Re: [RFC][PATCH v3 1/3] runtime interpreted power sequences

[Alex Courbot <acourbot@nvidia.com>]

On 07/31/2012 07:45 AM, Stephen Warren wrote:
>> +- Delay to wait before performing the action,
>> +- Delay to wait after performing the action.
>
> I don't see a need to have a delay both before and after an action;
> except at the start of the sequence, step n's post-delay is at the same
> point in the sequence as step n+1's pre-delay. Perhaps make a "delay"
> step type?

My first version used this actually - and you're right, having a "delay" 
step type would be more flexible and less redundant.

>> +Both new resources and parameters can be introduced, but the goal is of course
>> +to keep things as simple and compact as possible.
>
>> +The platform data is a simple array of platform_power_seq_step instances, each
>
> Rather than jumping straight into platform data here, I'd expect an
> enumeration of legal resource types, and what actions can be performed
> on each, followed by a description of a sequence (very simply, just a
> list of actions and their parameters). This could be followed by a
> section describing the mapping of the abstract concepts to concrete
> platform data representation (and concrete device tree representation).

Keeping that in mind for the next revision.

>> +instance describing a step. The type as well as one of id or gpio members
>> +(depending on the type) must be specified. The last step must be of type
>> +POWER_SEQ_STOP.
>
> I'd certainly suggest having a step count rather than a sentinel value
> in the list.

As Thierry did - I think I will go that way.

>> Regulator and PWM resources are identified by name. GPIO are
>> +identified by number.
>
> That's a little implementation-specific. I guess it's entirely true for
> a platform data representation, but not when mapping this into device tree.

If we can come with a way to properly use phandles within DT sequences 
(and we should), then this will only apply to platform data.

>> +You will need an instance of power_seq_resources to keep track of the resources
>> +that are already allocated. On success, the function returns a devm allocated
>> +resolved sequence that is ready to be passed to power_seq_run(). In case of
>> +failure, and error code is returned.
>
> If the result is devm-allocated, the function probably should be named
> devm_power_seq_build().

Right - more generally this needs to have both devm and non-devm variants.

> I wonder if using the same structure/array as input and output would
> simplify the API; the platform data would fill in the fields mentioned
> above, and power_seq_build() would parse those, then set other fields in
> the same structs to the looked-up handle values?

The thing is that I am not sure what happens to the platform data once 
probe() is done. Isn't it customary to mark it with __devinit and have 
it freed after probing is successful?

More generally, I think it is a good practice to have data structures 
tailored right for what they need to do - code with members that are 
meaningful only at given points of an instance's life tends to be more 
confusing.

> You can make a custom devm free routine for the power_seq_resources
> itself, so the overall free'ing of the content can be triggered by devm,
> but the free'ing function can then call whatever non-devm APIs it wants
> for the non-devm-allocated members.

That sounds good.

>> +Device tree
>> +-----------
>> +All the same, power sequences can be encoded as device tree nodes. The following
>> +properties and nodes are equivalent to the platform data defined previously:
>> +
>> +		power-supply = <&mydevice_reg>;
>> +		enable-gpio = <&gpio 6 0>;
>> +
>> +		power-on-sequence {
>> +			regulator@0 {
>
> As Thierry mentioned, the step nodes should be named for the type of
> object they are (a "step") not the type or name of resource they act
> upon ("regulator" or "gpio").

Will fix that.

> If the nodes have a unit address (i.e. end in "@n"), which they will
> have to if all named "step" and there's more than one of them, then they
> will need a matching reg property. Equally, the parent node will need
> #address-cells and #size-cells too. So, the last couple lines would be:
>
> 		power-on-sequence {
> 			#address-cells = <1>;
> 			#size-cells = <0>;
> 			step@0 {
> 				reg = <0>;

That's precisely what I would like to avoid - I don't need the steps to 
be numbered and I certainly have no use for a reg property. Isn't there 
a way to make it simpler?

>> +				id = "power";
>
> "id" is usually a name or identifier. I think you mean "type" or perhaps
> "action" here:
>
> 				type = "regulator";
> 				action = "enable";
>
> or:
>
> 				action = "enable-regulator";

Right, that was a clear misuse.

> Oh I see. That's a little confusing. Why not just reference the relevant
> resources directly in each step; something more like:
>
> 		gpio@1 {
> 			action = "enable-gpio";
> 			gpio = <&gpio 1 0>;
> 		};
>
> I guess that might make parsing/building a little harder, since you'd
> have to detect when you'd already done gpio_request() on a given GPIO
> and not repeat it or something like that, but to me this makes the DT a
> lot easier to comprehend.

You can see my reply to Thierry for the reason - the only issue with 
that is caused by PWM phandles. If we overcome this, then I agree we 
should use phandles. The code should not even get more complex as I have 
to check whether a resource is already allocated with strings as well.

Thanks,
Alex.