Re: [PATCH bpf-next v3 5/9] selftests/bpf: Add kfunc __arena and __arena_nullable argument tests

[Eduard Zingerman <eddyz87@gmail.com>]

On Tue, 2026-08-04 at 13:33 -0700, Eduard Zingerman wrote:
> On Tue, 2026-08-04 at 22:28 +0200, Kumar Kartikeya Dwivedi wrote:
> > On Tue Aug 4, 2026 at 10:26 PM CEST, Eduard Zingerman wrote:
> > > On Tue, 2026-08-04 at 13:24 -0700, Eduard Zingerman wrote:
> > > > On Tue, 2026-08-04 at 22:14 +0200, Kumar Kartikeya Dwivedi wrote:
> > > > 
> > > > ...
> > > > 
> > > > > > Which leads to several conclusions:
> > > > > > - We should reach a consensus regarding KF_ARENA* usage, as these
> > > > > >   kfuncs would clearly benefit from having __arena annotations on
> > > > > >   parameters (on the BPF side, on the host side these should be
> > > > > >   declared as plain pointers).
> > > > > 
> > > > > Yes, but even if we do, for the time begin, we won't have support for
> > > > > automatically emitting it until resolve_btfids changes are done. That said, I
> > > > > can manually add the annotation for now.
> > > > 
> > > > We can fast-track resolve_btfids.
> > > > 
> > > > > > - In patch #1, it appears that there is no need to allow PTR_TO_ARENA
> > > > > >   for parameters passed to such a function. At-least I don't see a C
> > > > > >   code not involving manual casts that would require such support.
> > > > > 
> > > > > This one I don't understand. Why wouldn't we get PTR_TO_ARENA typed paramters as
> > > > > arguments?
> > > > 
> > > > `(u64 *)val` is translated as an address space cast instruction.
> > > > This instruction yields PTR_TO_ARENA. Except for explicit casts,
> > > > how else can you pass PTR_TO_ARENA to such a function?
> > > 
> > > For example, `foo->ptr` yields a scalar from the verifier point of view.
> > > 
> > 
> > Sure, but I don't think we should reject the case where someone does explicit
> > cast. See e.g. the global function __arena tag, which also accepts both cases.
> 
> PTR_TO_ARENA is already a valid kernel space pointer, it does not need
> additional conversion.

Nah, that's not true. It is still a 32-bit value.