On Wed, 2023-11-15 at 09:51 +0100, Peter Zijlstra wrote:
> On Wed, Nov 15, 2023 at 01:11:20AM +0000, Edgecombe, Rick P wrote:
> > On Tue, 2023-11-14 at 21:14 +0100, Peter Zijlstra wrote:
> > > Urgh, thanks!
> > >
> > > Confirmed, the below cures things. Although I should probably
> > > make
> > > that
> > > FLAGS_SIZE_32 | FLAGS_SHARED against Linus' tree.
> > >
> > > Let me go do a proper patch.
> >
> > I saw these fail on the glibc shadow stack branch today, and I also
> > saw
> > this one failing:
> > FAIL: nptl/tst-robustpi8
>
> tip/locking/urgent (branch with the fix on) gets me:
>
> root@ivb-ep:/usr/local/src/glibc# ./build/nptl/tst-robustpi8
> running child
> verifying locks
> running child
> verifying locks
> running child
> verifying locks
> running child
> verifying locks
> running child
> verifying locks
> root@ivb-ep:/usr/local/src/glibc#
>
> Which, to my untrained eye, looks like a pass to me.
It bisects to this for me:
fbeb558b0dd0 ("futex/pi: Fix recursive rt_mutex waiter state")
Reading the patch, I'm not immediately clear what is going on but a few
comments stood out: "There be dragons here" "What could possibly go
wrong..." "This is a somewhat dangerous proposition".
Seems a likelihood of some race, but it reproduces reliably on my
machine. Haven't dug into debugging it yet. Any pointers?
