Re: [PATCH 01/16] move copyout of timespec into do_cpu_nanosleep()

[Thomas Gleixner]

On Tue, 13 Jun 2017, Thomas Gleixner wrote:
> On Wed, 7 Jun 2017, Al Viro wrote:
> > @@ -1242,12 +1243,14 @@ static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
> >  
> >  		error = -ERESTART_RESTARTBLOCK;
> > +		/*
> > +		 * Report back to the user the time still remaining.
> > +		 */
> > +		rmtp = restart->nanosleep.rmtp;
> > +		if (rmtp) {
> > +			struct timespec ts;
> > +			ts = timespec64_to_timespec(it.it_value);
> > +			if (copy_to_user(rmtp, &ts, sizeof(*rmtp)))
> > +				return -EFAULT;
> > +		}
> > +		restart->nanosleep.expires = timespec64_to_ns(rqtp);
> 
> Hmm, this looks wrong. But that was wrong before this patch already.
> 
> The restart code needs the absolute expiry time of the timer, but in case
> of flags != ABSTIME rqtp contains the relative expiry time.
> 
> So that want's to be fixed first. Need to figure out what's the least ugly
> way to do that. I'll have a look tomorrow.

Misread the code. rqtp is updated with the absolute expiry time when a
signal hits. Though with your changes we can simplify that further and
avoid the whole back and forth timespec conversion.

Thanks,

	tglx