[parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain + loop

[parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain + loop

[Carlos O'Donell]

2.6.7-pa0 fails to build on newer 64-bit toolchain:

  CC      drivers/block/loop.o
{standard input}: Assembler messages:
{standard input}:3652: Error: Field not properly aligned [8] (-183).
{standard input}:3652: Error: Invalid operands 
make[2]: *** [drivers/block/loop.o] Error 1
make[1]: *** [drivers/block] Error 2
make: *** [drivers] Error 2

Just a heads up.

c.

_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain + loop

[Kyle McMartin]

On Wed, Jun 16, 2004 at 03:44:11PM -0400, Carlos O'Donell wrote:
> Just a heads up.
> 
I know you already know this, but for posterity's sake, I'm following up
with the GCC bug that Randolph filed for this.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14782

Regards,
-- 
Kyle McMartin
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up,

[John David Anglin]

> On Fri, Jun 18, 2004 at 05:25:05PM -0400, Kyle McMartin wrote:
> > gcc-3.4-hppa (or gcc-3.3-hppa)
> 
> Should I be testing 64-bit kernels built with this toolchain
> or stick with my trusty gcc 3.0.4-hppa64 handmade debs?

>>From my perspective, the 3.4 branch is the best yet.  By this, I
mean it has the fewest bugs in the PA backend.  However, there are
lots of other changes that may bite ...

Randolph's PR (the one that started this thread) isn't fixed and
hasn't been seriously researched (too many other issues).  However,
it's obviously a serious issue for the kernel.  So, maybe you should
use 3.0.4 for kernel builds until it's resolved.  For other uses, I
suggest 3.3.4 (released?) or the soon to be released 3.4.1.  3.5 is
for Joel to test ;-)

If anyone out there wants to learn more than they really wanted
to know about GCC and binutils internals, let me know as there
are a number of projects that need working on.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up,

[John David Anglin]

> I've been building hppa64 kernels with the gcc 3.0.4 debs on the FTP
> server, until this issue is resolved. :/

I'm 99% certain that this issue is not a target related problem.  The way to
move this forward is to provide more analysis and keep the PR visible.  This
could be more details on the nature of the failure, etc.

It's important for others to come forward on problem reports as this
shows an interest from the community at large in resolving problems on
the PA.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain + loop

[Joel Soete]

Hello all,

And there is also this pb mentionned by James:
<http://lists.parisc-linux.org/pipermail/parisc-linux/2004-February/022335.html>

regarding the pb of pointer comparison with gcc-3.3 and 3.4 64bits (sorry I didn't find back the gcc bug report reference) which 
iirc jda re-qualified it in a binutils pb. (That why of my interest in binutils 2.15 :) to see if I can help.)

That said, my very interest in 3.4 (and 3.5 now because is became the debian gcc-snapshot ;) ) is to continue the glibc-2.3.3 test 
and see if additional integer lib (hp one iirc) can help for the few test which still failled?

I was focused last month on my c110 ncr53c8xx bug; now I trust it is fixed, so jda if you think that I can be of better help else 
where I can always try to have a look ;). Please advise.

Joel




Kyle McMartin wrote:
> On Fri, Jun 18, 2004 at 07:17:14PM -0600, Grant Grundler wrote:
> 
>>Should I be testing 64-bit kernels built with this toolchain
>>or stick with my trusty gcc 3.0.4-hppa64 handmade debs?
>>
> 
> There's a bug in hppa64 gcc sometime after 3.0.4 (it shows up in the
> gcc-3.3-hppa64 and gcc-3.4-hppa64 debs) that causes builds to fail in
> (at least) loop.c.
> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14782 is the bug Randolph
> filed.
> 
> I've been building hppa64 kernels with the gcc 3.0.4 debs on the FTP
> server, until this issue is resolved. :/
> 
> Regards,

_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain

[John David Anglin]

> regarding the pb of pointer comparison with gcc-3.3 and 3.4 64bits (sorry I didn't find back the gcc bug report reference) which 
> iirc jda re-qualified it in a binutils pb. (That why of my interest in binutils 2.15 :) to see if I can help.)

I don't know of any hppa64 bugs involving pointer comparison.  The cause
of the binutils bug is known:

<http://sources.redhat.com/ml/binutils/2004-06/msg00211.html>.

Until this is fixed, don't use cvs binutils from the trunk after
2004-05-11 17:08:32.

