[Pkg-xen-devel] Bug#900746: xen toolstack xl causes a Segmentation fault on create domu

[Stephen Gelman]

On Thu, 28 Jun 2018 18:58:24 +0200 Hans van Kranenburg <hans at knorrie.org> wrote:
> Hi,
>
> On 06/28/2018 09:58 AM, Damian Pietras wrote:
> > I've also hit it on one of my boxes with
> > 4.8.3+xsa267+shim4.10.1+xsa267-1+deb9u8
> >
> > This is related to too small stack size set for threads in XEN utils
> > which explicitly set it to use 16KB. Similar issue is reported here for
> > NTP: https://bugzilla.redhat.com/show_bug.cgi?id=1564527
> >
> > I've recompilled the package with the attached patch to increase the
> > stack size from 16KB to 32KB and it works.
>
> Thanks for doing "bug triaging"!
>
> > Technical details:
> >
> > The issue appears with modern CPU that support AVX-512 instruction set,
> > in my case it's Intel(R) Xeon(R) Gold 6148. More details are in this
> > bug report against glibc: https://bugzilla.redhat.com/show_bug.cgi?id=1
> > 527887#c18
> >
> > There was a post on xen-users acknowledging the bug that says it's
> > fixed in XEN 4.11: https://lists.xenproject.org/archives/html/xen-users
> > /2018-05/msg00034.html
>
> Following that post and what happened afterwards leads me to upstream
> commit 448c03b3cb "tools/xenstore: try to get minimum thread stack size
> for watch thread", which seems to solve this problem without hardcoding
> some size.
>
> Maybe this should be a nice candidate for upstream backport to stable
> branches, since users are buying newer hardware and otherwise cannot use
> Debian Stable without recompiling their Xen packages?
>
> Ian?
>
> Thanks,
> Hans

This patch fixed segfault problems for us on servers with Intel Xeon
Gold 6148 CPUs.  What would it take to get this patch included in the
next stretch update?  It's a pretty big deal for us right now and
while we'll be ok running this patched version for now it'd be great
to not have to maintain our own patch.

Thanks,

Stephen