calamari (ppc e500v2) failure

calamari (ppc e500v2) failure

[Tom Rini]

Hey all,

With DISTRO=minimal and MACHINE=calamari, I see the following failure in
gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
so this is unrelated.

-- 
Tom Rini <tom_rini@mentor.com>
Mentor Graphics Corporation

Re: calamari (ppc e500v2) failure

[Frans Meulenbroeks]

2010/5/7 Tom Rini <tom_rini@mentor.com>:
> Hey all,
>
> With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> so this is unrelated.

Which eglibc is that?
The issues I had seemed due to some catch22 situation that I cured; I
started a clean build yesterday before I left work & can tell the
results on monday.
My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
minimal also using 2.10 or is it at 2.11

I can also try to kick off a minimal build on monday

Frans

Re: calamari (ppc e500v2) failure

[Tom Rini]

On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> > Hey all,
> >
> > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> > so this is unrelated.
> 
> Which eglibc is that?
> The issues I had seemed due to some catch22 situation that I cured; I
> started a clean build yesterday before I left work & can tell the
> results on monday.
> My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> minimal also using 2.10 or is it at 2.11
> 
> I can also try to kick off a minimal build on monday

minimal is eglibc 2.10.  I'll try an Angstrom build.

-- 
Tom Rini <tom_rini@mentor.com>
Mentor Graphics Corporation

Re: calamari (ppc e500v2) failure

[Tom Rini]

On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
> On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> > > Hey all,
> > >
> > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> > > so this is unrelated.
> > 
> > Which eglibc is that?
> > The issues I had seemed due to some catch22 situation that I cured; I
> > started a clean build yesterday before I left work & can tell the
> > results on monday.
> > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> > minimal also using 2.10 or is it at 2.11
> > 
> > I can also try to kick off a minimal build on monday
> 
> minimal is eglibc 2.10.  I'll try an Angstrom build.

With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
cc1: error: not configured for ABI: 'spe'
configure:2591: $? = 1
configure: failed program was:
... standard empty program ...
configure:2605: error: cannot compute suffix of object files: cannot
compile

Ubuntu 9.10 x86_64 host


-- 
Tom Rini <tom_rini@mentor.com>
Mentor Graphics Corporation

Re: calamari (ppc e500v2) failure

[Khem Raj]

[-- Attachment #1: Type: text/plain, Size: 1815 bytes --]

On (09/05/10 11:24), Tom Rini wrote:
> On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
> > On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> > > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> > > > Hey all,
> > > >
> > > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> > > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> > > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> > > > so this is unrelated.
> > > 
> > > Which eglibc is that?
> > > The issues I had seemed due to some catch22 situation that I cured; I
> > > started a clean build yesterday before I left work & can tell the
> > > results on monday.
> > > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> > > minimal also using 2.10 or is it at 2.11
> > > 
> > > I can also try to kick off a minimal build on monday
> > 
> > minimal is eglibc 2.10.  I'll try an Angstrom build.
> 
> With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
> cc1: error: not configured for ABI: 'spe'
> configure:2591: $? = 1
> configure: failed program was:
> ... standard empty program ...
> configure:2605: error: cannot compute suffix of object files: cannot
> compile
> 
> Ubuntu 9.10 x86_64 host

Could you test with attached gcc patch please. This should fix the
undefined symbol error. You could use minimal distro for SPE targets
it should work. Please let me know if it works for you.

Thanks
-Khem




> 
> 
> -- 
> Tom Rini <tom_rini@mentor.com>
> Mentor Graphics Corporation
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

[-- Attachment #2: optspace.diff --]
[-- Type: text/x-diff, Size: 2064 bytes --]

diff --git a/recipes/gcc/gcc-configure-common.inc b/recipes/gcc/gcc-configure-common.inc
index fe8b486..892d684 100644
--- a/recipes/gcc/gcc-configure-common.inc
+++ b/recipes/gcc/gcc-configure-common.inc
@@ -21,10 +21,7 @@ OBJC ?= ",objc"
 # hidden symbols in libgcc.a which linker complains
 # when linking shared libraries further in the build like (gnutls)
 
