Re: [syzbot] upstream boot error: BUG: unable to handle kernel NULL pointer dereference in __dabt_svc

From: Dmitry Vyukov
Date: Wed Apr 26 2023 - 06:43:41 EST


On Wed, 26 Apr 2023 at 12:30, Miguel Ojeda
<miguel.ojeda.sandonis@xxxxxxxxx> wrote:
> > In which of the dozens of kernel testing systems? ;)
> > And also in heads of thousands of kernel developers and users?
> > All of them use get_maintainer.pl.
>
> I am aware, but `get_maintainer.pl` is fine as it is -- we still want
> to know about things that touch things that mention Rust in general,
> so that we can possibly be helpful to others, especially early on.
>
> However, if a bot is testing the kernel with Rust actually disabled at
> runtime, what I am saying is that the chance that it has something to
> do with Rust is quite low, especially if matched via `K:` rather than
> `F:`. Thus my request.
>
> Now, it could be nice to have some logic like that in
> `get_maintainer.pl` encoded for all bots to filter things out based on
> the kernel config and the type of match; but otherwise, yes, the bots
> would need to add the logic.
>
> Cc'ing Joe in case this is already possible in `get_maintainer.pl` or
> whether there could be a better approach.

I understand your intentions and they make sense.
But adding this logic to syzbot won't help thousands of users of
get_maintainer.pl and dozens of other testing systems. There also will
be a bit of get_maintainer.pl inside of syzbot code, so now all kernel
developers will need to be aware of it and also submit changes to
syzbot when they want to change maintainers logic.

I think this also equally applies to all other users of K:.
And a number of them had similar complaints re how K; works.

I am thinking if K: should actually apply just to patches and be
ignored for source files?
If there are files that belong to "rust" (or "bpf" or any other user
of K:), then I think these should be just listed explicitly in the
subsystem (that should be a limited set of files that can be
enumerated with wildcards).
It's also reasonable to apply K: to patches.
But if a random source file happened to mention "rust" somewhere once,
I am not sure you want to be CCed on all issues in that file.
Does it sound reasonable?