> Kyle McMartin wrote:
> > On Fri, Jun 18, 2004 at 07:17:14PM -0600, Grant Grundler wrote:
> > 
> >>Should I be testing 64-bit kernels built with this toolchain
> >>or stick with my trusty gcc 3.0.4-hppa64 handmade debs?
> >>
> > 
> > There's a bug in hppa64 gcc sometime after 3.0.4 (it shows up in the
> > gcc-3.3-hppa64 and gcc-3.4-hppa64 debs) that causes builds to fail in
> > (at least) loop.c.
> > 
> > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14782 is the bug Randolph
> > filed.

I believe that I understand the cause of this bug and have started
testing a fix.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain

[John David Anglin]

> > Kyle McMartin wrote:
> > > On Fri, Jun 18, 2004 at 07:17:14PM -0600, Grant Grundler wrote:
> > > 
> > >>Should I be testing 64-bit kernels built with this toolchain
> > >>or stick with my trusty gcc 3.0.4-hppa64 handmade debs?
> > >>
> > > 
> > > There's a bug in hppa64 gcc sometime after 3.0.4 (it shows up in the
> > > gcc-3.3-hppa64 and gcc-3.4-hppa64 debs) that causes builds to fail in
> > > (at least) loop.c.
> > > 
> > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14782 is the bug Randolph
> > > filed.
> 
> I believe that I understand the cause of this bug and have started
> testing a fix.

Enclosed is the patch that I am testing.  It seems to generated reasonable
code for the testcase that Carlos submitted today.  As far as I can tell,
the bug has been present in the FSF sources since the first support for
the 64-bit port (3.0 branch).  Thus, I expect the bug is present in debian
3.0.4, just not triggered.  The bug is triggered by the loop optimizer.

I would be interested in hearing if this solves the kernel build problem.
The patch is for 3.5.  It might need a bit of hacking to apply it to
earlier branches.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

2004-06-19  John David Anglin  <dave.anglin@nrc-cnrc.gc.ca>

	* pa.h (GO_IF_LEGITIMATE_ADDRESS): Require constant offsets to be
	correctly aligned for DImode loads and stores.

Index: config/pa/pa.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/pa/pa.h,v
retrieving revision 1.218
diff -u -3 -p -r1.218 pa.h
--- config/pa/pa.h	13 May 2004 06:40:03 -0000	1.218
+++ config/pa/pa.h	20 Jun 2004 00:32:04 -0000
@@ -1467,13 +1467,18 @@ extern int may_call_alloca;
       if (base								\
 	  && GET_CODE (index) == CONST_INT				\
 	  && ((INT_14_BITS (index)					\
-	       && (TARGET_SOFT_FLOAT					\
-		   || (TARGET_PA_20					\
-		       && ((MODE == SFmode				\
-			    && (INTVAL (index) % 4) == 0)		\
-			   || (MODE == DFmode				\
+	       && ((TARGET_64BIT && (MODE) == DImode			\
+		    && (INTVAL (index) % 8) == 0)			\
+		   || ((MODE) == SFmode					\
+		       && (TARGET_SOFT_FLOAT				\
+			   || (TARGET_PA_20				\
+			       && (INTVAL (index) % 4) == 0)))		\
+		   || ((MODE) == DFmode					\
+		       && (TARGET_SOFT_FLOAT				\
+			   || (TARGET_PA_20				\
 			       && (INTVAL (index) % 8) == 0)))		\
-		   || ((MODE) != SFmode && (MODE) != DFmode)))		\
+		   || ((!TARGET_64BIT || (MODE) != DImode)		\
+		       && (MODE) != SFmode && (MODE) != DFmode)))	\
 	       || INT_5_BITS (index)))					\
 	goto ADDR;							\
       if (!TARGET_DISABLE_INDEXING					\
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain

[Matthias Klose]

John David Anglin writes:
> > > Kyle McMartin wrote:
> > > > On Fri, Jun 18, 2004 at 07:17:14PM -0600, Grant Grundler wrote:
> > > > 
> > > >>Should I be testing 64-bit kernels built with this toolchain
> > > >>or stick with my trusty gcc 3.0.4-hppa64 handmade debs?
> > > >>
> > > > 
> > > > There's a bug in hppa64 gcc sometime after 3.0.4 (it shows up in the
> > > > gcc-3.3-hppa64 and gcc-3.4-hppa64 debs) that causes builds to fail in
> > > > (at least) loop.c.
> > > > 
> > > > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14782 is the bug Randolph
> > > > filed.
> > 
> > I believe that I understand the cause of this bug and have started
> > testing a fix.
> 
> Enclosed is the patch that I am testing.  It seems to generated reasonable
> code for the testcase that Carlos submitted today.  As far as I can tell,
> the bug has been present in the FSF sources since the first support for
> the 64-bit port (3.0 branch).  Thus, I expect the bug is present in debian
> 3.0.4, just not triggered.  The bug is triggered by the loop optimizer.
> 
> I would be interested in hearing if this solves the kernel build problem.
> The patch is for 3.5.  It might need a bit of hacking to apply it to
> earlier branches.

