs12
without holding vcpu->mutex, otherwise any queries are inherently unstable.
It's just that vCPU creation runs without taking vcpu->mutex, because the vCPU
is otherwise unreachable.

I'm pretty sure we can squash the WARN by grabbing vmcs12 if and only if the vCPU
is actually in guest mode.

diff --git arch/x86/kvm/vmx/vmx.c arch/x86/kvm/vmx/vmx.c
index c548f22375ad..332fbcd924f2 100644
--- arch/x86/kvm/vmx/vmx.c
+++ arch/x86/kvm/vmx/vmx.c
@@ -6876,11 +6876,10 @@ int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
 
 void vmx_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
 {
-       struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
        int tpr_threshold;
 
        if (is_guest_mode(vcpu) &&
-               nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
+           nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_TPR_SHADOW))
                return;
 
        guard(vmx_vmcs01)(vcpu);


Longer term, I'll work on figuring out how to handle this in get_vmcs12(), because
to_hv_vcpu() has the solve the same fundamental problem:

https://lore.kernel.org/all/[email protected]