[PATCH v3 04/13] x86: Handle KCOV __init vs inline mismatches

[Ard Biesheuvel]

On Sat, 19 Jul 2025 at 08:51, Kees Cook <kees at kernel.org> wrote:
>
> On Fri, Jul 18, 2025 at 11:36:32AM +0300, Mike Rapoport wrote:
> > Hi Kees,
> >
> > On Thu, Jul 17, 2025 at 04:25:09PM -0700, Kees Cook wrote:
> > > When KCOV is enabled all functions get instrumented, unless the
> > > __no_sanitize_coverage attribute is used. To prepare for
> > > __no_sanitize_coverage being applied to __init functions, we have to
> > > handle differences in how GCC's inline optimizations get resolved. For
> > > x86 this means forcing several functions to be inline with
> > > __always_inline.
> > >
> > > Signed-off-by: Kees Cook <kees at kernel.org>
> >
> > ...
> >
> > > diff --git a/include/linux/memblock.h b/include/linux/memblock.h
> > > index bb19a2534224..b96746376e17 100644
> > > --- a/include/linux/memblock.h
> > > +++ b/include/linux/memblock.h
> > > @@ -463,7 +463,7 @@ static inline void *memblock_alloc_raw(phys_addr_t size,
> > >                                       NUMA_NO_NODE);
> > >  }
> > >
> > > -static inline void *memblock_alloc_from(phys_addr_t size,
> > > +static __always_inline void *memblock_alloc_from(phys_addr_t size,
> > >                                             phys_addr_t align,
> > >                                             phys_addr_t min_addr)
> >
> > I'm curious why from all memblock_alloc* wrappers this is the only one that
> > needs to be __always_inline?
>
> Thread-merge[1], adding Will Deacon, who was kind of asking the same
> question.
>
> Based on what I can tell, GCC has kind of fragile inlining logic, in the
> sense that it can change whether or not it inlines something based on
> optimizations. It looks like the kcov instrumentation being added (or in
> this case, removed) from a function changes the optimization results,
> and some functions marked "inline" are _not_ inlined. In that case, we end up
> with __init code calling a function not marked __init, and we get the
> build warnings I'm trying to eliminate.
>
> So, to Will's comment, yes, the problem is somewhat fragile (though
> using either __always_inline or __init will deterministically solve it).
> We've tripped over this before with GCC and the solution has usually
> been to just use __always_inline and move on.
>

Given that 'inline' is already a macro in the kernel, could we just
add __attribute__((__always_inline__)) to it when KCOV is enabled?