Re: [discuss] RE: [PATCH] [1/2] x86_64: Pass -32 to the assembler when compiling the 32bit vsyscall pages

Re: [discuss] RE: [PATCH] [1/2] x86_64: Pass -32 to the assembler when compiling the 32bit vsyscall pages

[H. J. Lu]

On Fri, Apr 21, 2006 at 09:25:33AM -0500, Baker, Brian (ISS - Houston) wrote:
> The credit for this should actually go to Dave Altobelli, not myself.
> 
> -----Original Message-----
> From: ak@suse.de [mailto:ak@suse.de] 
> Sent: Friday, April 21, 2006 2:42 AM
> To: torvalds@osdl.org
> Cc: discuss@x86-64.org; akpm@osdl.org; Baker, Brian (ISS - Houston)
> Subject: [PATCH] [1/2] x86_64: Pass -32 to the assembler when compiling
> the 32bit vsyscall pages
> 
> 
> This quietens warnings and actually fixes a bug. The unwind tables would
> come out wrong without -32, causing pthread cancellation during them 
> to crash in the gcc runtime.
> 
> The problem seems to only happen with newer binutils
> (it doesn't happen with 2.16.91.0.2 but happens wit 2.16.91.0.5) 

If you believe it is true, please open a bug report at

http://www.sourceware.org/bugzilla/

> 
> Thanks to Brian Baker @ HP for test case and initial analysis.
> 
> Cc: brian.b@hp.com
> 
> Signed-off-by: Andi Kleen <ak@suse.de>
> 
> ---
>  arch/x86_64/ia32/Makefile |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
> 
> Index: linux/arch/x86_64/ia32/Makefile
> ===================================================================
> --- linux.orig/arch/x86_64/ia32/Makefile
> +++ linux/arch/x86_64/ia32/Makefile
> @@ -27,5 +27,5 @@ $(obj)/vsyscall-sysenter.so $(obj)/vsysc
>  $(obj)/vsyscall-%.so: $(src)/vsyscall.lds $(obj)/vsyscall-%.o FORCE
>  	$(call if_changed,syscall)
>  
> -AFLAGS_vsyscall-sysenter.o = -m32
> -AFLAGS_vsyscall-syscall.o = -m32
> +AFLAGS_vsyscall-sysenter.o = -m32 -Wa,-32
> +AFLAGS_vsyscall-syscall.o = -m32 -Wa,-32

I can't reproduce it with binutils 2.16.91.0.5. Unless you have a bad
gcc driver, -m32 will pass --32 to assembler, which is the same as -32.


H.J.

Re: [discuss] RE: [PATCH] [1/2] x86_64: Pass -32 to the assembler when compiling the 32bit vsyscall pages

[Andi Kleen]

> > Index: linux/arch/x86_64/ia32/Makefile
> > ===================================================================
> > --- linux.orig/arch/x86_64/ia32/Makefile
> > +++ linux/arch/x86_64/ia32/Makefile
> > @@ -27,5 +27,5 @@ $(obj)/vsyscall-sysenter.so $(obj)/vsysc
> >  $(obj)/vsyscall-%.so: $(src)/vsyscall.lds $(obj)/vsyscall-%.o FORCE
> >  	$(call if_changed,syscall)
> >
> > -AFLAGS_vsyscall-sysenter.o = -m32
> > -AFLAGS_vsyscall-syscall.o = -m32
> > +AFLAGS_vsyscall-sysenter.o = -m32 -Wa,-32
> > +AFLAGS_vsyscall-syscall.o = -m32 -Wa,-32
>
> I can't reproduce it

What did you try exactly? 

You have to recompile the file and either run the pthreads
cancellation test case with the right glibc version, or examine
the unwind tables in the .so by hand. Or just check if the warning
appears or not.

> with binutils 2.16.91.0.5. Unless you have a bad 
> gcc driver, -m32 will pass --32 to assembler, which is the same as -32.

It definitely fixed the problem here and the warning went away.
gcc used was 4.1

But you're right it could have been the gcc - in the case where it worked
I was using gcc 4.0.