-OPTSPACE_linux-gnuspe ?= ""
-OPTSPACE_linux-uclibcspe ?= ""
-OPTSPACE ?= "--enable-target-optspace"
-
+OPTSPACE = "${@base_contains('TARGET_ARCH', 'powerpc', '', '--enable-target-optspace',d)}"
 # gcc 3.x expects 'f77', 4.0 expects 'f95', 4.1 and 4.2 expect 'fortran'
 FORTRAN ?= ",f77"
 LANGUAGES ?= "c,c++${OBJC}${FORTRAN}${JAVA}"
diff --git a/recipes/gcc/gcc-cross-initial.inc b/recipes/gcc/gcc-cross-initial.inc
index 97002f9..25d7ba5 100644
--- a/recipes/gcc/gcc-cross-initial.inc
+++ b/recipes/gcc/gcc-cross-initial.inc
@@ -12,11 +12,11 @@ EXTRA_OECONF = "--with-local-prefix=${STAGING_DIR_TARGET}${target_prefix} \
 		--disable-multilib \
 		--disable-__cxa_atexit \
 		--enable-languages=c \
-		--enable-target-optspace \
 		--program-prefix=${TARGET_PREFIX} \
 		--with-sysroot=${STAGING_DIR_TARGET} \
 		--with-build-sysroot=${STAGING_DIR_TARGET} \
 		${EXTRA_OECONF_INITIAL} \
+		${OPTSPACE} \
 		${@get_gcc_fpu_setting(bb, d)} \
 		${@get_gcc_mips_plt_setting(bb, d)}"
 
diff --git a/recipes/gcc/gcc-cross-intermediate.inc b/recipes/gcc/gcc-cross-intermediate.inc
index 0c06d6f..cbc8638 100644
--- a/recipes/gcc/gcc-cross-intermediate.inc
+++ b/recipes/gcc/gcc-cross-intermediate.inc
@@ -10,10 +10,10 @@ EXTRA_OECONF = "--with-local-prefix=${STAGING_DIR_TARGET}${target_prefix} \
 		--disable-multilib \
 		--disable-threads \
 		--enable-languages=c \
-		--enable-target-optspace \
 		--program-prefix=${TARGET_PREFIX} \
 		--with-sysroot=${STAGING_DIR_TARGET} \
 		--with-build-sysroot=${STAGING_DIR_TARGET} \
+		${OPTSPACE} \
 		${EXTRA_OECONF_INTERMEDIATE} \
 		${@get_gcc_fpu_setting(bb, d)} \
 		${@get_gcc_mips_plt_setting(bb, d)}"

Re: calamari (ppc e500v2) failure

[Frans Meulenbroeks]

2010/5/9 Tom Rini <tom_rini@mentor.com>:
> On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
>> On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
>> > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
>> > > Hey all,
>> > >
>> > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
>> > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
>> > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
>> > > so this is unrelated.
>> >
>> > Which eglibc is that?
>> > The issues I had seemed due to some catch22 situation that I cured; I
>> > started a clean build yesterday before I left work & can tell the
>> > results on monday.
>> > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
>> > minimal also using 2.10 or is it at 2.11
>> >
>> > I can also try to kick off a minimal build on monday
>>
>> minimal is eglibc 2.10.  I'll try an Angstrom build.
>
> With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
> cc1: error: not configured for ABI: 'spe'
> configure:2591: $? = 1
> configure: failed program was:
> ... standard empty program ...
> configure:2605: error: cannot compute suffix of object files: cannot
> compile
>
> Ubuntu 9.10 x86_64 host
>

I made a mistake when doing the clean build, and kicked it off again.
I don't think I saw your error.
In  my local.conf I have:

MACHINE = "calamari"
DISTRO = "angstrom-2008.1"
ANGSTROMLIBC = "eglibc"

# e500v2 supports the SPE instruction set.
TARGET_OS = "linux-gnuspe"

Perhaps the TARGET_OS line influences how libgcc is build.

Frans

Re: calamari (ppc e500v2) failure

[Tom Rini]

On Mon, 2010-05-10 at 10:40 +0200, Frans Meulenbroeks wrote:
> 2010/5/9 Tom Rini <tom_rini@mentor.com>:
> > On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
> >> On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> >> > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> >> > > Hey all,
> >> > >
> >> > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> >> > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> >> > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> >> > > so this is unrelated.
> >> >
> >> > Which eglibc is that?
> >> > The issues I had seemed due to some catch22 situation that I cured; I
> >> > started a clean build yesterday before I left work & can tell the
> >> > results on monday.
> >> > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> >> > minimal also using 2.10 or is it at 2.11
> >> >
> >> > I can also try to kick off a minimal build on monday
> >>
> >> minimal is eglibc 2.10.  I'll try an Angstrom build.
> >
> > With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
> > cc1: error: not configured for ABI: 'spe'
> > configure:2591: $? = 1
> > configure: failed program was:
> > ... standard empty program ...
> > configure:2605: error: cannot compute suffix of object files: cannot
> > compile
> >
> > Ubuntu 9.10 x86_64 host
> >
> 
> I made a mistake when doing the clean build, and kicked it off again.
> I don't think I saw your error.
> In  my local.conf I have:
> 
> MACHINE = "calamari"
> DISTRO = "angstrom-2008.1"
> ANGSTROMLIBC = "eglibc"

This should just be LIBC, btw.

