Re: [PATCH 1/2] remoteproc: maintain a generic child device for each rproc

[Stephen Boyd <sboyd@codeaurora.org>]

(Sorry your mail was lost due to mail outage)

On 05/30/12 05:16, Ohad Ben-Cohen wrote:
>
> On Wed, May 30, 2012 at 11:36 AM, Stephen Boyd <sboyd@codeaurora.org> wrote:
>
>> One complaint I've gotten is that the error messages are essentially
>> useless now. I believe there are some ongoing discussions on lkml to fix
>> this by traversing the device hierarchy to find the "real" device but
>> the hard part is finding the real device.
> You probably refer to the discussions around the input subsystem's pull request.
>
> I was thinking about that too when creating this patch, and it looks
> like whatever Greg will come up with on that matter will benefit us
> too. So taking that into account, it might make more sense to do stick
> with the virtual device rather than use the real one here (we'll end
> up having more information in the long run).

Fair enough. Hopefully something comes out of that discussion since this
will need it.

>> I'm not clear on busses versus classes.
> I think that busses is a whole lot more complex beast. Probably the
> main indication we want one is when we need to match drivers to
> devices.
>
> In this case, I was more wondering between using a class to a device type.
>
>> I recall seeing a thread where
>> someone said classes were on the way out and shouldn't be used but I
>> can't find it anymore.
> I also remembered a similar discussion at a plumbers mini-conf about
> 2-3 years ago too, so I looked at device_type as an alternative to
> class. The former looks somewhat simpler, but I couldn't find any
> major advantage for using one over the other, and both seem to be in
> use by many subsystems.
>
>> Should we use classes for devices that will never
>> have a matching driver?
> It's not strictly required, but in case we want to provide these
> devices some common behavior (and in our case we want them all to have
> the same release handler, and very soon, the same PM handlers, too),
> then a class (or a type) is helpful.
>
> It looks like moving from a class to a type is quite trivial, in case
> classes do eventually go away (or an advantage of using the latter
> shows up), but I'm not aware of any other viable alternative for us
> other than class/type.
>

Ok. Will moving from a class to a device type disrupt the kernel ABI?
First it will be under /sys/class/ and then under /sys/bus? Greg, can
you shed some light on when to use a class versus a device type?

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.