Re: [PATCH v3 08/10] clk: ns2: add clock support for Broadcom Northstar 2 SoC

[Jon Mason <jonmason@broadcom.com>]

On Wed, Oct 21, 2015 at 05:22:47PM -0700, Stephen Boyd wrote:
> On 10/15, Scott Branden wrote:
> > On 15-10-15 02:15 PM, Ray Jui wrote:
> > >On 10/15/2015 2:10 PM, Jon Mason wrote:
> > >>On Thu, Oct 15, 2015 at 02:04:09PM -0700, Scott Branden wrote:
> > >>>On 15-10-15 01:55 PM, Ray Jui wrote:
> > >>>>On 10/15/2015 1:40 PM, Scott Branden wrote:
> > >>>>
> > >>>>If using CONFIG_CLK_NS2, how is it going to be enabled/selected?
> > >>>
> > >>>Since CONFIG_ARCH_BCM_NS2 isn't "allowed" to be introduced we will
> > >>>need to create and select a CONFIG_CLK_BCM_NS2 in the defconfig
> > >>>instead.
> > >>
> > >>Is this better than the binary becoming slightly bigger?  I thought
> > >>the extra complexity was worse than having an unused chunk of clk code
> > >>(and Kona is already doing the same thing above).  I believe Ray was
> > >>in agreement with me during the internal review of this code.
> > >>
> > >>Thanks,
> > >>Jon
> > >>
> > >
> > >Yes, I'm okay with leaving it as it is. I even prefer changing the
> > >current Makefile to make all iProc based core clock drivers and SoC
> > >specific clock tables under CONFIG_COMMON_CLK_IPROC, which is what some
> > >of the other vendors do.
> > >
> > I'd leave it exactly as is then rather than pulling in more dead
> > code when not needed.  This ns2 clock code is very minor compared to
> > other code bloat in the kernel and drivers.
> 
> We should really make these visible options that can be selected
> by anyone. Having selects in the ARCH config area is simple, but
> also has some downsides:
> 
>  1) select is a reverse dependency and is hard for people to
>  understand and can sometimes be a pain to track down
> 
>  2) build coverage goes down because configs are hidden
> 
>  3) we get code bloat like is being discussed here
> 
> So I'd really like to see someone take a good look at this whole
> Makefile situation that's going on and clean it up so that
> they're user visible options and then throw the config options
> into the defconfig. It isn't going to block this series, but it
> would be nice to do at some later point.

I'll do a pass at this today and send it out for review.

Thanks,
Jon

> 
> -- 
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> a Linux Foundation Collaborative Project