> # e500v2 supports the SPE instruction set.
> TARGET_OS = "linux-gnuspe"

This should also be picked out automatically, iirc.

-- 
Tom Rini <tom_rini@mentor.com>
Mentor Graphics Corporation

Re: calamari (ppc e500v2) failure

[Khem Raj]

On (10/05/10 10:40), Frans Meulenbroeks wrote:
> 2010/5/9 Tom Rini <tom_rini@mentor.com>:
> > On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
> >> On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> >> > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> >> > > Hey all,
> >> > >
> >> > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> >> > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> >> > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> >> > > so this is unrelated.
> >> >
> >> > Which eglibc is that?
> >> > The issues I had seemed due to some catch22 situation that I cured; I
> >> > started a clean build yesterday before I left work & can tell the
> >> > results on monday.
> >> > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> >> > minimal also using 2.10 or is it at 2.11
> >> >
> >> > I can also try to kick off a minimal build on monday
> >>
> >> minimal is eglibc 2.10.  I'll try an Angstrom build.
> >
> > With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
> > cc1: error: not configured for ABI: 'spe'
> > configure:2591: $? = 1
> > configure: failed program was:
> > ... standard empty program ...
> > configure:2605: error: cannot compute suffix of object files: cannot
> > compile
> >
> > Ubuntu 9.10 x86_64 host
> >
> 
> I made a mistake when doing the clean build, and kicked it off again.
> I don't think I saw your error.
> In  my local.conf I have:
> 
> MACHINE = "calamari"
> DISTRO = "angstrom-2008.1"
> ANGSTROMLIBC = "eglibc"
> 
> # e500v2 supports the SPE instruction set.
> TARGET_OS = "linux-gnuspe"
> 
> Perhaps the TARGET_OS line influences how libgcc is build.

yes. if you look at the gcc-configure-cross.inc then it disables
optspace for SPE targets. thats why it works for gnuspe but when I
did that I only saw the problem on gnuspe and wanted to keep the 
effect minimal. But it seems now we need this ppc wide thats why
the new patch. 

Although using -Os in kernel might cause the same problem even if we
sort out libgcc because kernel does not use libgcc and I dont know if
it implements these functions.

If this causes the issue then we have to do same fix for kernel where
we compile kernel with -O2 and not with -Os

Thx
-Khem

> 
> Frans
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: calamari (ppc e500v2) failure

[Frans Meulenbroeks]

2010/5/10 Tom Rini <tom_rini@mentor.com>:
> On Mon, 2010-05-10 at 10:40 +0200, Frans Meulenbroeks wrote:

>> MACHINE = "calamari"
>> DISTRO = "angstrom-2008.1"
>> ANGSTROMLIBC = "eglibc"
>
> This should just be LIBC, btw

Angstrom has ANGSTROMLIBC

From the code (grepping in the conf dir):
conf/distro/include/angstrom.inc:ANGSTROMLIBC ?= "glibc"
conf/distro/include/angstrom.inc:LIBC ?= "${ANGSTROMLIBC}"

No idea why we have ANGSTROMLIBC next to LIBC.

>
>> # e500v2 supports the SPE instruction set.
>> TARGET_OS = "linux-gnuspe"
>
> This should also be picked out automatically, iirc.

Hm, It was there in the conf that I got when I took over this project.
I seem to recall having tried to build without it, and that it failed,
but it is a while ago, so not 100% sure about that.

Frans.

Re: calamari (ppc e500v2) failure

[Tom Rini]

On Tue, 2010-05-11 at 08:53 +0200, Frans Meulenbroeks wrote:
> 2010/5/10 Tom Rini <tom_rini@mentor.com>:
> > On Mon, 2010-05-10 at 10:40 +0200, Frans Meulenbroeks wrote:
> 
> >> MACHINE = "calamari"
> >> DISTRO = "angstrom-2008.1"
> >> ANGSTROMLIBC = "eglibc"
> >
> > This should just be LIBC, btw
> 
> Angstrom has ANGSTROMLIBC
> 
> From the code (grepping in the conf dir):
> conf/distro/include/angstrom.inc:ANGSTROMLIBC ?= "glibc"
> conf/distro/include/angstrom.inc:LIBC ?= "${ANGSTROMLIBC}"
> 
> No idea why we have ANGSTROMLIBC next to LIBC.

Because Angstrom initiated the hook and then I made it generic and made
all of the distros use it.  ANGSTROMLIBC is still there for
compatibility.

