Re: [PATCH] psi: avoid divide-by-zero crash inside virtual machines

[Johannes Weiner]

On Thu, Feb 14, 2019 at 11:58:55AM -0800, Andrew Morton wrote:
> On Thu, 14 Feb 2019 14:31:57 -0500 Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> 
> > --- a/kernel/sched/psi.c
> > +++ b/kernel/sched/psi.c
> > @@ -322,7 +322,7 @@ static bool update_stats(struct psi_group *group)
> >  	expires = group->next_update;
> >  	if (now < expires)
> >  		goto out;
> > -	if (now - expires > psi_period)
> > +	if (now - expires >= psi_period)
> >  		missed_periods = div_u64(now - expires, psi_period);
> >  
> >  	/*
> 
> It seems appropriate to use time_after64() and friends in this code.

These timestamps are all sourced from sched_clock(), which is defined
to be monotonic and never wrap in practice. From the "sched_clock()"
section in Documentation/timers/timekeeping.txt:

	"This function shall return the number of nanoseconds since
	 the system was started."

	"The sched_clock() function may wrap only on unsigned long
	 long boundaries, i.e. after 64 bits. Since this is a
	 nanosecond value this will mean it wraps after circa 585
	 years. (For most practical systems this means "never".)"

As far as readability goes, I have to say I find the naked comparisons
a bit easier to understand (and I'm glad we can use those here since
the code is already complicated):

	if (now < expires)

vs.

	if (time_before64(now, expires))

These macros always have me double check the argument order.