RE: [RFC PATCH] watchdog: ep93xx: Use the WatchDog Timer DriverCore.

[H Hartley Sweeten]

On Wednesday, August 03, 2011 3:07 AM, Wim Van Sebroeck wrote:
> Hi H Hartley,
>
>>> [...]
>>>> @@ -210,43 +126,31 @@ static int __init ep93xx_wdt_init(void)
>>>>  {
>>>>  	int err;
>>>>  
>>>> -	err = misc_register(&ep93xx_wdt_miscdev);
>>>> +	ep93xx_wdd.bootstatus = readl(EP93XX_WDT_WATCHDOG);
>>>> +	ep93xx_wdd.timeout = timeout;
>>>> +
>>>> +	err = watchdog_register_device(&ep93xx_wdd);
>>>> +	if (err)
>>>> +		return err;
>>>>  
>>>> -	boot_status = __raw_readl(EP93XX_WDT_WATCHDOG) & 0x01 ? 1 : 0;
>>>> +	setup_timer(&timer, ep93xx_timer_ping, 1);
>>>
>>> Shouldn't the bootstatus setting be:
>>> ep93xx_wdd.bootstatus = readl(EP93XX_WDT_WATCHDOG) & 0x01 ? 1 : 0;
>>> (or something similar with the WDIOF_OVERHEAT, ... flags).
>>
>> The ep93xx watchdog status register doesn't line up nicely with the standard
>> WDIOF_* flags.  The register has these bits defined:
>> 
>> PLSDN		6	Pulse Disable Not
>> OVRID		5	Software Override of HWDIS
>> SWDIS		4	Software Watchdog Disable
>> HWDIS		3	Hardware Watchdog Disable
>> URST		2	User Reset Detected
>> 3KRST		1	Three-Key Reset Detected
>> WD		0	Watchdog Reset Detected
>
> Hmm, it would be nice to have this info also in the driver. Certainly at
> least a define for the WD bit.

I was doing to document these bits in a follow-up patch after the conversion.
Instead I'll do it now in v2 of this patch.  It does make the driver a bit clearer.

> Secondly: what it is doing now is allready incorrect: The WD bit is returned as
> bit 0 of the bootstatus value.
> This is actually: WDIOF_OVERHEAT (Reset due to CPU overheat) instead of
> WDIOF_CARDRESET (0x0020 Card previously reset the CPU)...

I agree, the current driver was already abusing the bootstatus use.

>> The original bootstatus setting just checked for the WD bit.  I would like
>> to pass the full status register so that userspace can figure out what caused
>> the reset.  Is this an inappropriate abuse of WDIOC_GETBOOTSTATUS?
>
> It's indeed an inappropriate abuse of WDIOC_GETBOOTSTATUS.
> The WDIOC_GETBOOTSTATUS should return a result for all watchdog drivers.
> The result is based on the WDIOF_* flags
> This however does not mean that we can add flags. The 'User Reset detected' looks
> to me as something that can be usefull in embedded environments. We need to think
> about this and see what would be usefull in general.

I'll post a v2 of this patch shortly and try to address this and your other comments.
 
Thanks,
Hartley
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/