PowerPC Port: Question about magic numbers in sigsetjmp.s

PowerPC Port: Question about magic numbers in sigsetjmp.s

[Dan Parrot]

I'm trying to modify the PowerPC port in order to run it on a 64-bit
Little Endian system. I haven't been able to get pthreads to pass the
tests in libc-testsuite and I believe the culprit to be sigsetjmp.s.

I would be grateful for an explanation of the offsets added to register
3 in these two lines in sigsetjmp.s:

stw 5, 448(3)
stw 16, 448+4+8(3)


Thanks.
Dan.

Re: PowerPC Port: Question about magic numbers in sigsetjmp.s

[Bobby Bingham]

On Sat, Apr 30, 2016 at 12:55:19PM -0500, Dan Parrot wrote:
> I'm trying to modify the PowerPC port in order to run it on a 64-bit
> Little Endian system. I haven't been able to get pthreads to pass the

I've already submitted a patch to the mailing list to add a ppc64le port.
http://www.openwall.com/lists/musl/2016/03/27/8

I'm about to submit a new version to address the remaining comments that
have been made on it, but it should be close to done.

> tests in libc-testsuite and I believe the culprit to be sigsetjmp.s.
> 
> I would be grateful for an explanation of the offsets added to register
> 3 in these two lines in sigsetjmp.s:
> 
> stw 5, 448(3)
> stw 16, 448+4+8(3)
> 

See this commit: http://git.musl-libc.org/cgit/musl/commit/?id=583e55122e767b1586286a0d9c35e2a4027998ab

The sigjmp_buf structure includes a __jmp_buf at the beginning (see the
definition in include/setjmp.h).

__jmp_buf is 448 bytes large on ppc32 (it should be larger on ppc64).
So, these are accessing the part of the sigjmp_buf beyond the initial
__jmp_buf.

The first line is saving the return address to the __fl member, and the
second line is saving r16 to __ss[2].  The first two elements of __ss
are used by __sigsetjmp_tail to save the signal mask.  The remainder of
__ss is unused, so it's a convenient place to spill a register to.

> 
> Thanks.
> Dan.
> 
>

Bobby

Re: PowerPC Port: Question about magic numbers in sigsetjmp.s

[Dan Parrot]

Thanks for the quick response.

I'll take a look at the patch and comments.

Dan.

-----Original Message-----From: Bobby Bingham <koorogi@koorogi.info>
Reply-to: musl@lists.openwall.com
To: musl@lists.openwall.com
Subject: Re: [musl] PowerPC Port: Question about magic numbers in
sigsetjmp.s
Date: Sat, 30 Apr 2016 14:20:48 -0500

On Sat, Apr 30, 2016 at 12:55:19PM -0500, Dan Parrot wrote:
> I'm trying to modify the PowerPC port in order to run it on a 64-bit
> Little Endian system. I haven't been able to get pthreads to pass the

I've already submitted a patch to the mailing list to add a ppc64le port.
http://www.openwall.com/lists/musl/2016/03/27/8

I'm about to submit a new version to address the remaining comments that
have been made on it, but it should be close to done.

> tests in libc-testsuite and I believe the culprit to be sigsetjmp.s.
> 
> I would be grateful for an explanation of the offsets added to register
> 3 in these two lines in sigsetjmp.s:
> 
> stw 5, 448(3)
> stw 16, 448+4+8(3)
> 

See this commit: http://git.musl-libc.org/cgit/musl/commit/?id=583e55122e767b1586286a0d9c35e2a4027998ab

The sigjmp_buf structure includes a __jmp_buf at the beginning (see the
definition in include/setjmp.h).

__jmp_buf is 448 bytes large on ppc32 (it should be larger on ppc64).
So, these are accessing the part of the sigjmp_buf beyond the initial
__jmp_buf.

The first line is saving the return address to the __fl member, and the
second line is saving r16 to __ss[2].  The first two elements of __ss
are used by __sigsetjmp_tail to save the signal mask.  The remainder of
__ss is unused, so it's a convenient place to spill a register to.

> 
> Thanks.
> Dan.
> 
>

Bobby