Re: [PATCH v3 4/5] clocksource: Enable clocksource_cyc2ns() to cover big cycles

[Feng Tang <feng.tang@intel.com>]

On Wed, Mar 06, 2013 at 05:10:53PM +0100, Thomas Gleixner wrote:
> On Wed, 6 Mar 2013, Feng Tang wrote:
> > Hi Thomas,
> > 
> > Thanks for the reviews.
> > 
> > On Wed, Mar 06, 2013 at 03:09:26PM +0100, Thomas Gleixner wrote:
> > > On Wed, 6 Mar 2013, Feng Tang wrote:
> > > 
> > > > Current clocksource_cyc2ns() has a implicit limit that the (cycles * mult)
> > > > can not exceed 64 bits limit. Jason Gunthorpe proposed a way to
> > > > handle this big cycles case, and this patch put the handling into
> > > > clocksource_cyc2ns() so that it could be used unconditionally.
> > > 
> > > Could be used if it wouldn't break the world and some more.
> > 
> > Exactly.
> > 
> > One excuse I can think of is usually the clocksource_cyc2ns() will be called
> > for cycles less than 600 seconds, based on which the "mult" and "shift" are
> > calculated for a clocksource.
> 
> That's not an excuse for making even the build fail on ARM and other
> 32bit archs.

That's a huge mistake I made in my patch, and I didn't meant to excuse for it :)

> > 
> > trying to avoid expensieve maths. But as Jason pointed, there is some accuracy
> > lost. 
> 
> Right, but if you precalculate the max_fast_cycles value you can avoid
> at least the division in the fast path and then do
> 
>    if (cycles > max_fast_cycles)
>       return clocksource_cyc2ns_slow();
>    return ((u64) cycles * mult) >> shift;	      
> 
> clocksource_cyc2ns_slow() should be out of line and there you can do
> all the slow 64 bit operations. That keeps the fast path sane and we
> don't need extra magic for the large cycle values.

Yeah! This should well cover all possilbe cycles and solve the fast/slow
problem. Thanks. Will try to make a new patch.

- Feng