In the Linux kernel, the following vulnerability has been resolved:

KVM: Don't null dereference ops->destroy

A KVM device cleanup happens in either of two callbacks:
1) destroy() which is called when the VM is being destroyed;
2) release() which is called when a device fd is closed.

Most KVM devices use 1) but Book3s's interrupt controller KVM devices
(XICS, XIVE, XIVE-native) use 2) as they need to close and reopen during
the machine execution. The error handling in kvm_ioctl_create_device()
assumes destroy() is always defined which leads to NULL dereference as
discovered by Syzkaller.

This adds a checks for destroy!=NULL and adds a missing release().

This is not changing kvm_destroy_devices() as devices with defined
release() should have been removed from the KVM devices list by then.

https://git.kernel.org/stable/c/170465715a60cbb7876e6b961b21bd3225469da8

https://git.kernel.org/stable/c/3616776bc51cd3262bb1be60cc01c72e0a1959cf

https://git.kernel.org/stable/c/d4a5a79b780891c5cbdfdc6124d46fdf8d13dba1

https://git.kernel.org/stable/c/e8bc2427018826e02add7b0ed0fc625a60390ae5

https://git.kernel.org/stable/c/e91665fbbf3ccb268b268a7d71a6513538d813ac