Re: [PATCH] mm: larger stack guard gap, between vmas

[Ben Hutchings <ben@decadent.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 2725 bytes --]

On Wed, 2017-06-21 at 11:47 +0100, Ben Hutchings wrote:
> On Wed, 2017-06-21 at 11:24 +0200, Michal Hocko wrote:
> > On Wed 21-06-17 02:38:21, Ben Hutchings wrote:
> > > On Mon, 2017-06-19 at 16:23 +0200, Willy Tarreau wrote:
> > > > On Mon, Jun 19, 2017 at 08:44:24PM +0800, Linus Torvalds wrote:
> > > > > The distros are in a different situation and don't have that
> > > > > two-week
> > > > > window until a release, and presumably would not want to cut
> > > > > over to
> > > > > something new and fairly untested on such short notice.
> > > > > 
> > > > > The timing for this all sucks, but if somebody has some final
> > > > > comments, please speak up now..
> > > > 
> > > > What do you suggest the stable maintainers do here ? I've just
> > > > backported
> > > > this patch back to 3.10 and could boot it on i386 where it
> > > > apparently
> > > > works. But we may need more tests. On the other hand we benefit
> > > > from the
> > > > automated tests on tens of platforms when we push the queues so
> > > > at least
> > > > we'll quickly know if it builds and boots. I just don't feel
> > > > confident in
> > > > my work just because it builds and boots, you know.
> > > > 
> > > > I'm appending the patches I currently have if anyone wants to
> > > > have a
> > > > glance. Ben, 3.2 requires much more changes than 3.10 and I'm
> > > > pretty
> > > > sure you won't change your patches at the last minute so I gave
> > > > up.
> > > 
> > > Well I'm now dealing with fall-out from the Debian stable updates,
> > > which used a backport of Michal's patch series.  That unfortunately
> > > seems to break programs running Java code in the main thread (the
> > > 'java' command doesn't do this, but e.g. 'jsvc' does).
> > 
> > Could you share more details please?
> 
> https://bugs.debian.org/865303
> https://bugs.debian.org/865311
> https://bugs.debian.org/865343

Unfortunately these regressions have not been completely fixed by
switching to Hugh's fix.

Firstly, some Rust programs are crashing on ppc64el with 64 KiB pages. 
Apparently Rust maps its own guard page at the lower limit of the stack
(determined using pthread_getattr_np() and pthread_attr_getstack()).  I
don't think this ever actually worked for the main thread stack, but it
now also blocks expansion as the default stack size of 8 MiB is smaller
than the stack gap of 16 MiB.  Would it make sense to skip over
PROT_NONE mappings when checking whether it's safe to expand?

Secondly, LibreOffice is crashing on i386 when running components
implemented in Java.  I don't have a diagnosis for this yet.

Ben.

-- 
Ben Hutchings
The world is coming to an end.	Please log off.


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 833 bytes --]