IMO the big problem is that it's very hard (and AFAIK no way has even be proposed) to restrict from the outside which syscall-alikes are allowed via io_uring.

IMO the big problem is that it's very hard (and AFAIK no way has even be proposed) to restrict from the outside which syscall-alikes are allowed via io_uring. This is trivial with regular syscalls. That's important if you ever co-host applications with different security domains.