Dear Alexander Stein,
In message 201007220817.17441.alexander.stein@systec-electronic.com you wrote:
This seems the wrong approach to me - when CONFIG_HW_WATCHDOG is defined and requires CONFIG_SYS_WDTC_WDMR_VAL, but the latter is missing, than this should raise n error condition. We must not silently ignore errors.
Well, my problem is, that CONFIG_SYS_WDTC_WDMR_VAL is used to program the internal watchdog. But this watchdog can only be programmed once until a reset occurs. So there is no possibility for linux to reprogram it.
This is normal. Any watchdog that is worth the name will behave similar.
So, if CONFIG_SYS_WDTC_WDMR_VAL is not defined and the watchdog is not programed using my patch, the watchdog still runs with default settings (timeout of 16s). So a user may choose to trigger the watchdog from u-boot (define CONFIG_AT91SAM9_WATCHDOG and CONFIG_HW_WATCHDOG) or let it run silently. To summarize, CONFIG_SYS_WDTC_WDMR_VAL is not needed to use CONFIG_AT91SAM9_WATCHDOG and CONFIG_HW_WATCHDOG.
Then the subject is misleading - it suggests you do not initialize / let run the watchdog at all if CONFIG_SYS_WDTC_WDMR_VAL is undefined.
I think we should make sure that everything is in a sane and consistent state - if CONFIG_HW_WATCHDOG (and CONFIG_AT91SAM9_WATCHDOG) are set, this indicates that U-Boot is supposed to use the watchdog, which in turn means they should initialize it.
Best regards,
Wolfgang Denk