> >> # e500v2 supports the SPE instruction set.
> >> TARGET_OS = "linux-gnuspe"
> >
> > This should also be picked out automatically, iirc.
> 
> Hm, It was there in the conf that I got when I took over this project.
> I seem to recall having tried to build without it, and that it failed,
> but it is a while ago, so not 100% sure about that.

Well, more accurately, sane-toolchain.inc has a function to get
TARGET_OS right and I assume angstrom is either using it or has a
similar one.

-- 
Tom Rini <tom_rini@mentor.com>
Mentor Graphics Corporation

Re: calamari (ppc e500v2) failure

[Tom Rini]

On Sun, 2010-05-09 at 22:49 -0700, Khem Raj wrote:
> On (09/05/10 11:24), Tom Rini wrote:
> > On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
> > > On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> > > > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> > > > > Hey all,
> > > > >
> > > > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> > > > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> > > > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> > > > > so this is unrelated.
> > > > 
> > > > Which eglibc is that?
> > > > The issues I had seemed due to some catch22 situation that I cured; I
> > > > started a clean build yesterday before I left work & can tell the
> > > > results on monday.
> > > > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> > > > minimal also using 2.10 or is it at 2.11
> > > > 
> > > > I can also try to kick off a minimal build on monday
> > > 
> > > minimal is eglibc 2.10.  I'll try an Angstrom build.
> > 
> > With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
> > cc1: error: not configured for ABI: 'spe'
> > configure:2591: $? = 1
> > configure: failed program was:
> > ... standard empty program ...
> > configure:2605: error: cannot compute suffix of object files: cannot
> > compile
> > 
> > Ubuntu 9.10 x86_64 host
> 
> Could you test with attached gcc patch please. This should fix the
> undefined symbol error. You could use minimal distro for SPE targets
> it should work. Please let me know if it works for you.

This fixes the e300c3 / ppc603e issue, but not the gnutls issue (which
is now seen on those platforms).  So the general problem of -Os + ppc
not being right with GNU.org toolchains is still there, on 4.4.3 at
least.

-- 
Tom Rini <tom_rini@mentor.com>
Mentor Graphics Corporation

Re: calamari (ppc e500v2) failure

[Khem Raj]

On (11/05/10 07:32), Tom Rini wrote:
> On Sun, 2010-05-09 at 22:49 -0700, Khem Raj wrote:
> > On (09/05/10 11:24), Tom Rini wrote:
> > > On Sat, 2010-05-08 at 08:11 -0700, Tom Rini wrote:
> > > > On Sat, 2010-05-08 at 09:06 +0200, Frans Meulenbroeks wrote:
> > > > > 2010/5/7 Tom Rini <tom_rini@mentor.com>:
> > > > > > Hey all,
> > > > > >
> > > > > > With DISTRO=minimal and MACHINE=calamari, I see the following failure in
> > > > > > gnutls (using fe4ff6940b06c73d5b88224b99750bf65accf13d in oe.dev):
> > > > > > http://pastebin.com/cjAStcLb and I confirmed this prior to the RP merge,
> > > > > > so this is unrelated.
> > > > > 
> > > > > Which eglibc is that?
> > > > > The issues I had seemed due to some catch22 situation that I cured; I
> > > > > started a clean build yesterday before I left work & can tell the
> > > > > results on monday.
> > > > > My distro was angstrom, machine calamari, this uses eglibc 2.10. Is
> > > > > minimal also using 2.10 or is it at 2.11
> > > > > 
> > > > > I can also try to kick off a minimal build on monday
> > > > 
> > > > minimal is eglibc 2.10.  I'll try an Angstrom build.
> > > 
> > > With Angstrom, gcc-cross-initial fails building libgcc, config.log has:
> > > cc1: error: not configured for ABI: 'spe'
> > > configure:2591: $? = 1
> > > configure: failed program was:
> > > ... standard empty program ...
> > > configure:2605: error: cannot compute suffix of object files: cannot
> > > compile
> > > 
> > > Ubuntu 9.10 x86_64 host
> > 
> > Could you test with attached gcc patch please. This should fix the
> > undefined symbol error. You could use minimal distro for SPE targets
> > it should work. Please let me know if it works for you.
> 
> This fixes the e300c3 / ppc603e issue, but not the gnutls issue (which
> is now seen on those platforms).  So the general problem of -Os + ppc
> not being right with GNU.org toolchains is still there, on 4.4.3 at
> least.

out of order epilogue/prologue are added in gcc 4.4 so I expect 4.4+ to
show this issue. That said, when I tried for calamari from scratch 
bitbake gnutls it went through fine on Kubuntu 10.04/x86_64 build machine.
> 
> -- 
> Tom Rini <tom_rini@mentor.com>
> Mentor Graphics Corporation
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel