Messages in this thread | | | From | Mike Turquette <> | Subject | Re: [PATCH] clk: shmobile: div6: Avoid division by zero in .round_rate() | Date | Thu, 05 Feb 2015 09:46:10 -0800 |
| |
Quoting Laurent Pinchart (2015-02-05 09:19:14) > Hi Sergei, > > On Thursday 05 February 2015 01:14:46 Sergei Shtylyov wrote: > > On 02/05/2015 01:04 AM, Sergei Shtylyov wrote: > > >>>>> Anyone may call clk_round_rate() with a zero rate value, so we have to > > >>>>> protect against that. > > >>>>> > > >>>>> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> > > >>>> > > >>>> Acked-by: Wolfram Sang <wsa+renesas@sang-engineering.com> > > >>>> > > >>>> I agree that this should not be fixed in the core because the fixup is > > >>>> really driver dependant. > > >>>> > > >>> Dunno, zero frequency seems generally insane to me. > > >> > > >> It is useful to find the lowest frequency a clock can support. Basically > > >> it is a search for the floor frequency. > > >> > > > Why not just use 1? Or are you assuming that some hardware could actually > > > support 0 Hz? > > > > Replying to myself: yes, this has happened to me, when I forgot to override > > the EXTAL frequency in the board .dts file (default was 0). > > So it was a good thing that the driver crashed, it let you find a bug ;-) > > Jokes aside, a zero frequency is the usual way to find the lowest frequency, > but I agree that there aren't many integers between 0 and 1. Mike, do you have > an opinion ?
Yes, I think we should support passing a zero rate for two reasons:
1) it's crazy to not sanitize a value that is passed into a function and used as a divisor. This is not really a shortcoming of the framework.
2) during the fractional divider discussion there was the idea of making unsigned long rate into something like millihertz. E.g. rate = 1000 is 1Hz. If we start cheating by passing a rate of 1 into .round_rate, then we've just created a bug for ourselves if we ever move to millihertz.
Regards, Mike
> > -- > Regards, > > Laurent Pinchart >
| |