[PATCH v2 04/16] arm64: Downgrade flush_icache_range to invalidate

[Mark Rutland]

On Tue, May 18, 2021 at 06:02:32PM +0200, Ard Biesheuvel wrote:
> On Tue, 18 May 2021 at 17:53, Mark Rutland <mark.rutland at arm.com> wrote:
> >
> > On Mon, May 17, 2021 at 08:51:12AM +0100, Fuad Tabba wrote:
> > > Since __flush_dcache_area is called right before,
> > > invalidate_icache_range is sufficient in this case.
> > >
> > > Rewrite the comment to better explain the rationale behind the
> > > cache maintenance operations used here.
> > >
> > > No functional change intended.
> > > Possible performance impact due to invalidating only the icache
> > > rather than invalidating and cleaning both caches.
> > >
> > > Reported-by: Catalin Marinas <catalin.marinas at arm.com>
> > > Reported-by: Will Deacon <will at kernel.org>
> > > Link: https://lore.kernel.org/linux-arch/20200511110014.lb9PEahJ4hVOYrbwIb_qUHXyNy9KQzNFdb_I3YlzY6A@z/
> > > Signed-off-by: Fuad Tabba <tabba at google.com>
> > > ---
> > >  arch/arm64/kernel/machine_kexec.c | 10 +++++++---
> > >  1 file changed, 7 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/arch/arm64/kernel/machine_kexec.c b/arch/arm64/kernel/machine_kexec.c
> > > index 90a335c74442..ecd8915e02e1 100644
> > > --- a/arch/arm64/kernel/machine_kexec.c
> > > +++ b/arch/arm64/kernel/machine_kexec.c
> > > @@ -68,10 +68,14 @@ int machine_kexec_post_load(struct kimage *kimage)
> > >       kimage->arch.kern_reloc = __pa(reloc_code);
> > >       kexec_image_info(kimage);
> > >
> > > -     /* Flush the reloc_code in preparation for its execution. */
> > > +     /*
> > > +      * For execution with the MMU off and I-cache on, reloc_code needs to be
> > > +      * cleaned to the PoC and invalidated from the I-cache.
> > > +      */
> >
> > Minor nit, but the I-cache is *always* on (SCTLR.I affects the
> > attributes used for fetches into the I-caches), so it would be slightly
> > better to drop the "and I-cache on" words.
> 
> This may be true, but it may not be obvious to someone reading the
> 'MMU off' part of the comment. Bottom line is that we will be running
> in a mode where we may hit in the I-cache, so it needs to be
> invalidated. If we miss in the I-cache, we should fetch from the PoC,
> hence the D-cache clean.

No disagreement there. I literally meant dropping the words "and I-cache
on", leaving the whole comment as:

	/*
	 * For execution with the MMU off, reloc_code needs to be
	 * cleaned to the PoC and invalidated from the I-cache.
	 */

Thanks,
Mark.