Locking in the clk API, part 2: clk_prepare/clk_unprepare

On Tue, Feb 01, 2011 at 05:11:29PM +0800, Jeremy Kerr wrote:

Hi all,

quoted

I suggested that clk_prepare() be callable only from non-atomic contexts,
and do whatever's required to ensure that the clock is available.  That
may end up enabling the clock as a result.

I think that clk_prepare/clk_unprepare looks like the most promising solution, 
so will try to get some preliminary patches done. Here's what I'm planning:

-----

The changes to the API are essentially:

1) Document clk_enable/clk_disable as callable from atomic contexts, and
   so clock implementations must not sleep within this function.

2) For clock implementations that may sleep to turn on a clock, we add a
   new pair of functions to the clock API: clk_prepare and clk_unprepare.

   These will provide hooks for the clock implmentation to do any sleepable
   work (eg, wait for PLLs to settle) in preparation for a later clk_enable.

   For the most common clock implemntation cases (where clocks can be enabled 
   atomically), these functions will be a no-op, and all of the enable/disable
   work can be done in clk_enable/clk_disable.

   For implementations where clocks require blocking on enable/disable, most
   of the work will be done in clk_prepare/clk_unprepare. The clk_enable
   and clk_disable functions may be no-ops.

For drivers, this means that clk_prepare must be called (and have returned) 
before calling clk_enable.

== Enable/Prepare counts ==

I intend to do the enable and prepare "counting" in the core clock API, 
meaning that that the clk_ops callbacks will only invoked on the first 
prepare/enable and the last unprepare/disable.

== Concurrency ==

Splitting the prepare and enable stages introduces the concurrency 
requirements:

1) clk_enable must not return before the clock is outputting a valid clock 
   signal.

2) clk_prepare must not return before the clock is fully prepared (ie, it is 
   safe to call clk_enable).

It is not possible for clk_enable to wait for the clock to be prepared, 
because that would require synchronisation with clk_prepare, which may then 
require blocking. Therefore:

3) The clock consumer *must* respect the proper ordering of clk_prepare and 
   clk_enable. For example, drivers that call clk_enable during an interrupt 
   must ensure that the interrupt handler will not be invoked until 
   clk_prepare has returned.

== Other considerations ==

The time that a clock spends "prepared" is a superset of the the time that a 
clock spends "enabled". Therefore, clocks that are switched on during 
clk_prepare (ie, non-atomic clocks) will be running for a larger amount of 
time. In some cases, this can be mitigated by moving some of the final 
(atomic) switching functionality to the clk_enable function.

== Implementation ==

Basically:

struct clk {
	const struct clk_ops *ops
	int                  enable_count;
	spinlock_t           enable_lock;
	int                  prepare_count;
	struct mutex         prepare_lock;
};

int clk_enable(struct clk *clk)
{
	int ret = 0;

	spin_lock(&clk->enable_lock);
	if (!clk->enable_count)
		ret = clk->ops->enable(clk);

	if (!ret)
		clk->enable_count++;
	spin_unlock(&clk->enable_lock);

	return ret;
}

Why do we not call parent's clk_enable in this function? For flexible? How many
different cases is causing us to move the effert to platform clock driver?

int clk_prepare(struct clk *clk)
{
	int ret = 0;

	mutex_lock(&clk->prepare_lock);
	if (!clk->prepare_count)
		ret = clk->ops->prepare(clk);

	if (!ret)
		clk->prepare_count++;
	mutex_unlock(&clk->prepare_lock);

	return ret;
}

Same as above.
And for most clocks, prepare/unprepare may be NULL. So in such case, is it
better to call parent's prepare and increase its own prepare_count here?

Thanks
Richard

-----

Comments welcome, code coming soon.

Cheers,


Jeremy


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel