Message-ID: <5538C6C2.5010508@linaro.org>
Date:	Thu, 23 Apr 2015 13:17:38 +0300
From:	"Grygorii.Strashko@...aro.org" <grygorii.strashko@...aro.org>
To:	Nishanth Menon <nm@...com>,
	Alexandre Belloni <alexandre.belloni@...e-electrons.com>,
	Alessandro Zummo <a.zummo@...ertech.it>
CC:	linux-omap@...r.kernel.org, linux-kernel@...r.kernel.org,
	rtc-linux@...glegroups.com
Subject: Re: [PATCH V2] drivers/rtc/rtc-ds1307.c: Enable the mcp794xx alarm
 after programming time

On 04/23/2015 03:00 AM, Nishanth Menon wrote:
> On 04/22/2015 08:26 AM, Grygorii.Strashko@...aro.org wrote:
>> Hi,
>>
>> On 04/21/2015 03:51 AM, Nishanth Menon wrote:
>>> Alarm interrupt enable register is at offset 0x7, while the time
>>> registers for the alarm follow that. When we program Alarm interrupt
>>> enable prior to programming the time, it is possible that previous
>>> time value could be close or match at the time of alarm enable
>>> resulting in interrupt trigger which is unexpected (and does not match
>>> the time we expect it to trigger).
>>>
>>> To prevent this scenario from occuring, program the ALM0_EN bit only
>>> after the alarm time is appropriately programmed.
>>>
>>> Ofcourse, I2C programming is non-atomic, so there are loopholes where
>>> the interrupt wont trigger if the time requested is in the past at
>>> the time of programming the ALM0_EN bit. However, we will not have
>>> unexpected interrupts while the time is programmed after the interrupt
>>> are enabled.
>>
>> I think it will be nice if you will mention that you going to follow
>> vendor recommendations - AN1491 Configuring the MCP794XX RTCC Family
>> http://ww1.microchip.com/downloads/en/AppNotes/01491A.pdf
>> ;)
>> "Also, it is recommended that the alarm registers be loaded
>> before the alarm is enabled."
>>
> 
> Hmm... i did not know that existed, thanks for digging it up.. that
> teaches me to look for docs before putting a scope/LA on the board
> (not that I regret doing that)... That said, reading the app note, I
> kind of realized:
> a) that playing with ST bit for programming time is not done, but
> then, that implies that oscillator will have to be restarted (upto a
> few seconds for certain crystals).. but that said, it does not seem
> mandatory or seem to (yet seen) functional issues...
> 
> b) We dont have flexibility yet to describe if we do indeed have a
> backup battery or not - VBATEN should be set only if we have a backup
> battery on the platform :( - on some it might even be optional thanks
> to certain compliance requirements of shipping boards internationally
> and general "unlike" of lithium ion in cargo hold..
> 
> c) we dont have capability to control the alarm polarity in the driver
> which, by the way, we probably should also control OUT polarity (when
> ALARM is not asserted)..
> 
> d) we dont have support for external 32k oscillator(X1 only) instead
> of assuming we always have a 32k crystal(X1 and X2)...
> 
> Ugghhh... more cleaning up to do for the future..
> 
> that said, the sequence it does recommend (in page 4):
> The following steps show how the Alarm 0 is config-
> ured. Alarm 1 can be configured in the same manner.
> 1.    Write 0x23 to the Alarm0 Seconds register
> [0x0A].
> 2.    Write 0x47 to the Alarm0 Minutes register
> [0x0B].
> 3.    Write 0x71 to the Alarm0 Hours register [0x0C]
> – 11 hours in 12-hour format.
> 4.    Write 0x72 to the Alarm0 Day register [0x0D] –
> Tuesday + Alarm Polarity Low + Match on all.
> The Alarm0 Interrupt Flag is also cleared.
> 5.    Write 0x14 to the Alarm0 Date register [0x0E].
> 6.    Write 0x08 to the Alarm0 Month register [0x0F].
> With all the Alarm0 registers set we can now activate
> the Alarm0 on the Control register.
> 7.    Write 0x10 to the Control register [0x07] –
> Alarm0 enabled no CLKOUT, Alarm1 disabled
> 
> before this patch we do ( http://pastebin.ubuntu.com/10863880/)
> 	CONTROL r[7] = 0x90 (OUT=1, ALM0EN=1)
> 	OSCTRIM r[8] = 0x00
> 	EEUNLOCK r[9] = 0x00
> 	ALM0SEC r[A] = 0x01
> 	ALM0MIN r[B] = 0x45
> 	ALM0HOUR r[C] = 0x23
> 	ALM0WKDAY r[D] = 0x75 <-ALMOIF is cleared
> 	ALM0DATE r[E] = 0x09
> 	ALM0MTH r[F] = 0x04
> 	RSRVED r[10] = 0x01
> 
> with this patch, we do:
> burst(	CONTROL r[7] = 0x80 (OUT=1)
> 	OSCTRIM r[8] = 0x00
> 	EEUNLOCK r[9] = 0x00
> 	ALM0SEC r[A] = 0x01
> 	ALM0MIN r[B] = 0x45
> 	ALM0HOUR r[C] = 0x23
> 	ALM0WKDAY r[D] = 0x75 <-ALMOIF is cleared
> 	ALM0DATE r[E] = 0x09
> 	ALM0MTH r[F] = 0x04
> 	RSRVED r[10] = 0x01
> )
> 	CONTROL r[7] = 0x90 (OUT=1, ALM0EN=1)
> 
> Which is slightly unoptimal way of what the app note recommends. - as
> I mentioned earlier in this thread, I will try and do optimizations in
> a later patch.
> 
> Given that Andrew had picked up this patch, I dont see a reason to
> respin this yet. but will include the app note for future patches -
> thanks for pointing it out to me.

