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

[Ard Biesheuvel <ardb@kernel.org>]

On Tue, 22 Jul 2025 at 06:49, Kees Cook <kees@kernel.org> wrote:
>
> On Mon, Jul 21, 2025 at 01:14:36PM -0700, Kees Cook wrote:
> > On Mon, Jul 21, 2025 at 01:47:55PM +0100, Will Deacon wrote:
> > > On Sun, Jul 20, 2025 at 04:10:01PM +1000, Ard Biesheuvel wrote:
> > > > On Sat, 19 Jul 2025 at 08:51, Kees Cook <kees@kernel.org> wrote:
> > > > > On Fri, Jul 18, 2025 at 11:36:32AM +0300, Mike Rapoport wrote:
> > > > > > 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@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.
> > >
> > > Got it, thanks for the explanation!
> > >
> > > > > 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?
> > >
> > > That sounds like a more robust approach and, by the sounds of it, we
> > > could predicate it on GCC too. That would also provide a neat place for
> > > a comment describing the problem.
> > >
> > > Kees, would that work for you?
> >
> > That seems like an extremely large hammer for this problem, IMO. It
> > feels like it could cause new strange corner cases. I'd much prefer the
> > small fixes I've currently got since it keeps it focused. KCOV is
> > already enabled for "allmodconfig", so any new instances would be found
> > very quickly, etc. (And GCC's fragility in this regard has already been
> > exposed to these cases -- it's just that I changed one of the
> > combinations of __init vs inline vs instrumentation.
> >
> > I could give it a try, if you really prefer the big hammer approach...
>
> I gave it a try -- it fails spectacularly. ;) Let's stick to my small
> fixes instead?
>

Fair enough :-)