Re: [PATCH V2 1/3] lib: add support for stmp-style devices

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 21 Mar 2012 23:21:57 +0100 Wolfram Sang <w.sang@pengutronix.de> wrote:

> MX23/28 use IP cores which follow a register layout I have first seen on
> STMP3xxx SoCs. In this layout, every register actually has four u32:
> 
>  1.) to store a value directly
>  2.) a SET register where every 1-bit sets the corresponding bit,
>      others are unaffected
>  3.) same with a CLR register
>  4.) same with a TOG (toggle) register
> 
> Also, the 2 MSBs in register 0 are always the same and can be used to reset
> the IP core.
> 
> All this is strictly speaking not mach-specific (but IP core specific) and,
> thus, doesn't need to be in mach-mxs/include. At least, mx6 and mx50 also
> utilize IP cores following this stmp-style. So:
> 
> Introduce a stmp-style device, put the code and defines for that in a public
> place (lib/), and let drivers for stmp-style devices select that code.
> To avoid regressions and ease reviewing, the actual code is simply copied from
> mach-mxs. It definately wants updates, but those need a seperate patch series.
> 
> Voila, mach dependency gone, reusable code introduced. Note that I didn't
> remove the duplicated code from mach-mxs yet, first the drivers have to be
> converted.
> 
> ...
>
>  include/linux/stmp_device.h |   20 +++++++++++
>  lib/Kconfig                 |    3 ++
>  lib/Makefile                |    2 +
>  lib/stmp_device.c           |   80 +++++++++++++++++++++++++++++++++++++++++++

It's good that this is being presented as library code, rather than
being buried in some device-specific directory then copied and pasted
ten times.

But ./lib/ does seem rather a strange place for it.  Perhaps we need a
drivers/lib/ or something.  We can use ./lib/ for now - it can always
be moved later on.

> --- /dev/null
> +++ b/lib/stmp_device.c

The functions in this file look terribly racy on SMP, or even with
preemption or interrupts.  What happens if two CPUs or threads run
stmp_reset_block() against the same device at the same time?

Perhaps the caller is supposed to prevent that, and the documentation
which isn't there forgot to mention it ;)