Re: [PATCH v2] sched/rt: Fix race in push_rt_task

[Phil Auld <pauld@redhat.com>]

On Mon, Feb 17, 2025 at 11:54:09AM -0500 Steven Rostedt wrote:
> On Fri, 14 Feb 2025 17:08:44 +0000
> Harshit Agarwal <harshit@nutanix.com> wrote:
> 
> > Co-developed-by: Jon Kohler <jon@nutanix.com>
> > Signed-off-by: Jon Kohler <jon@nutanix.com>
> > Co-developed-by: Gauri Patwardhan <gauri.patwardhan@nutanix.com>
> > Signed-off-by: Gauri Patwardhan <gauri.patwardhan@nutanix.com>
> > Co-developed-by: Rahul Chunduru <rahul.chunduru@nutanix.com>
> > Signed-off-by: Rahul Chunduru <rahul.chunduru@nutanix.com>
> > Signed-off-by: Harshit Agarwal <harshit@nutanix.com>
> > Tested-by: Will Ton <william.ton@nutanix.com>
> > ---
> > Changes in v2:
> > - As per Steve's suggestion, removed some checks that are done after
> >   obtaining the lock that are no longer needed with the addition of new
> >   check.
> > - Moved up is_migration_disabled check.
> > - Link to v1:
> >   https://lore.kernel.org/lkml/20250211054646.23987-1-harshit@nutanix.com/
> > ---
> >  kernel/sched/rt.c | 54 +++++++++++++++++++++++------------------------
> >  1 file changed, 26 insertions(+), 28 deletions(-)
> 
> 
> Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
>

We've got some cases that look to be hitting this as well.

I'm a little concerned about turning some runtime checks into
BUG_ON()s but in this case I think we are really just going to
trap out on !has_pushable_tasks() check first and if not, pick
a different task and don't drop the lock so it should pass the
BUG_ON()s and fail to match the original task.  So...

Reviewed-by: Phil Auld <pauld@redhat.com>



Cheers,
Phil

> Peter or Ingo,
> 
> Care to take his patch
> 
> Thanks,
> 
> -- Steve
> 

--