Re: Build broken on s390 and ia64 [was: Linux 2.6.32.58]

[Willy Tarreau <w@1wt.eu>]

Hi Ben,

On Fri, Mar 09, 2012 at 05:29:13AM +0000, Ben Hutchings wrote:
> On Mon, 2012-03-05 at 23:33 +0100, Willy Tarreau wrote:
> > Hi,
> > 
> > On Mon, Mar 05, 2012 at 01:48:24PM +0100, Jiri Slaby wrote:
> > > >>> Linus Torvalds (2): Fix autofs compile without CONFIG_COMPAT
> > > >>
> > > >> But who defines is_compat_task *with* CONFIG_COMPAT on ia64?
> > > >>
> > > >> fs/autofs4/inode.c: In function 'autofs4_fill_super':
> > > >> fs/autofs4/inode.c:345: error: implicit declaration of function
> > > >> 'is_compat_task'
> > > > 
> > > > The ia64 compat code got entirely removed, since it was broken:
> > > > 
> > > > 32974ad4907cdde6c9de612cd1b2ee0568fb9409 "[IA64] Remove COMPAT_IA32 support"
> > > 
> > > Yes, but that is even in 2.6.34. So the fix for autofs is incomplete in
> > > .32 as it breaks build on configs which used to work.
> > 
> > So in the end, does anybody have an idea what is missing from this patch ?
> > I'm not sure that reverting the autofs fix is a right solution either :-/
> 
> Either cherry-pick commit 32974ad4907cdde6c9de612cd1b2ee0568fb9409 or
> use this reduced version.

Thanks but now I have a doubt, maybe Jiri and Tony can help. In the
commit above, Tony says that CONFIG_COMPAT has been broken for a long
time (2008) and nobody apparently uses it. Still, Jiri got a build
failure because he used CONFIG_COMPAT. So either there are some valid
uses and we should not remove the feature that late in a stable branch,
or Jiri only encountered it upon a make allyesconfig then it might
be safe to remove it.

You see, I wouldn't want to have to issue .60 with a revert of this
patch because someone complains about CONFIG_COMPAT disappearing on
IA64.

I'm fine with either patch (#define is_compat_task 0 or remove COMPAT)
but I want to ensure we're not breaking working setups which are currently
stuck to .57 due to the recent breakage.

Regards,
Willy