Re: [syzbot ci] Re: KVM: x86/hyperv: Fix racy usage of vcpu->arch.hyperv

[Sean Christopherson]

On Thu, Apr 23, 2026, syzbot ci wrote:
> syzbot ci has tested the following series
> ***
> 
> WARNING in kvm_hv_vcpu_uninit
> 
> tree:      linux-next
> URL:       https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next
> base:      85f871f6ba46f20d7fbc0b016b4db648c33220dd
> arch:      amd64
> compiler:  Debian clang version 21.1.8 (++20251221033036+2078da43e25a-1~exp1~20251221153213.50), Debian LLD 21.1.8
> config:    https://ci.syzbot.org/builds/4f8c36c7-e865-4f65-bb81-e22962e8a1e0/config
> syz repro: https://ci.syzbot.org/findings/60fac5d8-1296-4ec7-a8e3-40f561d007e9/syz_repro
> 
> ------------[ cut here ]------------
> debug_locks && !(lock_is_held(&(&vcpu->mutex)->dep_map) || !refcount_read(&vcpu->kvm->users_count))
> WARNING: arch/x86/kvm/hyperv.h:79 at to_hv_vcpu arch/x86/kvm/hyperv.h:78 [inline], CPU#1: syz.2.19/5974
> WARNING: arch/x86/kvm/hyperv.h:79 at kvm_hv_vcpu_uninit+0x163/0x1b0 arch/x86/kvm/hyperv.c:906, CPU#1: syz.2.19/5974
> Modules linked in:
> CPU: 1 UID: 0 PID: 5974 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:to_hv_vcpu arch/x86/kvm/hyperv.h:78 [inline]
> RIP: 0010:kvm_hv_vcpu_uninit+0x163/0x1b0 arch/x86/kvm/hyperv.c:906
> Call Trace:
>  <TASK>
>  kvm_arch_vcpu_destroy+0x1a9/0x380 arch/x86/kvm/x86.c:12963
>  kvm_vm_ioctl_create_vcpu+0x69a/0x930 virt/kvm/kvm_main.c:4269
>  kvm_vm_ioctl+0x893/0xd50 virt/kvm/kvm_main.c:5168
>  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+0x14d/0xf80 arch/x86/entry/syscall_64.c:94
>  entry_SYSCALL_64_after_hwframe+0x77/0x7f

Argh, what a pain.  It's effectively the same issue that I fudged around in
kvm_hv_set_cpuid(): KVM queries HyperV state during vCPU creation, before taking
vcpu->mutex makes any sense.

One thought I had was to initialize vcpu_idx to -1, so that to_hv_vcpu() could
detect that the vCPU isn't yet visible to others.  Arguably that would be also
nice-to-have as it would harden against consuming vcpu->vcpu_idx before it's
fully initialized.  As-is, goofs would result in KVM thinking its vCPU0.

diff --git virt/kvm/kvm_main.c virt/kvm/kvm_main.c
index 7fcb92c69dc8..35e92cfb2a45 100644
--- virt/kvm/kvm_main.c
+++ virt/kvm/kvm_main.c
@@ -4198,6 +4198,8 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, unsigned long id)
                goto vcpu_decrement;
        }
 
+       vcpu->vcpu_idx = -1;
+
        BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
        page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
        if (!page) {