Re: [PATCH 0/2] OMAP3: wdtimer: fix wdtimer blocking CORE idle

["Cousson, Benoit" <b-cousson@ti.com>]

Paul,

On 3/10/2011 2:54 PM, Cousson, Benoit wrote:
> On 3/10/2011 2:08 PM, Paul Walmsley wrote:
>> Hi Benoît,
>>
>> On Thu, 10 Mar 2011, Cousson, Benoit wrote:
>>
>>> On 3/10/2011 11:28 AM, kalle.jokiniemi@nokia.com wrote:
>>>>
>>>>> From: ext Paul Walmsley [mailto:paul@pwsan.com]
>>>>> Sent: 10. maaliskuuta 2011 11:50
>>>>>
>>>>> Thanks Kalle for tracking this down,
>>>>>
>>>>> could you try this patch along with the patch from this message:
>>>>>
>>>>>      http://www.spinics.net/lists/linux-omap/msg48115.html
>>>
>>> I have some question with that patch.
>>> By using pm_runtime API, you are suppose to idle the whole IP.
>>> In this case, just because you cannot idle a wdt that is running, it works,
>>> but the point is that the pm_runtime state will not reflect the HW state.
>>> For pm_runtime point of view, the wdt is supposed to be fully idle (both fclk
>>> and iclk). Whereas in that case, only the iclk will be gated.
>>>
>>> Did I miss something?
>>
>> Looking at it, I don't know how the old driver managed to work at all.
>> Commit 7ec5ad0f3c1e28b693185c35f768953c5db32291 ("OMAP: WDT: Use PM
>> runtime APIs instead of clk FW APIs") removed this from omap_wdt_probe():
>>
>> -       /* autogate OCP interface clock */
>> -       __raw_writel(0x01, wdev->base + OMAP_WATCHDOG_SYS_CONFIG);
>>
>> This would put the WDTIMER into force-idle, and nothing ever took it out
>> of force-idle.  So I don't quite understand why the PRCM wouldn't just
>> turn off the functional clock to the WDTIMER at this point.  Any thoughts
>> on this?
>
> The idle definition of the PRCM is just: idle of the ocp clock (iclk).
> The fclk can be active while the module is "idle" for the PRCM point of
> view.
> That's why we have a mismatch in that case with the pm_runtime idle
> definition, where the module should be fully idle.
>
> In that case, since the WDT is in the wakeup domain, we do not even have
> to re-enable the iclk since the wakeup domain is always on whenever the
> MPU is running.
>
> That fact that we have to explicitly do a force idle, probably means
> that the autoidle is broken in the IP.

Meanwhile, this patch is probably the best approach anyway.
I do not like going backward with a clock level management solution.

Regards,
Benoit
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html