Re: [PATCH v2 -next] sched/cputime: Fix mul_u64_u64_div_u64() precision for cputime

[Uwe Kleine-König]

Hello,

On Fri, Jul 26, 2024 at 12:44:29PM +0200, Oleg Nesterov wrote:
> On 07/26, Zheng Zucheng wrote:
> >
> > before call mul_u64_u64_div_u64(),
> > stime = 175136586720000, rtime = 135989749728000, utime = 1416780000.
> 
> So stime + utime == 175138003500000
> 
> > after call mul_u64_u64_div_u64(),
> > stime = 135989949653530
> 
> Hmm. On x86 mul_u64_u64_div_u64(175136586720000, 135989749728000, 175138003500000)
> returns 135989749728000 == rtime, see below.
> 
> Nevermind...
> 
> > --- a/kernel/sched/cputime.c
> > +++ b/kernel/sched/cputime.c
> > @@ -582,6 +582,12 @@ void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
> >  	}
> >  
> >  	stime = mul_u64_u64_div_u64(stime, rtime, stime + utime);
> > +	/*
> > +	 * Because mul_u64_u64_div_u64() can approximate on some
> > +	 * achitectures; enforce the constraint that: a*b/(b+c) <= a.
> > +	 */
> > +	if (unlikely(stime > rtime))
> > +		stime = rtime;
> 
> Thanks,
> 
> Acked-by: Oleg Nesterov <oleg@xxxxxxxxxx>
> 
> -------------------------------------------------------------------------------
> But perhaps it makes sense to improve the accuracy of mul_u64_u64_div_u64() ?

Note there is a patch by Nicolas Pitre currently in mm-nonmm-unstable
that makes mul_u64_u64_div_u64() precise. It was in next for a while as
commit 3cc8bf1a81ef ("mul_u64_u64_div_u64: make it precise always")
which might explain problems to reproduce the incorrect results.

An obvious alternative to backporting this change to
kernel/sched/cputime.c for stable is to backport Nicolas's patch
instead. Andrew asked me to provide a justification to send Nicolas's
patch for inclusion in the current devel cycle. So it might make it in
before 6.11.

Best regards
Uwe

Attachment: signature.asc
Description: PGP signature