On Sun, Sep 15, 2019 at 07:07:52AM -0700, Paul E. McKenney wrote:
> On Sat, Sep 14, 2019 at 07:33:58AM -0500, Eric W. Biederman wrote:
> >
> > In the ordinary case today the rcu grace period for a task_struct is
> > triggered when another process wait's for it's zombine and causes the
Oh, and "waits for its", just to hit the grammar en passant... ;-)
Thanx, Paul
> > kernel to call release_task(). As the waiting task has to receive a
> > signal and then act upon it before this happens, typically this will
> > occur after the original task as been removed from the runqueue.
> >
> > Unfortunaty in some cases such as self reaping tasks it can be shown
> > that release_task() will be called starting the grace period for
> > task_struct long before the task leaves the runqueue.
> >
> > Therefore use put_task_struct_rcu_user in finish_task_switch to
> > guarantee that the there is a rcu lifetime after the task
> > leaves the runqueue.
> >
> > Besides the change in the start of the rcu grace period for the
> > task_struct this change may cause perf_event_delayed_put and
> > trace_sched_process_free. The function perf_event_delayed_put boils
> > down to just a WARN_ON for cases that I assume never show happen. So
> > I don't see any problem with delaying it.
> >
> > The function trace_sched_process_free is a trace point and thus
> > visible to user space. Occassionally userspace has the strangest
> > dependencies so this has a miniscule chance of causing a regression.
> > This change only changes the timing of when the tracepoint is called.
> > The change in timing arguably gives userspace a more accurate picture
> > of what is going on. So I don't expect there to be a regression.
> >
> > In the case where a task self reaps we are pretty much guaranteed that
> > the rcu grace period is delayed. So we should get quite a bit of
> > coverage in of this worst case for the change in a normal threaded
> > workload. So I expect any issues to turn up quickly or not at all.
> >
> > I have lightly tested this change and everything appears to work
> > fine.
> >
> > Inspired-by: Linus Torvalds <[email protected]>
> > Inspired-by: Oleg Nesterov <[email protected]>
> > Signed-off-by: "Eric W. Biederman" <[email protected]>
> > ---
> > kernel/fork.c | 11 +++++++----
> > kernel/sched/core.c | 2 +-
> > 2 files changed, 8 insertions(+), 5 deletions(-)
> >
> > diff --git a/kernel/fork.c b/kernel/fork.c
> > index 9f04741d5c70..7a74ade4e7d6 100644
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -900,10 +900,13 @@ static struct task_struct *dup_task_struct(struct
> > task_struct *orig, int node)
> > if (orig->cpus_ptr == &orig->cpus_mask)
> > tsk->cpus_ptr = &tsk->cpus_mask;
> >
> > - /* One for the user space visible state that goes away when reaped. */
> > - refcount_set(&tsk->rcu_users, 1);
> > - /* One for the rcu users, and one for the scheduler */
> > - refcount_set(&tsk->usage, 2);
> > + /*
> > + * One for the user space visible state that goes away when reaped.
> > + * One for the scheduler.
> > + */
> > + refcount_set(&tsk->rcu_users, 2);
>
> OK, this would allow us to add a later decrement-and-test of
> ->rcu_users ...
>
> > + /* One for the rcu users */
> > + refcount_set(&tsk->usage, 1);
> > #ifdef CONFIG_BLK_DEV_IO_TRACE
> > tsk->btrace_seq = 0;
> > #endif
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 2b037f195473..69015b7c28da 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -3135,7 +3135,7 @@ static struct rq *finish_task_switch(struct
> > task_struct *prev)
> > /* Task is done with its stack. */
> > put_task_stack(prev);
> >
> > - put_task_struct(prev);
> > + put_task_struct_rcu_user(prev);
>
> ... which is here. And this looks to be invoked from the __schedule()
> called from do_task_dead() at the very end of do_exit().
>
> This looks plausible, but still requires that it no longer be possible to
> enter an RCU read-side critical section that might increment ->rcu_users
> after this point in time. This might be enforced by a grace period
> between the time that the task was removed from its lists and the current
> time (seems unlikely, though, in that case why bother with call_rcu()?) or
> by some other synchronization.
>
> On to the next patch!
>
> Thanx, Paul
>
> > }
> >
> > tick_nohz_task_switch();
> > --
> > 2.21.0.dirty
> >