-Andi

Re: [discuss] RE: [PATCH] [1/2] x86_64: Pass -32 to the assembler when compiling the 32bit vsyscall pages

[H. J. Lu]

On Sat, Apr 22, 2006 at 12:25:52AM +0200, Andi Kleen wrote:
> 
> > > Index: linux/arch/x86_64/ia32/Makefile
> > > ===================================================================
> > > --- linux.orig/arch/x86_64/ia32/Makefile
> > > +++ linux/arch/x86_64/ia32/Makefile
> > > @@ -27,5 +27,5 @@ $(obj)/vsyscall-sysenter.so $(obj)/vsysc
> > >  $(obj)/vsyscall-%.so: $(src)/vsyscall.lds $(obj)/vsyscall-%.o FORCE
> > >  	$(call if_changed,syscall)
> > >
> > > -AFLAGS_vsyscall-sysenter.o = -m32
> > > -AFLAGS_vsyscall-syscall.o = -m32
> > > +AFLAGS_vsyscall-sysenter.o = -m32 -Wa,-32
> > > +AFLAGS_vsyscall-syscall.o = -m32 -Wa,-32
> >
> > I can't reproduce it
> 
> What did you try exactly? 

I compared the outputs of vsyscall-sysenter.o with -m32 and -m32
-Wa,-32, using gcc 3.4. They are identical.

> 
> You have to recompile the file and either run the pthreads
> cancellation test case with the right glibc version, or examine
> the unwind tables in the .so by hand. Or just check if the warning
> appears or not.

There is no need to do that. If you believe the problem is due to
binutils 2.16.91.0.5, it is a binutils bug. Please open a bug report
at

http://www.sourceware.org/bugzilla/

and assign it to me. I need the preprocessed .s file, the expected .o
file and the expected .so file with procedures to reproduce the
difference. Since they are assembly codes, I will expect -m32 and
-m32 -Wa,-32 should generate the same output.

> 
> > with binutils 2.16.91.0.5. Unless you have a bad 
> > gcc driver, -m32 will pass --32 to assembler, which is the same as -32.
> 
> It definitely fixed the problem here and the warning went away.
> gcc used was 4.1
> 
> But you're right it could have been the gcc - in the case where it worked
> I was using gcc 4.0.
> 

Since they are assembly codes, it sounds like a gcc driver issue to me.


H.J.

Re: [discuss] RE: [PATCH] [1/2] x86_64: Pass -32 to the assembler when compiling the 32bit vsyscall pages

[Andi Kleen]

On Saturday 22 April 2006 01:32, H. J. Lu wrote:

> > You have to recompile the file and either run the pthreads
> > cancellation test case with the right glibc version, or examine
> > the unwind tables in the .so by hand. Or just check if the warning
> > appears or not.
> 
> There is no need to do that. If you believe the problem is due to
> binutils 2.16.91.0.5, it is a binutils bug. Please open a bug report
> at

I'm not sure - it is reasonable for binutils to expect --32 to be passed
And the linker did warn about it after all - just I made the mistake
of ignoring it before.

The only reason I mentioned binutils is that it didn't happen with 
older binutils/gcc releases.

> 
> http://www.sourceware.org/bugzilla/
> 
> and assign it to me. I need the preprocessed .s file, the expected .o
> file and the expected .so file with procedures to reproduce the
> difference. Since they are assembly codes, I will expect -m32 and
> -m32 -Wa,-32 should generate the same output.

I don't have time to extract a test case out of this right now, sorry.
I guess you could though.

> 
> > 
> > > with binutils 2.16.91.0.5. Unless you have a bad 
> > > gcc driver, -m32 will pass --32 to assembler, which is the same as -32.
> > 
> > It definitely fixed the problem here and the warning went away.
> > gcc used was 4.1
> > 
> > But you're right it could have been the gcc - in the case where it worked
> > I was using gcc 4.0.
> > 
> 
> Since they are assembly codes, it sounds like a gcc driver issue to me.

Might be. The way the assembly is built is a bit funky because it's a 
shared library.

-Andi