[PATCH] v2 rtc-at91rm9200: add support for at91sam9x5

[PATCH] v2 rtc-at91rm9200: add support for at91sam9x5

[Douglas Gilbert]

An unspecified number of SoCs from the Atmel's
at91sam9x5 family including the at91sam9g25 have a
broken AT91_RTC_IMR (interrupt mask) register which
always returns zero.

If this bug can be neutralized, then the existing
rtc-at91rm9200 driver will work. On the other hand,
leaving this bug in place, and starting the RTC
causes unhandled interrupts which result in the
"SYS" interrupt being disabled. And that takes down
several other interrupts wired or-ed through the
SYS interrupt including the DBG port.

This is the second version of this patch, and is
against lk 3.10.0-rc2 .

ChangeLog:
    - checks in probe() function if AT91_RTC_IMR is broken.
      If so, uses a shadow imr
    - apart from that check, SoCs with a good IMR register
      take the same paths through rtc-at91rm9200.c as before
    - SoCs with a broken IMR take a spinlock while changing
      the state of IER or IDR (and the shadow), mainly to
      disable interrupts **.

** similar technique used in arch/arm/mach-at91/clock.c

Tested on an Aria G25 (at91sam9g25 based).

Signed-off-by: Douglas Gilbert <dgilbert@interlog.com>


-------------- next part --------------
A non-text attachment was scrubbed...
Name: rtc-sam9x5_3100rc1.patch
Type: text/x-patch
Size: 5787 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20130521/44c67885/attachment.bin>

[PATCH] v2 rtc-at91rm9200: add support for at91sam9x5

[Johan Hovold]

On Tue, May 21, 2013 at 06:21:07PM -0400, Douglas Gilbert wrote:
> An unspecified number of SoCs from the Atmel's
> at91sam9x5 family including the at91sam9g25 have a
> broken AT91_RTC_IMR (interrupt mask) register which
> always returns zero.
> 
> If this bug can be neutralized, then the existing
> rtc-at91rm9200 driver will work. On the other hand,
> leaving this bug in place, and starting the RTC
> causes unhandled interrupts which result in the
> "SYS" interrupt being disabled. And that takes down
> several other interrupts wired or-ed through the
> SYS interrupt including the DBG port.
> 
> This is the second version of this patch, and is
> against lk 3.10.0-rc2 .

Your patch is almost identical to the one I submitted about two months
ago in this thread:

	https://lkml.org/lkml/2013/4/3/166

except that you do not use DT to the detect SoCs with broken IMR. (Your
patch is also lacking the register-read-back heuristic to make sure the
interrupts have been disabled before you update the mask.)

Are there sam9x5 SoCs with non-broken RTC_IMR which would prevent us
from using DT for detection? If so, I'd rather see your test for
broken-IMR implemented on top of my implementation.

Thanks,
Johan

> ChangeLog:
>     - checks in probe() function if AT91_RTC_IMR is broken.
>       If so, uses a shadow imr
>     - apart from that check, SoCs with a good IMR register
>       take the same paths through rtc-at91rm9200.c as before
>     - SoCs with a broken IMR take a spinlock while changing
>       the state of IER or IDR (and the shadow), mainly to
>       disable interrupts **.