[PATCH v2, RFC] RTC: PXA: Fix regression of interrupt before ioremap

[Petr Cvek]

On 2.2.2015 19:33, Robert Jarzmik wrote:
> Petr Cvek <petr.cvek at tul.cz> writes:
>
>> I agree that driver without .open looks ugly, but only thing in rtc-pxa .open
>> were two request_irq and I don't think it is good idea to have them there
>> (interrupts should be disabled trough register settings and not by handler
>> freeing).
>>
>> I'm not familiar with the linux RTC subsystem, so I don't know if it is OK to
>> get interrupt (and rtc_update_irq) without opened /dev/rtc. Intuitively I have
>> feeling it is OK, but even if not disabling can be done with some register flag.
>>
>> BTW It seems that kernel have only around 9 drivers in drivers/rtc which contain
>> .open function.
>>
>> OT: rtc-sa1100 seems to be compatible with PXAxxx (it is even in Kconfig). Are
>> there any reasons to have two drivers for one SoC?
> Yes, there is a reason :
>    http://marc.info/?l=linux-arm-kernel&m=122306289606732&w=2
>
> At that time we decided this were 2 different IPs (more or less) sharing the
> same IO region and IRQ. 2 IPs for pxa27x and greater, only 1 IP for pxa25x and
> lower.
>
> Now you should know that both rtc-sa1100 and rtc-pxa should be able to work
> together in the same kernel (at least that was the case so far). The open()
> decided who got a grip on the interrupt. This lets userland choose which rtc it
> relies on : either the increasing count, or the
> day/month/year/hour/minute/second counters (which are independant).
>

I see, thanks for info, I never thought that these two drivers were 
specially crafted to be able to coexist (if it is good idea I can write 
info for Kconfig help section).

I still don't think it is a good idea to do "mutex" by racing who gets 
both request_irq first, but I don't have better solution other than 
making both drivers fully independent on each other or merging them 
together (probably with checkbox for enabling enhanced features in PXA27x).

Actually only thing I want to know after reverting a44802f is how wakeup 
will work. Because a44802f suggests rtc-pxa needs to have interrupt 
enabled for waking up (and I cannot test it, because suspend subsystem 
on my machine needs to be fixed first).

> Moreover, if there are multiple rtc device, how on earth can it work, ie. how
> can an ioctl() be sent to a specific rtc device if there is no open() ???

It confuses me too, so I tried to look it up and it seems rtc_dev_open() 
in drivers/rtc/rtc-dev.c handles this by:

	err = ops->open ? ops->open(rtc->dev.parent) : 0;
     if (err == 0) {
         spin_lock_irq(&rtc->irq_lock);
         rtc->irq_data = 0;
         spin_unlock_irq(&rtc->irq_lock);

         return 0;
     }

, so without any .open() it just continues with success.

Petr