Re: [PATCH 0/9] Kernel API Specification Framework

[Jakub Kicinski <kuba@kernel.org>]

On Mon, 16 Mar 2026 08:05:53 +0100 Dmitry Vyukov wrote:
> On Sat, 14 Mar 2026 at 19:18, Jakub Kicinski <kuba@kernel.org> wrote:
> > On Fri, 13 Mar 2026 11:09:10 -0400 Sasha Levin wrote:  
> > > This enables static analysis tools to verify userspace API usage at compile
> > > time, test generation based on formal specifications, consistent error handling
> > > validation, automated documentation generation, and formal verification of
> > > kernel interfaces.  
> >
> > Could you give some examples? We have machine readable descriptions for
> > Netlink interfaces, we approached syzbot folks and they did not really
> > seem to care for those.  
> 
> I think our reasoning wrt syzkaller was that not all interfaces in all
> relevant kernels are described with netlink yml descriptions, so we
> need to continue using the extraction of interfaces from the source
> code. And if we have that code, then using yml as an additional data
> source only adds code/complexity. Additionally, we may extract some
> extra constraints/info from code that are not present in yml.
> 
> Realistically system call descriptions may have the same problem for
> us at this point, since we extract lots of info from the source code
> already:
> https://raw.githubusercontent.com/google/syzkaller/refs/heads/master/sys/linux/auto.txt

yup! we haven't tried to make the yml spec super useful to syzbot 
to be fair. I'm just flagging that example because in our case we
quickly went from "this will obviously be useful to syzbot" to
"although we could, it may not be super practical"

> (and LLMs obviously can allow us to extract more)

Didn't even think of that. LLMs should make short work of this sort of
extraction of information from source code..