Re: [PATCH] ibmaem: Fix 64-bit division on 32-bit platforms

["Carlos R. Mafra" <crmafra2@gmail.com>]

On Thu 15.May'08 at  1:27:04 +0300, Adrian Bunk wrote:
> On Thu, May 08, 2008 at 09:55:27PM -0700, Darrick J. Wong wrote:
> > On Thu, May 08, 2008 at 04:34:43PM -0700, Andrew Morton wrote:
> > 
> > > This driver is littered with 64-bit divides and doesn't link on i386.
> > > I'll make it depend on CONFIG_64BIT for now.
> > 
> > Oops, sorry, I didn't remember that one can't do 64-bit division on
> > i386.  The patch below fixes that.
> > ---
> > ibmaem: Fix 64-bit division on 32-bit platforms
> > 
> > Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
> > ---
> > 
> >  drivers/hwmon/ibmaem.c |    6 ++++--
> >  1 files changed, 4 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/hwmon/ibmaem.c b/drivers/hwmon/ibmaem.c
> > index 22fa7d6..f808ca3 100644
> > --- a/drivers/hwmon/ibmaem.c
> > +++ b/drivers/hwmon/ibmaem.c
> >...
> > @@ -864,9 +865,10 @@ static ssize_t aem_show_power(struct device *dev,
> >...
> >  	time = timespec_to_ns(&a) - timespec_to_ns(&b);
> > -	time /= 1000;
> > +	time = div_u64(time, 1000);
> >  
> > -	return sprintf(buf, "%llu\n", (after - before) * 1000000000 / time);
> > +	return sprintf(buf, "%llu\n",
> > +		       div64_u64((after - before) * 1000000000, time));
> >  }
> >...
> 
> What are you actually trying to do here?
> 
> Converting mJ/ns to uJ/s?
> 
> Or do you want to convert mJ/ns to Watt in which case you should
> multiply "time" with 1000 instead of dividing it through 1000?

To avoid these kind of doubts may I suggest using the time 
conversion definitions of include/linux/time.h?

It would be much easier to understand the intention with
NSEC_PER_SEC, USEC_PER_SEC etc (comments would also help)

> In any case you should only need one expensive 64bit division
> and you can fold the operation with 1000 into the second division.

I agree.