Re: TS-219 RTC issue with Debian Buster

Re: TS-219 RTC issue with Debian Buster

[Uwe Kleine-König]

Hello Oliver,

I added the RTC maintainers and list to Cc.

On Tue, Jul 23, 2019 at 05:30:48PM +0200, Oliver Hartkopp wrote:
> I upgraded my TS-219 to Debian Buster with Kernel 4.19.0-5-marvell.
> 
> They use dtbs/4.19.0-5-marvell/./kirkwood-ts219-6282.dtb
> 
> Unfortunately the RTC doesn't seem to work.
> 
> root@xxxxx:~# hwclock -v --test
> hwclock from util-linux 2.33.1
> System Time: 1563296223.425648
> Trying to open: /dev/rtc0
> Using the rtc interface to the clock.
> Last drift adjustment done at 1563291472 seconds after 1969
> Last calibration done at 1563291472 seconds after 1969
> Hardware clock is on UTC time
> Assuming hardware clock is kept in UTC time.
> Waiting for clock tick...
> hwclock: select() to /dev/rtc0 to wait for clock tick timed out
> ...synchronization failed
> Test mode: nothing was changed.
> With strace the clock tick problem looks like this:
> 
> write(1, "Waiting for clock tick...\n", 26Waiting for clock tick...
> ) = 26
> ioctl(4, RTC_UIE_ON)                    = 0
> _newselect(5, [4], NULL, NULL, {tv_sec=10, tv_usec=0}) = 0 (Timeout)
> write(2, "hwclock: ", 9hwclock: )                = 9
> write(2, "select() to /dev/rtc0 to wait fo"..., 54select() to /dev/rtc0 to
> wait for clock tick timed out) = 54
> write(2, "\n", 1)                       = 1
> ioctl(4, PHN_NOT_OH or RTC_UIE_OFF)     = 0

Maybe rtc-mv should set uie_unsupported if no irq is aquired? 

> It looks the same with
> 
> # hwclock --hctosys
> hwclock: select() to /dev/rtc0 to wait for clock tick timed out

hwclock (from util-linux) is being stupid here. The tradtional RTC on
the x86 platform has a resolution of 1s only and if you set it the next
increment is 1 second after the last increment instead of 1 second after
the new time was set. So hwclock waits for an update event and then sets
the new time (in some cases to a tad later than requested) to ensure the
PC clock is off by less than half a second. For most (if not all) other
RTCs this is wrong and still worse if the RTC in question doesn't
support UIE.

> I wonder whether the problem comes from a missing interrupt assignment
> 
> rtc: rtc@10300 {
>              compatible = "marvell,kirkwood-rtc", "marvell,orion-rtc";
>              reg = <0x10300 0x20>;
>              interrupts = <53>;            <- HERE!?!
>              clocks = <&gate_clk 7>;
>                };
> 
> ... I found in linux/arch/arm/boot/dts/kirkwood-6282.dtsi ?!?
> 
> In /proc/interrupts there's no rtc assigned to an interrupt 53.

What is the output of dmesg | grep 10300 ?
 
> Accessing the rtc values (time/date) via /sys/class/rtc/rtc0 entries works
> well and setting the date/time via "hwclock --systohc" does its job too. So
> I2C and the rtc_s35390a driver seem to work so far.

You're not using the s35390a driver, do you?
 
> Any ideas why hwclock fails to work properly?

Try if 

	busybox hwclock 

works better for you. At least this one doesn't have this strange PC
quirk.
 
Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

Re: TS-219 RTC issue with Debian Buster

[Oliver Hartkopp]

Hello Uwe,

nice to read from you again :-)

I think I already solved the problem, see here:
https://marc.info/?l=linux-arm-kernel&m=156390875629259&w=2

Will file a Debian bug for it ...

