Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Jamie Lenehan]

On Thu, Sep 28, 2006 at 01:14:05PM +0200, "pfalcon commit" wrote:
> glibc 2.4: Apply recently added patches only fo x86.
> * Fixes breakage on ARM.
> * Don't bump PR (previous commit neither did), to save long rebuilds.
> Few people affected should use "bitbake glibc -c rebuild".
> * Fixes #1443.
> * Approved by Michael 'Mickey' Lauer <mickey at tm.informatik.uni-frankfurt de>.

I've tested here and those two patches break glibc 2.4 for my main
targets which are sh4 (titan), i486 (wrap) and i586 (epia), building
on an x86_64 host. The breakage basically exactly the same as in
#1443.

I'm not sure what "x86" problem is that they are meant to fix (look
like they are just wiring up syscalls to use in place of glibc equivs
for the *at calls), but at least for me they actually break things on
i486 and i586.

Looks to me like the need more thinking/testing before being included
for anything.

-- 
 Jamie Lenehan <lenehan@twibble.org>

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Michael 'Mickey' Lauer]

Am Donnerstag, den 28.09.2006, 23:27 +1000 schrieb Jamie Lenehan:
> On Thu, Sep 28, 2006 at 01:14:05PM +0200, "pfalcon commit" wrote:
> > glibc 2.4: Apply recently added patches only fo x86.
> > * Fixes breakage on ARM.
> > * Don't bump PR (previous commit neither did), to save long rebuilds.
> > Few people affected should use "bitbake glibc -c rebuild".
> > * Fixes #1443.
> > * Approved by Michael 'Mickey' Lauer <mickey at tm.informatik.uni-frankfurt de>.
> 
> I've tested here and those two patches break glibc 2.4 for my main
> targets which are sh4 (titan), i486 (wrap) and i586 (epia), building
> on an x86_64 host. The breakage basically exactly the same as in
> #1443.
> 
> I'm not sure what "x86" problem is that they are meant to fix (look
> like they are just wiring up syscalls to use in place of glibc equivs
> for the *at calls), but at least for me they actually break things on
> i486 and i586.

1. patch fixes problem:
| ../sysdeps/unix/sysv/linux/openat.c: In function '__OPENAT':
| ../sysdeps/unix/sysv/linux/openat.c:156: error: 'va_start' used in
function with fixed args

This is glibc bug #3130, see
http://sourceware.com/7bugzilla/show_bug.cgi?id=03170

2. patch fixes problem:

| /local/pkg/oe/x86/tmp/work/i486-angstrom-linux/glibc-intermediate-2.4-r10/build-i486-angstrom-linux/libc_pic.os: In function `fchownat':
| (.text+0x78cc6): undefined reference to `__atfct_seterrno'
| collect2: ld returned 1 exit status

-- 
Regards,

Michael 'Mickey' Lauer | FreeLancer | http://www.Vanille-Media.de

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Jamie Lenehan]

On Thu, Sep 28, 2006 at 05:43:55PM +0200, Michael 'Mickey' Lauer wrote:
> Am Donnerstag, den 28.09.2006, 23:27 +1000 schrieb Jamie Lenehan:
> > On Thu, Sep 28, 2006 at 01:14:05PM +0200, "pfalcon commit" wrote:
> > > glibc 2.4: Apply recently added patches only fo x86.
> > > * Fixes breakage on ARM.
> > > * Don't bump PR (previous commit neither did), to save long rebuilds.
> > > Few people affected should use "bitbake glibc -c rebuild".
> > > * Fixes #1443.
> > > * Approved by Michael 'Mickey' Lauer <mickey at tm.informatik.uni-frankfurt de>.
> > 
> > I've tested here and those two patches break glibc 2.4 for my main
> > targets which are sh4 (titan), i486 (wrap) and i586 (epia), building
> > on an x86_64 host. The breakage basically exactly the same as in
> > #1443.
> > 
> > I'm not sure what "x86" problem is that they are meant to fix (look
> > like they are just wiring up syscalls to use in place of glibc equivs
> > for the *at calls), but at least for me they actually break things on
> > i486 and i586.
> 
> 1. patch fixes problem:

Right, but why are you seeing this issue when building for i486 and
I'm not? That's important since these patches break my currently
working builds.

Hmm, maybe it's those new linux-libc-headers that I see went in a few
days ago - they might now be exporting those *at functions when they
weren't before and therefore glibc is now picking them up. I'm still
using the 2.6.15.99 headers. I'll test with those newer ones now.

-- 
 Jamie Lenehan <lenehan@twibble.org>

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Michael 'Mickey' Lauer]

Am Freitag, den 29.09.2006, 07:39 +1000 schrieb Jamie Lenehan:
> Hmm, maybe it's those new linux-libc-headers that I see went in a few
> days ago - they might now be exporting those *at functions when they
> weren't before and therefore glibc is now picking them up. I'm still
> using the 2.6.15.99 headers. I'll test with those newer ones now.

Right. A couple of days ago, glibc-intermediate stopped building for me
wanting linux headers >= 2.6.17. I built the new 2.6.18 kernel-based
headers then
(which are still DEF_PREF=-1).

