[PATCH v2 1/2] OMAP: omap_device: Add omap_device_[alloc|delete] for DT integration

[grant.likely@secretlab.ca (Grant Likely)]

On Sat, Sep 17, 2011 at 05:27:01PM +0100, Russell King - ARM Linux wrote:
> On Sat, Sep 17, 2011 at 10:05:14AM -0600, Grant Likely wrote:
> > > +static struct omap_device *omap_device_alloc(struct platform_device *pdev,
> > > +					struct omap_hwmod **ohs, int oh_cnt,
> > > +					struct omap_device_pm_latency *pm_lats,
> > > +					int pm_lats_cnt)
> > > +{
> > > +	int ret = -ENOMEM;
> > > +	struct omap_device *od;
> > > +	struct resource *res = NULL;
> > > +	int i, res_count;
> > > +	struct omap_hwmod **hwmods;
> > > +
> > > +	od = kzalloc(sizeof(struct omap_device), GFP_KERNEL);
> > 
> > possible enhancement:  devm_kzalloc() perhaps?  Would simplify the cleanup
> > paths.
> 
> Are you sure about that - have you thought about the lifetime issues?
> devm_*() lifetime is supposed to be from driver binding to driver
> unbinding.
> 
> The lifetime of 'omap_device' appears to be from device creation to
> device destruction, which is different from what devm_*() gives you.
> 
> So, using devm_*() will result in the 'omap_device' being destroyed
> when a driver is unbound - and presumably an oops if its attempted to
> be re-bound later (because the 'omap_device' structure will have been
> freed.)

Ugh, yes.  You're absolutely right.  devm_*() is doesn't at all work here.

g.