Re: [PATCH]: fix 32bits integer overflow in loops_per_jiffy

[Yoann Vandoorselaere]

On Thu, 2002-08-22 at 19:46, Dominik Brodowski wrote:
> On Thu, Aug 22, 2002 at 09:35:16PM +0200, Benjamin Herrenschmidt wrote:
> > >IMHO per-arch functions are really not needed. The only architectures which
> > >have CPUFreq drivers by now are ARM and i386. This will change, hopefully;
> > >IMHO it should be enough to include some basic limit checking in 
> > >cpufreq_scale().
> > 
> > In this specific case, we were talking about PPC since the problem
> > occured when I implemented cpufreq support to switch the speed
> > of the latest powerbooks between 667 and 800Mhz
> 
> And the patch from Yoann solves this? 

Yep, the integer overflow resulted in an incorrectly computed
loops_per_jiffy :

Aug 21 19:50:41 titane kernel: adjust_jiffies: prechange cur=667000, new=800000
Aug 21 19:50:41 titane kernel: old loop_per_jiffy = 665.19 (cpufreq_ref_loops=3325952, cpufreq_ref_freq=667000).
Aug 21 19:50:41 titane kernel: new loop_per_jiffy = 669.02 (cpufreq_ref_loops=3325952, cpufreq_ref_freq=667000).

With the patch applied, it work fine :

Aug 22 11:33:40 titane kernel: adjust_jiffies: prechange cur=667000, new=800000
Aug 22 11:33:40 titane kernel: old loop_per_jiffy = 665.19 (cpufreq_ref_loops=3325952, cpufreq_ref_freq=667000).
Aug 22 11:33:40 titane kernel: new loop_per_jiffy = 797.82 (cpufreq_ref_loops=3325952, cpufreq_ref_freq=667000).

-- 
Yoann Vandoorselaere, http://www.prelude-ids.org

"Programming is a race between programmers, who try and make more and 
 more idiot-proof software, and universe, which produces more and more 
 remarkable idiots. Until now, universe leads the race"  -- R. Cook