Re: [PATCH 5.12 000/242] 5.12.18-rc1 review

Hi Greg,

Greg Kroah-Hartman [off-list ref] writes:

On Mon, Jul 19, 2021 at 07:40:21AM +0200, Sven Schnelle wrote:

quoted

If https://gitlab.com/Linaro/lkft/mirrors/stable/linux-stable-rc/-/jobs/1428532107
is the logfile for this problem, than i see the following in the log:

make --silent --keep-going --jobs=8 O=/home/tuxbuild/.cache/tuxmake/builds/current ARCH=s390 CROSS_COMPILE=s390x-linux-gnu- 'CC=sccache s390x-linux-gnu-gcc' 'HOSTCC=sccache gcc'
/bin/sh: 1: /builds/linux/arch/s390/kernel/vdso64/gen_vdso_offsets.sh: Permission denied

However, in the patch this script is 755, and other architecture are
using this for a while now - can you check what the permission are when
you're trying to build the kernel?

Yes, the problem is that when handling patches, we can not change the
permissions on files.  That causes this file to not be added with
execute permissions.  This has generally been considered a bad thing
anyway, and other scripts that relied on being executable have been
changed over time to not be that way and be explicitly run by the
calling script.

Hmm, right. I didn't thought about patches. So i'll adjust the patch and
sent it again. I guess prefixing it with $(CONFIG_SHELL) is then the way
to do it, at least i see a lot of location doing it that way.

But it looks like th gen_vdso_offsets.sh script has not been changed on
any arch to do that yet.  It is one of the few hold outs.

Also, this feels like a HUGE change for a stable tree, adding new
features like this, are you sure it's all needed?

Yes. This fixes syscall restarting in combination with signals coming in
on s390, which is broken since the conversion to generic entry. There's
no easy way to fix that, and we don't want to introduce another
workaround that is userspace visible, as we would have to carry that
forever.

Sven