Re: RTC seems broken on 2.6.38-rc1 for RTCs lacking alarms.

[David Daney <ddaney@caviumnetworks.com>]

On 01/20/2011 02:23 PM, John Stultz wrote:
> On Thu, 2011-01-20 at 22:24 +0100, Geert Uytterhoeven wrote:
>> On Thu, Jan 20, 2011 at 22:16, John Stultz<john.stultz@linaro.org>  wrote:
>>> On Thu, 2011-01-20 at 12:51 -0800, David Daney wrote:
>>>> open("/dev/rtc0", O_RDONLY|O_LARGEFILE) = 3
>>>> ioctl(3, PRESTO_GETMOUNT or RTC_UIE_ON, 0) = 0
>>>> _newselect(4, [3], NULL, NULL, {5, 0})  = 0 (Timeout)
>>>> write(2, "select() to /dev/rtc0 to wait for"..., 55select() to /dev/rtc0
>>>> to wait for clock tick timed out
>>>> ) = 55
>>>> ioctl(3, PRESTO_SETPID or RTC_UIE_OFF, 0) = 0
>>>> close(3)                                = 0
>>>> exit_group(1)                           = ?
>>>>
>>>>
>>>> The hwclock program is asking to put the clock in UIE mode and then
>>>> does a select() on it.  Since the alarm doesn't work, the select times out.
>>>>
>>>> Previously the ioctl(RTC_UIE_ON) would return EINVAL:
>>>
>>> Ah. Good diagnosis! Let me try to get a patch for you and Andreas to
>>> test.
>>
>> I'm also seeing this on m68k (ARAnyM, rtc-generic).
>
> Geert, David, Andreas,
> 	Could you try the following? Its a bit messy of a patch doing a couple
> of things:
>
> 1) Simplify the timer->enabled management by pushing it into
> rtc_timer_enqueue/remove (needed cleanup for #2).
>
> 2) Properly propagating errors from __rtc_set_alarm back through
> rtc_timer_enqueue and users.
>
> 3) Trivial clenaup making rtc_timer_enqueue/remove static.
>
> 4) Fixup virtualized rtc_read_alarm to check hardware capabilities and
> return errors (also restores zeroing of the rtc_wkalrm stucture).
>
> I'll be cleaning these up and breaking them into commits I can send
> upward, but I wanted to make sure it resolves the issue for you.
>
> Let me know if it fixes things.
>
> thanks
> -john
>

John,

With that patch, it works now.  Thanks.

You can add:

Tested-by: David Daney <ddaney@caviumnetworks.com>