[PATCH] gpio/omap: ensure gpio context is initialised

[Jon Hunter]

On 04/18/2013 03:22 AM, Santosh Shilimkar wrote:
> On Thursday 18 April 2013 02:01 AM, Jon Hunter wrote:
>> Commit a2797be (gpio/omap: force restore if context loss is not
>> detectable) broke gpio support for OMAP when booting with device-tree
>> because a restore of the gpio context being performed without ever
>> initialising the gpio context. In other words, the context restored was
>> bad.
>>
>> This problem could also occur in the non device-tree case, however, it
>> is much less likely because when booting without device-tree we can
>> detect context loss via a platform specific API and so context restore
>> is performed less often.
>>
>> Nevertheless we should ensure that the gpio context is initialised
>> during the probe for gpio banks that could lose their state regardless
>> of whether we are booting with device-tree or not.
>>
>> Reported-by: Tony Lindgren <tony at atomide.com>
>> Signed-off-by: Jon Hunter <jon-hunter at ti.com>
>> Tested-by: Tony Lindgren <tony at atomide.com>
>> ---
>>  drivers/gpio/gpio-omap.c |   53 +++++++++++++++++++++++++++++++++++++---------
>>  1 file changed, 43 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/gpio/gpio-omap.c b/drivers/gpio/gpio-omap.c
>> index 0557529..0ba5cb9 100644
>> --- a/drivers/gpio/gpio-omap.c
>> +++ b/drivers/gpio/gpio-omap.c
>> @@ -70,6 +70,7 @@ struct gpio_bank {
>>  	bool is_mpuio;
>>  	bool dbck_flag;
>>  	bool loses_context;
>> +	bool context_valid;
>>  	int stride;
>>  	u32 width;
>>  	int context_loss_count;
>> @@ -1085,6 +1086,7 @@ static void omap_gpio_chip_init(struct gpio_bank *bank)
>>  }
>>  
>>  static const struct of_device_id omap_gpio_match[];
>> +static void omap_gpio_init_context(struct gpio_bank *p);
>>  
>>  static int omap_gpio_probe(struct platform_device *pdev)
>>  {
>> @@ -1179,8 +1181,10 @@ static int omap_gpio_probe(struct platform_device *pdev)
>>  	omap_gpio_chip_init(bank);
>>  	omap_gpio_show_rev(bank);
>>  
>> -	if (bank->loses_context)
>> +	if (bank->loses_context) {
>>  		bank->get_context_loss_count = pdata->get_context_loss_count;
>> +		omap_gpio_init_context(bank);
>> +	}
>>  
>>  	pm_runtime_put(bank->dev);
>>  
>> @@ -1269,6 +1273,14 @@ static int omap_gpio_runtime_resume(struct device *dev)
>>  	int c;
>>  
>>  	spin_lock_irqsave(&bank->lock, flags);
>> +
>> +	/*
>> +	 * On the first resume during the probe, the context has not
>> +	 * been initialised and so if the context is not valid return.
>> +	 */
>> +	if (!bank->context_valid)
>> +		goto done;
>> +
> The use of 'context_valid' is just for the probe resume case and hence
> adding a state variable for GPIO bank structures looks bit too much.

Yes it is not ideal. However, it is safe.

> Ideally gpio_init_context() should make the first callback successful
> as well if ordered correctly but then you need clocks for that
> to happen which are enabled in runtime pm calls. 

Right.

> Do you really need that "context_valid" flag per bank ? I mean
> every gpio bank which can "loses_context", probe time context
> validity is an issue, right ? In that case, won't just an
> __init static variable do for all banks which can loose
> context?

I think that it would be fragile to have a global for all banks. What
happens if one bank failed during the probe? Probably unlikely but still.

The best solution would be to be able to plug the pm-runtime
resume/suspend handlers at the end of the probe or be able to enable the
module without the callback handlers being calling during the probe. I
am not sure if there is a way to do this. May be Kevin knows.

Another option is to move the initialisation of the bank->loses_context
to after the call to pm_runtime_get_sync() in the probe. This would
prevent the context restore occurring during the initial runtime resume.
However, if you look at the runtime resume function there are still some
register writes that do occur (which is basically what happens today).
This works but it is a little fragile, however, we can do that if
preferred. May be I could add a comment to the probe to say why we
initialise bank->loses_context after the pm_runtime_get_sync() so no one
moves this in the future (and more importantly I don't forget!).

Cheers
Jon