-- 
Regards,

Michael 'Mickey' Lauer | FreeLancer | http://www.Vanille-Media.de

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Jamie Lenehan]

On Fri, Sep 29, 2006 at 12:06:49AM +0200, Michael 'Mickey' Lauer wrote:
> Am Freitag, den 29.09.2006, 07:39 +1000 schrieb Jamie Lenehan:
> > Hmm, maybe it's those new linux-libc-headers that I see went in a few
> > days ago - they might now be exporting those *at functions when they
> > weren't before and therefore glibc is now picking them up. I'm still
> > using the 2.6.15.99 headers. I'll test with those newer ones now.
> 
> Right. A couple of days ago, glibc-intermediate stopped building for me
> wanting linux headers >= 2.6.17. I built the new 2.6.18 kernel-based
> headers then
> (which are still DEF_PREF=-1).

I've been building for for MACHINE=wrap (which is i486) with
linux-libc-headers 2.6.15.99 and everything is fine. If I change to
MACHINE=x86 (which is also i486), without changing anything else (ie,
still using linux-libc-headers 2.6.15.99), then it fails with the
"checking for kernel header at least 2.6.17" error.

If I change the OLDEST_KERNEL = "2.6.17" in machine/x86.conf to
OLDEST_KERNEL = "2.6.15" (still using linux-libc-headers 2.6.15.99)
then x86 works fine as well.

So it looks like with linux-libc-headers 2.6.15.99 (or earlier)
things are fine, but with linux-libc-headers 2.6.18 things break
unless you have those *at patches. And this is regardless of what
target you are building for.

Now the question is  how to make all these things play nicely
together...

-- 
 Jamie Lenehan <lenehan@twibble.org>

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Jamie Lenehan]

On Fri, Sep 29, 2006 at 09:30:42AM +1000, Jamie Lenehan wrote:
> On Fri, Sep 29, 2006 at 12:06:49AM +0200, Michael 'Mickey' Lauer wrote:
> > Am Freitag, den 29.09.2006, 07:39 +1000 schrieb Jamie Lenehan:
> > > Hmm, maybe it's those new linux-libc-headers that I see went in a few
> > > days ago - they might now be exporting those *at functions when they
> > > weren't before and therefore glibc is now picking them up. I'm still
> > > using the 2.6.15.99 headers. I'll test with those newer ones now.
> > 
> > Right. A couple of days ago, glibc-intermediate stopped building for me
> > wanting linux headers >= 2.6.17. I built the new 2.6.18 kernel-based
> > headers then
> > (which are still DEF_PREF=-1).
[...]
> Now the question is  how to make all these things play nicely
> together...

The upstream fix was different to what's in the bugzilla entry. Using the
following patch (thanks to lfs) in place of openat-bugzilla-fix-1220.patch
and fix-fchownat-20060808.patch seems to get everything working:

  http://www.linuxfromscratch.org/patches/downloads/glibc/glibc-2.4-openat-3.patch

I've tested this with with linux-libc-headers 2.6.15.99 and
linux-libc-headers 2.6.18 for wrap (i486) so far, and in both cases it
completed building glibc-intermediate-2.4.

-- 
 Jamie Lenehan <lenehan@twibble.org>

Re: [oe-commits] org.oe.dev glibc 2.4: Apply recently added patches only fo x86.

[Michael 'Mickey' Lauer]

Am Freitag, den 29.09.2006, 11:01 +1000 schrieb Jamie Lenehan:
> On Fri, Sep 29, 2006 at 09:30:42AM +1000, Jamie Lenehan wrote:
> > On Fri, Sep 29, 2006 at 12:06:49AM +0200, Michael 'Mickey' Lauer wrote:
> > > Am Freitag, den 29.09.2006, 07:39 +1000 schrieb Jamie Lenehan:
> > > > Hmm, maybe it's those new linux-libc-headers that I see went in a few
> > > > days ago - they might now be exporting those *at functions when they
> > > > weren't before and therefore glibc is now picking them up. I'm still
> > > > using the 2.6.15.99 headers. I'll test with those newer ones now.
> > > 
> > > Right. A couple of days ago, glibc-intermediate stopped building for me
> > > wanting linux headers >= 2.6.17. I built the new 2.6.18 kernel-based
> > > headers then
> > > (which are still DEF_PREF=-1).
> [...]
> > Now the question is  how to make all these things play nicely
> > together...
> 
> The upstream fix was different to what's in the bugzilla entry. Using the
> following patch (thanks to lfs) in place of openat-bugzilla-fix-1220.patch
> and fix-fchownat-20060808.patch seems to get everything working:
> 
>   http://www.linuxfromscratch.org/patches/downloads/glibc/glibc-2.4-openat-3.patch
> 
> I've tested this with with linux-libc-headers 2.6.15.99 and
> linux-libc-headers 2.6.18 for wrap (i486) so far, and in both cases it
> completed building glibc-intermediate-2.4.

Ok, I'll see whether it works on this machine here as well and then
commit.
Thanks for researching!

-- 
Regards,

Michael 'Mickey' Lauer | FreeLancer | http://www.Vanille-Media.de