Potential glibc 2.43 regression on big-endian targets

[Adhemerval Zanella Netto]

On 22/08/26 13:16, John Paul Adrian Glaubitz wrote:
> Hi,
> 
> there seems to be a regression in glibc causing fork to fail on
> big-endian targets, observed on powerpc/ppc64 and sparc64.
> 
> On powerpc/ppc64, it has been observed that a simple shell process
> can run out of memory on a system with plentyful memory [1]:
> 
> make[5]: /bin/sh: Cannot allocate memory
> make[5]: *** [/build/reproducible-path/linux-7.2~rc7/scripts/Makefile.modinst:124: /build/reproducible-path/linux-7.2~rc7/debian/build/build_ppc64_none_powerpc64/output/image-dbg/lib/modules/7.2-
> powerpc64/kernel/drivers/block/mtip32xx/mtip32xx.ko] Error 127
> make[5]: *** Waiting for unfinished jobs....
> 
> And on sparc64, the shell would just fail with:
> 
> /bin/sh: 1: Cannot fork
> 
> Since this happens on PowerPC and SPARC servers which got just
> upgraded to glibc 2.43, I'm suspecting a glibc regression here.
> 
> Adrian
> 
>> [1] https://buildd.debian.org/status/fetch.php?pkg=linux&arch=ppc64&ver=7.2%7Erc7-1%7Eexp1&stamp=1786539811&raw=0
> 

I run tests on powerpc64/powerpc32 from time to time on cfarm121 and I haven't
see any other than the usual regressions [1][2]. Did it started with 2.43?
Does it still happens on master? Do we have any other extra information of
why malloc has failed?

[1] https://sourceware.org/glibc/wiki/Release/2.43
[2] https://sourceware.org/glibc/wiki/Release/2.44