Re: BUG_ON(!newowner) in fixup_pi_state_owner()

[Mike Galbraith <efault@gmx.de>]

On Wed, 2020-11-04 at 01:56 +0100, Mike Galbraith wrote:
> On Tue, 2020-11-03 at 17:31 -0600, Gratian Crisan wrote:
> > Hi all,
> >
> > I apologize for waking up the futex demons (and replying to my own
> > email), but ...
> >
> > Gratian Crisan writes:
> > >
> > > Brandon and I have been debugging a nasty race that leads to
> > > BUG_ON(!newowner) in fixup_pi_state_owner() in futex.c. So far
> > > we've only been able to reproduce the issue on 4.9.y-rt kernels.
> > > We are still testing if this is a problem for later RT branches.
> >
> > I was able to reproduce the BUG_ON(!newowner) in fixup_pi_state_owner()
> > with a 5.10.0-rc1-rt1 kernel (currently testing 5.10.0-rc2-rt4).
>
> My box says it's generic.

---
 kernel/futex.c |   13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -2383,7 +2383,18 @@ static int fixup_pi_state_owner(u32 __us
 		 * Since we just failed the trylock; there must be an owner.
 		 */
 		newowner = rt_mutex_owner(&pi_state->pi_mutex);
-		BUG_ON(!newowner);
+
+		/*
+		 * Why? Because I know what I'm doing with these beasts?  Nope,
+		 * but what the hell, a busy restart loop let f_boosted become
+		 * owner, so go for it. Box still boots, works, no longer makes
+		 * boom with fbomb_v2, and as an added bonus, didn't even blow
+		 * futextests all up.  Maybe it'll help... or not, we'll see.
+		 */
+		if (unlikely(!newowner)) {
+			err = -EAGAIN;
+			goto handle_err;
+		}
 	} else {
 		WARN_ON_ONCE(argowner != current);
 		if (oldowner == current) {