Re: [PATCH v2 0/8] watchdog: w83627hf: Convert to watchdog infrastructure

[Guenter Roeck <linux@roeck-us.net>]

On Sat, Mar 23, 2013 at 01:57:42PM +0100, Wim Van Sebroeck wrote:
> Hi Guenter,
> 
> > > In 2011 I started something similar but then with the MFD approach in mind.
> > > Goal was also to clean-up the w836* watchdog drivers and get a clean driver that
> > > supports all Winbond super-I/O based watchdog drivers.
> > > 
> > > I dug op the development code again. I'll post it in a next e-mail so that we can
> > > see what the best way forward is. Note: I took the MFD approach because:
> > > 1) all superio shares the similar functions for using the Super-I/O registers.
> > > 2) Goal is to have low-level driver that support the specific super-I/O chipsets
> > > and that does the platform stuff for hwmon, watchdog, gpio, ...
> > > 
> > Hi Wim,
> > 
> > I started with a similar approach, only I used mfd cells to pass on platform
> > specific information such as the device type and the superio base address.
> > I still have the patchset for the mfd driver, in case you are interested.
> > My code is based on the patches submitted by Rodolfo Giometti a couple
> > of years ago. Want me to post it ?
> 
> If it's not v1 then I am interested.
> I think it depends on the super-I/O chipset of what info you can pass. I would
> not use mfd cells for the winbond driver but use platform data (similar to
> drivers/mfd/adp5520.c) because you don't need more then just pass some data.
> 
Actually one can pass platform data through mfd cells, which is what I do.
I pass the SIO address, the chip type (as enum), and the chip name to the
platform driver.

> > What I noticed in my testing is that the superio address range (2e or 4e),
> > which the drivers currently take as granted, is at least on my systems
> > (all three of them) reserved by ACPI. Unfortunately that means one can not
> > use the mfd infrastructure to pass on the superio memory region,
> > since it checks for acpi conflicts. With that I gave up on the idea and
> > reverted to using request_muxed_region. That seemed simpler and accomplish
> > the same as long as all drivers actually use it.
> 
> Noticed the same. That's why passing the platform data together with the
> superio-address and type seems the best way to go. I also kept the superio_enter
> and superio_exit a function in the low level driver because I used a lock to
> make sure that the hwmon code doesn't start doing things when the watchdog is
> doing things. But I think that the request_muxed_region is doing something similar
> also.
> 
Yes, it does the same. Only difference is that using a function in the mfd
driver increases the use count on it when a client module is loaded.

Patch follows.

Thanks,
Guenter