updated packages at

http://people.debian.org/~doko/gcc-3.3/gcc-3.3-hppa64_3.3.4-1_hppa.deb
http://people.debian.org/~doko/gcc-3.4/gcc-3.4-hppa64_3.4.0-3_hppa.deb

install the latter using dpkg -i --force-depends
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain

[Carlos O'Donell]

On Sat, Jun 19, 2004 at 09:14:28PM -0400, John David Anglin wrote:
> > I believe that I understand the cause of this bug and have started
> > testing a fix.
> 
> Enclosed is the patch that I am testing.  It seems to generated reasonable
> code for the testcase that Carlos submitted today.  As far as I can tell,
> the bug has been present in the FSF sources since the first support for
> the 64-bit port (3.0 branch).  Thus, I expect the bug is present in debian
> 3.0.4, just not triggered.  The bug is triggered by the loop optimizer.
> 
> I would be interested in hearing if this solves the kernel build problem.
> The patch is for 3.5.  It might need a bit of hacking to apply it to
> earlier branches.

I re-read the last email about "bugs getting fixed if you show that the
community has interest", so I simplified randolph's bug down to a simple
structure alignment issue with two structures :) I'm glad to see that
was of some help in clarifying the bug.

I also estatic that you fixed this bug, it means I can start compiling
64-bit kernels again.

Kudos to Matthias for the hppa64-linux-gcc debs :)

c.

_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain

[Kyle McMartin]

On Sat, Jun 19, 2004 at 09:14:28PM -0400, John David Anglin wrote:
> I would be interested in hearing if this solves the kernel build problem.
> The patch is for 3.5.  It might need a bit of hacking to apply it to
> earlier branches.
> 
Aye, it fixes the kernel compile! Hooray!

(I built 2.6.7-pa1 successfully on my c3000 with it, at least, building
SMP now to double check, and haven't booted)

Cheers,
-- 
Kyle McMartin
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain + loop

[Matthias Klose]

Joel Soete writes:
> That said, my very interest in 3.4 (and 3.5 now because is became
> the debian gcc-snapshot ;) )

gcc-3.4 is available as a package as well. Add the following line to
to your sources.list.

deb http://ftp.debian.org/debian ../project/experimental main contrib non-free

then install it using apt-get -t experimental gcc-3.4
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain + loop

[Matthias Klose]

Joel Soete writes:
> > deb http://ftp.debian.org/debian ../project/experimental main contrib non-free

sorry, should be ftp.de.debian.org, or every other mirror.

> ok i miss something in my sources.list I only have:
> deb http://ftp.de.debian.org/debian experimental main contrib non-free
> deb-src http://ftp.de.debian.org/debian experimental main contrib non-free
> 
> A link? (Packages, Sources, Release files only look like the same)

deb http://ftp.de.debian.org/debian ../project/experimental main contrib non-free
deb http://ftp.de.debian.org/debian experimental main contrib non-free

both lines should work.

> > then install it using apt-get -t experimental gcc-3.4

apt-get -t experimental install gcc-3.4


_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux

Re: [parisc-linux] Heads up, 2.6.7-pa0 FTBS with 64-bit toolchain

[John David Anglin]

> The build of the toolchain jsut failled as previously:
> <http://lists.parisc-linux.org/pipermail/parisc-linux/2004-April/022823.html>

Try

/CAD/parisc-linux/xc/bin/hppa-linux-gcc -print-file-name=libgcc.a
/CAD/parisc-linux/xc/bin/hppa-linux-gcc -print-file-name=libgcc_eh.a

This should show where thie compiler thinks libgcc.a and libgcc_eh.a
are.  You can see the full link command by adding "-Wl,-debug" to
the gcc command that failed.  GCC normally adds a "-L" option to the
directory where libgcc.a and libgcc_eh.a are installed.  If libgcc_eh.a
isn't there, you have a GCC build or install problem.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)
_______________________________________________
parisc-linux mailing list
parisc-linux@lists.parisc-linux.org
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux