Message-ID: <20090419085914.GA25898@elte.hu>
Date:	Sun, 19 Apr 2009 10:59:14 +0200
From:	Ingo Molnar <mingo@...e.hu>
To:	Sam Ravnborg <sam@...nborg.org>, "H. Peter Anvin" <hpa@...or.com>,
	Thomas Gleixner <tglx@...utronix.de>,
	Andrew Morton <akpm@...ux-foundation.org>
Cc:	Kyle Moffett <kyle@...fetthome.net>,
	linux-kbuild <linux-kbuild@...r.kernel.org>,
	linux arch <linux-arch@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Thomas Gleixner <tglx@...utronix.de>,
	Ingo Molnar <mingo@...hat.com>,
	"H. Peter Anvin" <hpa@...or.com>,
	"David S. Miller" <davem@...emloft.net>,
	Russell King <rmk+lkml@....linux.org.uk>
Subject: Re: kbuild - introduce support for subdir-ccflags-y


* Sam Ravnborg <sam@...nborg.org> wrote:

> On Sat, Apr 18, 2009 at 11:17:44PM -0400, Kyle Moffett wrote:
> > On Sat, Apr 18, 2009 at 2:09 PM, Russell King <rmk+lkml@....linux.org.uk> wrote:
> > > On Sat, Apr 18, 2009 at 05:07:50PM +0200, Ingo Molnar wrote:
> > >> * Russell King <rmk+lkml@....linux.org.uk> wrote:
> > >> > On Sat, Apr 18, 2009 at 02:51:59PM +0200, Sam Ravnborg wrote:
> > >> > > The typical use cases are an architecture or a subsystem that
> > >> > > decide to cover all files with -Werror.
> > >> > > Today alpha, mips and sparc uses -Werror in almost all their
> > >> > > Makefile - with subdir-ccflag-y it is now simpler to do so
> > >> > > as only the top-level directories needs to be covered.
> > >> >
> > >> > Hmm, this won't make sense for ARM.  We have things like #warning
> > >> > and deprecated functions in machine specific headers, and adding
> > >> > -Werror to the whole of arch/arm/ will result in these causing
> > >> > builds to fail.
> > >>
> > >> This is optional - if you dont want it, you dont set it.
> > >
> > > Please read _all_ of my mail, particularly the bit where it talks about
> > > it being useful for a certain subset.
> > 
> > It's my impression that on x86 it's a config option whether or not to
> > build with -Werror.  You could do the exact same thing with an
> > internal inverted-logic CONFIG_ARM_ALLOW_WARNINGS option and make all
> > the boards triggering warnings "select ARM_ALLOW_WARNINGS".  Then have
> > a user-visible config option "ARM_USER_ALLOW_WARNINGS" which also
> > selects the internal option.  That adds some additional kconfig-level
> > documentation on which subarch combos need some love.
> 
> Correct - but Russell's original comment was that the subdir-* 
> feature was of no use for arm. And this is correct with the 
> current way the individual directories are specified.
> 
> If you compare arch/x86/kernel/ and arch/arm/kernel/ you will see 
> that x86 has 3 subdirectories where arm has none.
> 
> Another measure: Arm has 12 Makefiles outside mach* and plat* 
> where x86 has 24 Makefiles.
> 
> Ao unless Russell decide to take the patch that refactor all the 
> core-* stuff out in a Kbuild file arm has no use of 
> subdir-ccflags-y and can use the already existing ccflags-y if 
> they decide to cover most of arm with -Werror.

Yes. This is an optional facility entirely - if some arch does
not need or does not want it, it does not have to care.

Sparc is the example to follow here: it has had unconditional 
-Werror builds (for its architecture files) for years. It has
done so without any explicit kbuild help beyond a single-line
ccflags-y := -Werror rule. Sparc has 7 makefiles and it has a 
well-tested compiler environment.

We cannot do the same on x86 straight away: it's being built
on a very wide variety of compiler versions, by a very wide
range of people: more than 95% of our Linux kernel testers are
on x86.

Some of those compiler versions emit bogus warnings that cannot
be worked around in the kernel in any sane way - and a forced
-Werror build would thus become a (needless) show-stopper there.
It will also take some time to flush out any remaining warnings.

All in one, we want arch/x86 to become a -Werror-clean architecture 
and we want all the advantages of that, but we didnt want to do any 
artificial 'break the world' flag day.

So we are doing a hybrid approach instead, we are adding a 
default-off .config option, and we are doing it all smoothly and 
gradually.

The reason we asked Sam whether he can think of any kbuild help 
(again, which is strictly optional) was that we didnt want to 
propagate a 5-6 lines hybrid Makefile rule into 24 makefiles.

We could have done this all via a special makefile rule entirely
in arch/x86/ that is included in every secondary makefile - but
we also wanted to give other architectures an optional, librarized
kbuild facility to do the same, if they so choose.

Thanks,

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists