Re: [PATCH] clk: fractional-divider: cast parent_rate to u64 before multiplying

["Heiko Stübner" <heiko@sntech.de>]

Am Montag, 1. September 2014, 17:26:29 schrieb Mike Turquette:
> Quoting Heiko Stübner (2014-08-28 03:46:10)
> 
> > On 32bit architectures, like ARM calculating the fractional rate will
> > do the multiplication before converting the value to u64 when it gets
> > assigned to ret, which can produce overflows.
> > 
> > The error in question happened with a parent_rate of 386MHz, m = 3000,
> > n = 60000, which resulted in a wrong rate value of 15812Hz.
> > 
> > Therefore cast parent_rate to u64 to make sure the multiplication
> > happens in a 64bit space and produces the correct 192MHz in the example.
> > 
> > Signed-off-by: Heiko Stuebner <heiko@sntech.de>
> 
> Looks good to me. Have you observed this issue using the vanilla
> kernel.org sources or on a downstream tree? I'm trying to decide if I
> should take this into clk-fixes or clk-next.
> 
> If this doesn't happen in practice on any merged platform then I'll take
> it for 3.18.

It happens on the Rockchip platform with the newly added fractional branches 
you just applied ... so it doesn't look like anybody else is affected right 
now.


Heiko


> 
> Regards,
> Mike
> 
> > ---
> > 
> >  drivers/clk/clk-fractional-divider.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/clk/clk-fractional-divider.c
> > b/drivers/clk/clk-fractional-divider.c index ede685c..82a59d0 100644
> > --- a/drivers/clk/clk-fractional-divider.c
> > +++ b/drivers/clk/clk-fractional-divider.c
> > @@ -36,7 +36,7 @@ static unsigned long clk_fd_recalc_rate(struct clk_hw
> > *hw,> 
> >         m = (val & fd->mmask) >> fd->mshift;
> >         n = (val & fd->nmask) >> fd->nshift;
> > 
> > -       ret = parent_rate * m;
> > +       ret = (u64)parent_rate * m;
> > 
> >         do_div(ret, n);
> >         
> >         return ret;