^^ Up to you. Np, Always yours!

>>>
>>> Signed-off-by: Nishanth Menon <nm@...com>
>>> ---
>>> Changes in v2:
>>> 	- minor typo fix in comments
>>> 	- merged up code that I missed committing in
>>>
>>> V1: https://patchwork.kernel.org/patch/6245041/
>>>
>>>    drivers/rtc/rtc-ds1307.c |   12 ++++++------
>>>    1 file changed, 6 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/drivers/rtc/rtc-ds1307.c b/drivers/rtc/rtc-ds1307.c
>>> index 4ffabb322a9a..3cd4783375a5 100644
>>> --- a/drivers/rtc/rtc-ds1307.c
>>> +++ b/drivers/rtc/rtc-ds1307.c
>>> @@ -742,17 +742,17 @@ static int mcp794xx_set_alarm(struct device *dev, struct rtc_wkalrm *t)
>>>    	regs[6] &= ~MCP794XX_BIT_ALMX_IF;
>>>    	/* Set alarm match: second, minute, hour, day, date, month. */
>>>    	regs[6] |= MCP794XX_MSK_ALMX_MATCH;
>>> -
>>> -	if (t->enabled)
>>> -		regs[0] |= MCP794XX_BIT_ALM0_EN;
>>> -	else
>>> -		regs[0] &= ~MCP794XX_BIT_ALM0_EN;
>>> +	/* Disable interrupt. We will not enable until completely programmed */
>>> +	regs[0] &= ~MCP794XX_BIT_ALM0_EN;
>>>    
>>>    	ret = ds1307->write_block_data(client, MCP794XX_REG_CONTROL, 10, regs);
>>>    	if (ret < 0)
>>>    		return ret;
>>>    
>>> -	return 0;
>>> +	if (!t->enabled)
>>> +		return 0;
>>> +	regs[0] |= MCP794XX_BIT_ALM0_EN;
>>> +	return i2c_smbus_write_byte_data(client, MCP794XX_REG_CONTROL, regs[0]);
>>
>> So, It seems, that right sequence should be:
>> - disable alarmX
>> - read alarmX regs
>> - configure alarmX regs
>> - load alarmX regs
>> - enable alarmX
> Not exactly.... see above. we can optimize this for a better sequence
> as follows - since there are already un-necessary reads being
> performed. probably just a couple of reads might be
> sufficient..(ALM0WKDAY has some control bits as well.. Ugggh..
> anyways..)...
> 
> 
> Will have to think more about optimizing more later.


Also I've done some fast investigation and I found that ~half of RTC drivers disable
 ALM IRQ before start accessing Alarm regs (twl-rtc.c)  while another half don't do that :)
(just FYI)

> 
>>
>> More over, looks like, alarm/alarm IRQ should be enabled/disabled separately from set_alarm/RTC_ALM_SET
>> by RTC_AIE_ON, RTC_AIE_OFF. Should it?
> 

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

Powered by blists - more mailing lists