linux-kernel - Re: [PATCH] mtd: spi-nor: don't build Cadence QuadSPI on non-ARM

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20160720032501.GA35298@google.com>
Date:	Tue, 19 Jul 2016 20:25:01 -0700
From:	Brian Norris <computersforpeace@...il.com>
To:	Marek Vasut <marex@...x.de>
Cc:	Stefan Roese <sr@...x.de>,
	kbuild test robot <fengguang.wu@...el.com>, x86@...nel.org,
	linux-kernel@...r.kernel.org, linux-mtd@...ts.infradead.org,
	kbuild-all@...org, Graham Moore <grmoore@...nsource.altera.com>
Subject: Re: [PATCH] mtd: spi-nor: don't build Cadence QuadSPI on non-ARM

On Wed, Jul 20, 2016 at 04:58:08AM +0200, Marek Vasut wrote:
> On 07/20/2016 04:50 AM, Brian Norris wrote:
> > On Wed, Jul 20, 2016 at 03:50:27AM +0200, Marek Vasut wrote:
> >> On 07/19/2016 10:05 PM, Brian Norris wrote:
> >>> On Tue, Jul 19, 2016 at 08:03:00AM +0200, Stefan Roese wrote:
> >>>> On 18.07.2016 22:20, Brian Norris wrote:
> >>>>> Hmm, does x86 not define readsl()/writesl()? I can never tell what
> >>>>> accessors are supposed to be "standard" across architectures.
> >>>>>
> >>>>> Either we need to drop the COMPILE_TEST or maybe make it (!X86 &&
> >>>>> COMPILE_TEST).
> >>>>
> >>>> iowrite32_rep() etc should work for x86 as well.
> >>>
> >>> Looks like it might. I'm not sure the original submitter can retest
> >>> right now (travel), so I'd probably rather just take the easy fix for
> >>> now, and we can widen to COMPILE_TEST later if desired.
> >>
> >> Isn't there a generic readsl() and writesl() implementation in
> >> include/asm-generic/io.h ?
> > 
> > Yes, but somehow x86 has managed to avoid that. I guess it's optional
> > for arch/<foo>/include/asm/io.h to include <asm-generic/io.h>? At any
> > rate, I double-checked myself by adding '#error "blah"' to
> > include/asm-generic/io.h, and x86 still seemed to build fine (at least
> > for the modules I was checking, like cadence-quadspi.o).
> 
> Yep, I just checked the same and it's not included from
> arch/x86/include/asm/io.h for whatever reason. Maybe this needs to be
> fixed on x86 level?

Maybe. That's why I added the x86 maintainers. Maybe they'd respond
better^Wmore loudly if I just sent a patch to do that :)

But seriously, doing the above really breaks things, even if I stick the
include at the end of asm/io.h. There's plenty of stuff that the
asm-generic version includes based on #ifndef some_accessor, except x86
uses a static inline for their definition. So it seems it's not trivial
to get an architecture to fall back gracefully to asm-generic; you have
to put in some work. It also may not be all that desirable to have some
allegedly generic version generate something that may not be safe on a
given architecture (and I don't purport to understand the x86 memory
model).

Additionally, it looks like asm-generic/io.h is actually only included
in 14 of 33 arch'es, so it seems like that's really not a designated
goal. It does make it awfully difficult to figure out what I/O accessors
are *actually* portable though...

Brian