Re: Random panic in load_balance() with 3.16-rc

[Markus Trippelsdorf <markus@trippelsdorf.de>]

On 2014.07.28 at 11:28 -0700, Linus Torvalds wrote:
> On Mon, Jul 28, 2014 at 11:09 AM, Markus Trippelsdorf
> <markus@trippelsdorf.de> wrote:
> >
> > It shouldn't be too hard to implement a simple check for the bug in the
> > next release. Just compile the gcc/testsuite/gcc.target/i386/pr61801.c
> > testcase with -fcompare-debug. If gcc returns 0 then
> > -fvar-tracking-assignments could safely be enabled again.
> 
> We don't really have any good infrastructure for things like this,
> though. We probably *should* have a way to generate config options by
> compiler version, but right now we don't. We do random ugly things
> from within Makefile shell escapes (see all the helpers for this we do
> in scripts/Kbuild.include, for example), and we could add yet another
> one. But this is a whole new level of "ugly hack". It would be better
> if we could do things like this at config time, not at build-time with
> Makefile hacks.
> 
> Also, the test-case seems to be very sensitive to compiler options: it
> passes with "-O", but fails with "-O2" or "-Os" for me. So I wonder
> how reliable it is in the face of compiler version differences (ie is
> it really robust wrt the bug actually being *fixed*, or is it a bit of
> a happenstance)

It is robust with -O2 and -Os for all supported series that I've
checked: 4.8, 4.9 and 5.0. I haven't checked older releases.

-- 
Markus