On 01/11/2011 06:35 AM, Jeremy Kerr wrote:

Hi Paul,

quoted

Again, you are approaching it from the angle that an atomic clock is a
special requirement rather than the default behaviour.

I'm not considering it a special requirement, but it's still a requirement
(that the called function does not sleep).

The problem with the inverse logic (clk_enable/clk_enable_sleepable) is that
now you've made the caller need to know what kind of clock it has, or might
have one day.

I think it's just a matter of how you interpret the name of the API in 
English. It doesn't make the decision making of the developer any easier.

Just having a _atomic suffix doesn't mean the driver developer doesn't 
need to know what type of clock it is. They are still making the 
assumption that the enable/disable for that clock can be done atomically 
-- namely an "atomic clock".

Similarly, when a driver developer calls the _sleepable APIs in their 
code, for all practical purposes, they are making an assumption that the 
enable/disable for that clock *needs to* (not may) sleep.

  * For clk_enable/clk_enable_atomic, the decision is: is this call in an
    atomic context?

  * For clk_enable/clk_enable_sleepable, the decision is: might the clock code
    have given us a sleeping clock?

Having said the above, I'm slightly leaning towards 
clk_enable/disable_atomic since it lines up with the 
.suspend/.suspend_noirq functions in pm_ops.

Also, since it's good to reduce the amount of work that needs to be done 
atomically, I think it would be good to make a developer explicitly 
state they need _atomic functions and make them think about if they 
really need to do that.

-Saravana

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