[PATCH 0/3] Introduce devm_clk_register()

[Russell King - ARM Linux]

On Mon, Sep 24, 2012 at 01:20:42PM -0700, Stephen Boyd wrote:
> On 09/22/12 03:06, Russell King - ARM Linux wrote:
> > On Tue, Sep 18, 2012 at 11:05:27PM -0700, Stephen Boyd wrote:
> >> The first patch in this series fixes error checking in the wm831x clock
> >> driver and is here to prevent context conflicts in the third patch.
> >> I split it out in case it needed to merge sooner rather than later.
> >>
> >> The goal of this series is to add devm_clk_register() so I can use it in
> >> some MSM clock code I'm sending out in the near future. The second 
> >> patch adds the API and the third patch moves over an existing user of
> >> clk_unregister() to the devm API.
> > Can we guarantee that the clocks are unused when the module is removed?
> > If we can't make that guarantee, then devm_* should not be used here,
> > and instead there should be refcounting done in the clocks (that's what
> > the __clk_get() and __clk_put() hooks are there for.)
> 
> We could guarantee that when clk_unregister() is actually implemented.
> __clk_get() would need to forward a call to the module providing the
> clock via try_module_get(). Similarly we would call module_put() in
> __clk_put(). That would prevent unbinding the driver from the device via
> module removal.

Strangely enough... mach-integrator's clkdev.h... precisely because it
has a module which may provide a clock.

static inline int __clk_get(struct clk *clk)
{
        return try_module_get(clk->owner);
}

static inline void __clk_put(struct clk *clk)
{
        module_put(clk->owner);
}