Re: [PATCH v3] sched/cputime: Protect some other sum_exec_runtime reads on 32 bit cpus

[Stanislaw Gruszka <sgruszka@redhat.com>]

On Tue, Sep 06, 2016 at 09:01:29PM +0200, Peter Zijlstra wrote:
> On Tue, Sep 06, 2016 at 02:49:08PM +0200, Stanislaw Gruszka wrote:
> > diff --git a/kernel/exit.c b/kernel/exit.c
> > index 2f974ae..a46f96f 100644
> > --- a/kernel/exit.c
> > +++ b/kernel/exit.c
> > @@ -134,7 +134,7 @@ static void __exit_signal(struct task_struct *tsk)
> >  	sig->inblock += task_io_get_inblock(tsk);
> >  	sig->oublock += task_io_get_oublock(tsk);
> >  	task_io_accounting_add(&sig->ioac, &tsk->ioac);
> > -	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
> > +	sig->sum_sched_runtime += read_sum_exec_runtime(tsk);
> >  	sig->nr_threads--;
> >  	__unhash_process(tsk, group_dead);
> >  	write_sequnlock(&sig->stats_lock);
> 
> If I'm not mistaken, at this point @p is dead, sum_exec_runtime will not
> ever be updated again.

I think at this point find_task_by_vpid() will still return not-null
pointer, hence we can update sum_exec_runtime via:

posix_cpu_{clock_get(),timer_set(),timer_get()} ->
cpu_clock_sample() -> task_sched_runtime() -> update_curr()

if some thread will do cpuclock/cputimer related syscall for
different thread and CPUCLOCK_SCHED clock.

> > diff --git a/kernel/time/posix-cpu-timers.c b/kernel/time/posix-cpu-timers.c
> > index 39008d7..a2d753b 100644
> > --- a/kernel/time/posix-cpu-timers.c
> > +++ b/kernel/time/posix-cpu-timers.c
> > @@ -848,7 +848,7 @@ static void check_thread_timers(struct task_struct *tsk,
> >  	tsk_expires->virt_exp = expires_to_cputime(expires);
> >  
> >  	tsk_expires->sched_exp = check_timers_list(++timers, firing,
> > -						   tsk->se.sum_exec_runtime);
> > +						   read_sum_exec_runtime(tsk));
> >  
> >  	/*
> >  	 * Check for the special case thread timers.
> 
> @tsk == current and IRQs are disabled, sum_exec_runtime cannot be
> updated.

It still can be updated via syscalls on different thread, same paths
as above.
 
> > @@ -1115,7 +1115,7 @@ static inline int fastpath_timer_check(struct task_struct *tsk)
> >  		struct task_cputime task_sample;
> >  
> >  		task_cputime(tsk, &task_sample.utime, &task_sample.stime);
> > -		task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
> > +		task_sample.sum_exec_runtime = read_sum_exec_runtime(tsk);
> >  		if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
> >  			return 1;
> >  	}
> 
> Same.

Same :-) But this could be optimized to avoid taking lock if not needed:
	
	if (expires->sum_exec_runtime != 0)
		task_sample.sum_exec_runtime = read_sum_exec_runtime(tsk);

Stanislaw