updated libunwind patch for v2.6 kernel

updated libunwind patch for v2.6 kernel

[David Mosberger]

An updated version of the libunwind patch for the 2.6-kernel is now
available at:

 ftp://ftp.hpl.hp.com/pub/linux-ia64/unwind-2.6-050609.diff.gz

The only change is that the libunwind API has been exported to
modules.  The problem I reported earlier with strace'ing an x86 task
apparently had nothing to do with the libunwind changes as it went
away before I even had a chance to investigate.

This patch passes all my testing so far.  The only thing that I
haven't tested (at all) is the oprofile call-stack tracing.  I'd be
interested to hear if/how that works.  (I did drop the old compiler
support there --- I hope nobody uses such old compilers anymore and
the code was just gross; but if we have to, we can resurrect it).

Tony, I know you're in the middle of making the test-2.6 tree work
better for you.  For now, the patch above is relative to Linus 2.6
tree, but if you have no objections, I'd like to feed it into your
test-2.6 as soon as possible (just let me know when you're ready and I
will respin the patch relative to test-2.6).

Thanks,

	--david

Re: updated libunwind patch for v2.6 kernel

[Bob Picco]

David Mosberger wrote:	[Fri Jun 10 2005, 01:31:34AM EDT]
> An updated version of the libunwind patch for the 2.6-kernel is now
> available at:
> 
>  ftp://ftp.hpl.hp.com/pub/linux-ia64/unwind-2.6-050609.diff.gz
> 
> The only change is that the libunwind API has been exported to
> modules.  The problem I reported earlier with strace'ing an x86 task
> apparently had nothing to do with the libunwind changes as it went
> away before I even had a chance to investigate.
> 
> This patch passes all my testing so far.  The only thing that I
> haven't tested (at all) is the oprofile call-stack tracing.  I'd be
> interested to hear if/how that works.  (I did drop the old compiler
> support there --- I hope nobody uses such old compilers anymore and
> the code was just gross; but if we have to, we can resurrect it).
> 
> Tony, I know you're in the middle of making the test-2.6 tree work
> better for you.  For now, the patch above is relative to Linus 2.6
> tree, but if you have no objections, I'd like to feed it into your
> test-2.6 as soon as possible (just let me know when you're ready and I
> will respin the patch relative to test-2.6).
> 
> Thanks,
> 
> 	--david
> -
Hi David,

I have build problems after applying this patch against
rsync://rsync.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git.
I've also have kgdb patches applied but don't believe they are the
offending issue.

Appended is the relevant part of the kernel make output.

thanks,

bob


  ia64-linux-gcc -Wp,-MD,arch/ia64/unwind/.init_local.o.d  -nostdinc -isystem /usr/local/lib/gcc-lib/ia64-linux/3.3/include -D__KERNEL__ -Iinclude  -DHAVE_BUGGY_SEGREL -DHAVE_ATOMIC_H  -Wall -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -ffreestanding -O2     -fomit-frame-pointer -g -pipe  -ffixed-r13 -mfixed-rangeñ2-f15,f32-f127 -falign-functions2 -frename-registers -fno-optimize-sibling-calls     -mconstant-gp -DKBUILD_BASENAME=init_local -DKBUILD_MODNAME=init_local -c -o arch/ia64/unwind/init_local.o arch/ia64/unwind/init_local.c
In file included from arch/ia64/unwind/init.h:26,
                 from arch/ia64/unwind/init_local.c:26:
arch/ia64/unwind/unwind_i.h:33:25: libunwind_i.h: No such file or directory
In file included from arch/ia64/unwind/unwind_i.h:34,
                 from arch/ia64/unwind/init.h:26,
                 from arch/ia64/unwind/init_local.c:26:
arch/ia64/unwind/rse.h:13:23: libunwind.h: No such file or directory
In file included from arch/ia64/unwind/init.h:26,
                 from arch/ia64/unwind/init_local.c:26:
arch/ia64/unwind/unwind_i.h:52: error: syntax error before "void"
arch/ia64/unwind/unwind_i.h: In function `inlined_sw_addr':
arch/ia64/unwind/unwind_i.h:55: error: `unw_word_t' undeclared (first use in this function)
[snip]