Many thanks,
Oliver

On 23.07.19 21:06, Uwe Kleine-König wrote:
> Hello Oliver,
> 
> I added the RTC maintainers and list to Cc.
> 
> On Tue, Jul 23, 2019 at 05:30:48PM +0200, Oliver Hartkopp wrote:
>> I upgraded my TS-219 to Debian Buster with Kernel 4.19.0-5-marvell.
>>
>> They use dtbs/4.19.0-5-marvell/./kirkwood-ts219-6282.dtb
>>
>> Unfortunately the RTC doesn't seem to work.
>>
>> root@xxxxx:~# hwclock -v --test
>> hwclock from util-linux 2.33.1
>> System Time: 1563296223.425648
>> Trying to open: /dev/rtc0
>> Using the rtc interface to the clock.
>> Last drift adjustment done at 1563291472 seconds after 1969
>> Last calibration done at 1563291472 seconds after 1969
>> Hardware clock is on UTC time
>> Assuming hardware clock is kept in UTC time.
>> Waiting for clock tick...
>> hwclock: select() to /dev/rtc0 to wait for clock tick timed out
>> ...synchronization failed
>> Test mode: nothing was changed.
>> With strace the clock tick problem looks like this:
>>
>> write(1, "Waiting for clock tick...\n", 26Waiting for clock tick...
>> ) = 26
>> ioctl(4, RTC_UIE_ON)                    = 0
>> _newselect(5, [4], NULL, NULL, {tv_sec=10, tv_usec=0}) = 0 (Timeout)
>> write(2, "hwclock: ", 9hwclock: )                = 9
>> write(2, "select() to /dev/rtc0 to wait fo"..., 54select() to /dev/rtc0 to
>> wait for clock tick timed out) = 54
>> write(2, "\n", 1)                       = 1
>> ioctl(4, PHN_NOT_OH or RTC_UIE_OFF)     = 0
> 
> Maybe rtc-mv should set uie_unsupported if no irq is aquired?
> 
>> It looks the same with
>>
>> # hwclock --hctosys
>> hwclock: select() to /dev/rtc0 to wait for clock tick timed out
> 
> hwclock (from util-linux) is being stupid here. The tradtional RTC on
> the x86 platform has a resolution of 1s only and if you set it the next
> increment is 1 second after the last increment instead of 1 second after
> the new time was set. So hwclock waits for an update event and then sets
> the new time (in some cases to a tad later than requested) to ensure the
> PC clock is off by less than half a second. For most (if not all) other
> RTCs this is wrong and still worse if the RTC in question doesn't
> support UIE.
> 
>> I wonder whether the problem comes from a missing interrupt assignment
>>
>> rtc: rtc@10300 {
>>               compatible = "marvell,kirkwood-rtc", "marvell,orion-rtc";
>>               reg = <0x10300 0x20>;
>>               interrupts = <53>;            <- HERE!?!
>>               clocks = <&gate_clk 7>;
>>                 };
>>
>> ... I found in linux/arch/arm/boot/dts/kirkwood-6282.dtsi ?!?
>>
>> In /proc/interrupts there's no rtc assigned to an interrupt 53.
> 
> What is the output of dmesg | grep 10300 ?
>   
>> Accessing the rtc values (time/date) via /sys/class/rtc/rtc0 entries works
>> well and setting the date/time via "hwclock --systohc" does its job too. So
>> I2C and the rtc_s35390a driver seem to work so far.
> 
> You're not using the s35390a driver, do you?
>   
>> Any ideas why hwclock fails to work properly?
> 
> Try if
> 
> 	busybox hwclock
> 
> works better for you. At least this one doesn't have this strange PC
> quirk.
>   
> Best regards
> Uwe
> 

Re: TS-219 RTC issue with Debian Buster

[Uwe Kleine-König]

Hello Oliver,

On Tue, Jul 23, 2019 at 09:21:06PM +0200, Oliver Hartkopp wrote:
> I think I already solved the problem, see here:
> https://marc.info/?l=linux-arm-kernel&m=156390875629259&w=2

IMHO the right fix is teach hwclock to handle your rtc correctly.
Changing the driver from =m to =y is only a workaround.

Maybe also using the time stuff from systemd instead of hwclock will
help^Wwork around your problem (not sure though).

Also I wonder why the ioctl to enable UIE doesn't fail if the driver set
.uie_unsupported (but I didn't look closely).

> Will file a Debian bug for it ...

Not sure this will have success. These were switched from =y to =m
because the kernel images got to big.
(https://salsa.debian.org/kernel-team/linux/commit/c55043a43eaacb9f876dabd71273b084cb024441)

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

Re: TS-219 RTC issue with Debian Buster

[Oliver Hartkopp]

Hello Uwe,

On 23.07.19 21:34, Uwe Kleine-König wrote:

> On Tue, Jul 23, 2019 at 09:21:06PM +0200, Oliver Hartkopp wrote:
>> I think I already solved the problem, see here:
>> https://marc.info/?l=linux-arm-kernel&m=156390875629259&w=2
> 
> IMHO the right fix is teach hwclock to handle your rtc correctly.
> Changing the driver from =m to =y is only a workaround.

Agreed.

> Maybe also using the time stuff from systemd instead of hwclock will
> help^Wwork around your problem (not sure though).
> 
> Also I wonder why the ioctl to enable UIE doesn't fail if the driver set
> .uie_unsupported (but I didn't look closely).

Right. When executing the ioctl(4, RTC_UIE_ON) syscall it just get's 
stuck and the select() timeout captures the call after 10 seconds.

I would have expected that ioctl(4, RTC_UIE_ON) returns with some 
-EINVAL or -ENOTSUPP value.

I'm not sure if hwclock needs the trigger for precision reasons:

https://github.com/karelzak/util-linux/blob/master/sys-utils/hwclock.c#L276

... or if this can be omitted.

>> Will file a Debian bug for it ...
> 
> Not sure this will have success. These were switched from =y to =m
> because the kernel images got to big.
> (https://salsa.debian.org/kernel-team/linux/commit/c55043a43eaacb9f876dabd71273b084cb024441)

Too late ;-)
I will just wait for the answer.

Best regards,
Oliver