Re: [PATCH 2/2] soc/fsl: add ftm alarm driver for ls1021a platform

[Linus Walleij]

On Fri, Aug 14, 2015 at 5:12 AM, Wang Dongsheng
<Dongsheng.Wang@xxxxxxxxxxxxx> wrote:
>> On Wed, Aug 12, 2015 at 7:53 AM, Dongsheng Wang
>> <dongsheng.wang@xxxxxxxxxxxxx> wrote:
>>
>> > From: Wang Dongsheng <dongsheng.wang@xxxxxxxxxxxxx>
>> >
>> > Only Ftm0 can be used when system going to deep sleep. So this driver
>> > to support ftm0 as a wakeup source.
>> >
>> > Signed-off-by: Wang Dongsheng <dongsheng.wang@xxxxxxxxxxxxx>
>> > ---
>> > *V2*
>> > Change Copyright 2014 to 2015.
>> (...)
>> > +config FTM_ALARM
>> > +       bool "FTM alarm driver"
>> > +       depends on SOC_LS1021A
>> > +       default n
>> > +       help
>> > +         Say y here to enable FTM alarm support.  The FTM alarm provides
>> > +         alarm functions for wakeup system from deep sleep.  There is only
>> > +         one FTM can be used in ALARM(FTM 0).
>> (...)
>> > +static u32 time_to_cycle(unsigned long time)
>> > +static u32 cycle_to_time(u32 cycle)
>> > +static int ftm_set_alarm(u64 cycle)
>> > +static irqreturn_t ftm_alarm_interrupt(int irq, void *dev_id)
>> > +static ssize_t ftm_alarm_show(struct device *dev,
>> > +                             struct device_attribute *attr,
>> > +                             char *buf)
>> > +static ssize_t ftm_alarm_store(struct device *dev,
>> > +                              struct device_attribute *attr,
>> > +                              const char *buf, size_t count)
>> (...)
>> > +static struct device_attribute ftm_alarm_attributes = __ATTR(ftm_alarm, 0644,
>> > +                       ftm_alarm_show, ftm_alarm_store);
>>
>> If you're gonna invent ABIs, document then in Documentation/ABI/testing/*.
>>
>> But I don't get it. Why is this driver not in drivers/rtc?
>>
>> It does a subset of what an RTC does. The ioctl()'s of an RTC
>> can do what you want to do. And much much more.
>>
>> If it can't do all an RTC can do, surely the RTC subsystem
>> can be augmented to host it anyway. It's way to close to
>> an RTC to have it's own random sysfs driver like this.
>>
>> Unless I'm totally off, rewrite this to an RTC driver and post
>> it to the RTC maintainers.
>
> FlexTimer is not a RTC device and not have any rtc deivce function. They belong to
> different devices, why we need to register this to RTC framework? I am confused about this.
>
> Now in freescale layerscape platform this driver is only for FlexTimer0, and not
> fit for each flextimer. Because only FlexTimer0 still turn-on when system in the Deep Sleep.
>
> If the "alarm" make you feel confused or mislead you think this is a RTC devices. I think
> I need to change the "alarm" to "timer".

I think it is an RTC, it is just that the hardware engineer
designed it with a wakeup usecase in mind and did not call
it an RTC. Wakeup is one of the things RTCs do.

If you inspect a few drivers in drivers/rtc such as drivers/rtc/rtc-pl030.c
you will find that they are just as crude as this "alarm" thing.

It has a counter that counts cycles, it has a comparator
and an alarm function. It is an on-chip RTC, just like PL030
no matter what the datasheet or hardware engineer thinks it
should be called, the Linux kernel calls this an RTC, and it
has a subsystem for handling it, so we should use it and
not invent random new stuff.

If the hardware is really so strange that the counter can only
be started if you also put an alarm at the same time (I doubt
it, but OK if you say so) it is a subset of an RTC that can
only be used for alarms but not timekeeping, but it should
*still* live in drivers/rtc.

Think for a moment on the huge effort that John Stultz put into
integrating Android alarm timers with POSIX and the RTC
subsystem and fixing it all from the smallest handset to
the largest S360 supercomputer. The approach of a custom
device just throws all of that out the window and reinvents the
mechanism in userspace, forcing all standardized userspace to
have special code to handle this special alarm with its
special sysfs ABI.

Check
commit ff3ead96d17f47ee70c294a5cc2cce9b61e82f0f
"timers: Introduce in-kernel alarm-timer interface"
for example.

Even if you persist on keeping it in its own magic driver
like this, it should implement the alarm timer interface
from <linux/alarmtimer.h> and I bet after that you don't
need the sysfs files anymore, as the system will sleep
and wake up from the regular syscalls instead of using
magic poking in sysfs from userspace. AFAICT this hardware
is designed for exactly this usecase.

tools/testing/selftests/timers/alarmtimer-suspend.c
is there for you to test your driver with alarmtimer
support.

Needless to say: if you implement it as an RTC you get the
alarmtimer interaction for free. That is why we have the
subsystem after all: to be helpful.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/