[PATCH v2 1/2] Input: tegra-kbc - add device tree bindings

[Grant Likely]

On Thu, Dec 29, 2011 at 01:23:08AM -0800, Dmitry Torokhov wrote:
> On Wed, Dec 28, 2011 at 10:00:06PM -0800, Stephen Warren wrote:
> > Dmitry Torokhov wrote at Tuesday, December 27, 2011 11:49 PM:
> > > On Tue, Dec 27, 2011 at 10:19:29PM -0800, Olof Johansson wrote:
> > > > This adds a simple device tree binding to the tegra keyboard controller.
> > ...
> > > > +	pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
> > > 
> > > Error handling is missing. I also dislike devm_* facilities as it causes
> > > inconsistencies in the way we handle releasing of resources: some of
> > > them will be released automatically while others need t be released
> > > manually. I prefer having consistent model.
> > 
> > I have to say that I also disagree here. Weren't the devm_* function
> > specifically added to allow people not to write all the error-handling
> > code themselves.
> 
> I guess they were. Unfortunately they also introduce inconsistency in
> the way resources are tracked and freed since some of them support
> devm_* facilities while others don't. I might be OK with drivers that
> use one model for all resources but this patch was mixing the two.
> 
> devm_* facilities are also not free; you are saving on error handling
> code but pay with memory footprint required to implement tracking.

FWIW, I agree with Olof and Stephen.  The devm_* functions are a good thing
and I strongly encourage driver authors to use them.  The cost in memory
footprint is pretty negligible, but making it easier for driver authors to
get things right is huge.

g.