Re: [PATCH] ppc44x/watchdog: Select WATCHDOG_NOWAYOUT option

Josh Boyer wrote:

I have no idea about FSL cores, but the 4xx maximum value selects TBU
bit 31.  When that bit flips is going to be determined by the speed of
the clock driving TB.  Most of the 4xx implementations I have seen use
the CPU clock, so to take the example from the 44x user manuals, a 400
MHz clock results in a maximum time period of ~21 seconds.  Multiply
by 3 to get the time before the WDT actually does a reset and you're
at about 1min.

Ah, that is a problem.

NOWAYOUT is working.  The patch was basically forcing it on, even if
the end user doesn't want the behavior described.  From what you've
said, that simply doesn't matter in the FSL case because the machine
will run for a really long time.  On 4xx, something needs to be done
to keep the machine running if the user doesn't want the NOWAYOUT
behavior, which is why I suggested a kernel timer.

Yes, I agree now -- a kernel timer is a better idea.  And it doesn't need
to be 4xx-specific.  Although, I wonder if it's the time is reliable enough.

We already have an exception handler:

#ifdef CONFIG_BOOKE_WDT
/*
 * Default handler for a Watchdog exception,
 * spins until a reboot occurs
 */
void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
{
	/* Generic WatchdogHandler, implement your own */
	mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
	return;
}

void WatchdogException(struct pt_regs *regs)
{
	printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
	WatchdogHandler(regs);
}
#endif

Maybe instead of disabling interrupts, we should ping the watchdog instead?

-- 
Timur Tabi
Linux kernel developer at Freescale