Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 17 May 2015 19:20:25 +0200
From: Szabolcs Nagy <>
Subject: Re: ppc soft-float regression

* Rich Felker <> [2015-05-17 12:35:21 -0400]:
> On Sun, May 17, 2015 at 10:03:21AM +0200, Waldemar Brodkorb wrote:
> > Hi,
> > 
> > as mentioned the days on IRC.
> > 1.1.9 produces a lot of segfaults on Qemu-PPC with Soft-Float.
> > 1.1.8 was fine.
> Soft-float has never been a supported configuration for PowerPC, as
> you can see from the fact that there's no separate dynamic linker name
> for it. I'm surprised it ever seemed to work -- surely setjmp/longjmp
> would be broken since they save/restore FPU registers. I don't think
> it would be hard to add though, and I'd welcome patches for it. What's
> needed is basically:
> - detection in configure, setting $SUBARCH there
> - dynamic linker name variants in arch/powerpc/reloc.h
> - separate subarch dir for soft-float in src/setjmp
> If you add soft-float it would probably make sense to add
> little-endian variant at the same time, if that's useful to anyone,
> since once you do the subarch work there's hardly any more work to
> make an endian variant too.

note that the current musl-cross powerpc patches don't do
the include reordering so the gcc includes will be the
first in the path.  this is fixed in

if there is a ppc soft-float abi musl may support then please
invent an abi name for it so i can update the gcc patch before
it gets committed.

(i think powerpc{el}{-sf} and powerpc64{el}{-sf} should work:
it is consistent with the naming for mips).

attached the musl dynamic linker name patch for "el" support,
similar would be needed for -sf (+setjmp.s etc updates).

View attachment "powerpcel.diff" of type "text/x-diff" (834 bytes)

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.