Re: updated libunwind patch for v2.6 kernel

[David Mosberger]

Bob,

>>>>> On Wed, 15 Jun 2005 12:55:50 -0400, Bob Picco <bob.picco@hp.com> said:

  Bob> ia64-linux-gcc -Wp,-MD,arch/ia64/unwind/.init_local.o.d  -nostdinc -isystem /usr/local/lib/gcc-lib/ia64-linux/3.3/include -D__KERNEL__ -Iinclude  -DHAVE_BUGGY_SEGREL -DHAVE_ATOMIC_H  -Wall -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -ffreestanding -O2     -fomit-frame-pointer -g -pipe  -ffixed-r13 -mfixed-rangeñ2-f15,f32-f127 -falign-functions2 -frename-registers -fno-optimize-sibling-calls     -mconstant-gp -DKBUILD_BASENAME=init_local -DKBUILD_MODNAME=init_local -c -o arch/ia64/unwind/init_local.o arch/ia64/unwind/init_local.c
  Bob> In file included from arch/ia64/unwind/init.h:26,
  Bob> from arch/ia64/unwind/init_local.c:26:
  Bob> arch/ia64/unwind/unwind_i.h:33:25: libunwind_i.h: No such file or directory

Looks like the compiler command is missing the -Iinclude/linux, which
should get added via CPPFLAGS in arch/ia64/unwind/Makefile:

CPPFLAGS        += -DHAVE_ATOMIC_H $(call addtree,-Iinclude/linux)

are you building in the source tree itself?  If so, can you try
building with make O=objdir and see if that fixes the problem.  I'm
unsure if addtree is safe to be called from the Makefile, but I didn't
want to worry about relatively minor build questions at this time.

Thanks,

	--david

Re: updated libunwind patch for v2.6 kernel

[Bob Picco]

David Mosberger wrote:	[Wed Jun 15 2005, 01:35:44PM EDT]
> Bob,
> 
> >>>>> On Wed, 15 Jun 2005 12:55:50 -0400, Bob Picco <bob.picco@hp.com> said:
> 
>   Bob> ia64-linux-gcc -Wp,-MD,arch/ia64/unwind/.init_local.o.d  -nostdinc -isystem /usr/local/lib/gcc-lib/ia64-linux/3.3/include -D__KERNEL__ -Iinclude  -DHAVE_BUGGY_SEGREL -DHAVE_ATOMIC_H  -Wall -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -ffreestanding -O2     -fomit-frame-pointer -g -pipe  -ffixed-r13 -mfixed-rangeñ2-f15,f32-f127 -falign-functions2 -frename-registers -fno-optimize-sibling-calls     -mconstant-gp -DKBUILD_BASENAME=init_local -DKBUILD_MODNAME=init_local -c -o arch/ia64/unwind/init_local.o arch/ia64/unwind/init_local.c
>   Bob> In file included from arch/ia64/unwind/init.h:26,
>   Bob> from arch/ia64/unwind/init_local.c:26:
>   Bob> arch/ia64/unwind/unwind_i.h:33:25: libunwind_i.h: No such file or directory
> 
> Looks like the compiler command is missing the -Iinclude/linux, which
> should get added via CPPFLAGS in arch/ia64/unwind/Makefile:
> 
> CPPFLAGS        += -DHAVE_ATOMIC_H $(call addtree,-Iinclude/linux)
> 
> are you building in the source tree itself?  If so, can you try
Normal kernel build and build tree.
> building with make O=objdir and see if that fixes the problem.  I'm
> unsure if addtree is safe to be called from the Makefile, but I didn't
> want to worry about relatively minor build questions at this time.
No problem. I just hacked around it.
CPPFLAGS        += -DHAVE_ATOMIC_H -Iinclude/linux
> 
> Thanks,
> 
> 	--david
thanks,

bob