Re: [PATCH bpf-next v2 3/4] bpf: replace min/max fields with struct cnum{32,64}

[Eduard Zingerman <eddyz87@gmail.com>]

On Fri, 2026-04-24 at 09:12 +0000, bot+bpf-ci@kernel.org wrote:
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index b91d2789e7b9..34900247efba 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> 
> [ ... ]
> 
> > @@ -2031,340 +2029,17 @@ static void __update_reg_bounds(struct bpf_reg_state *reg)
> >  /* Uses signed min/max values to inform unsigned, and vice-versa */
> >  static void deduce_bounds_32_from_64(struct bpf_reg_state *reg)
> >  {
> > -	/* If upper 32 bits of u64/s64 range don't change, we can use lower 32
> > -	 * bits to improve our u32/s32 boundaries.
> 
> [ ... ]
> 
> > -	}
> > +	cnum32_intersect_with(&reg->r32, cnum32_from_cnum64(reg->r64));
> >  }
> 
> The comment on the function still says "Uses signed min/max values to
> inform unsigned, and vice-versa" but the implementation is now just a
> single call to cnum32_intersect_with() that projects a 64-bit circular
> number onto the 32-bit subregister. With the cnum abstraction, signed
> and unsigned views are derived on-demand from a unified representation,
> so the cross-domain signed<->unsigned deduction described in the comment
> no longer happens. Should this comment be updated to describe the actual
> operation (e.g., "Project 64-bit range onto 32-bit subregister range")?

The comment has to be updated.