Re: [PATCH v2] watchdog: mediatek: acknowledge pretimeout interrupt

[Guenter Roeck <linux@roeck-us.net>]

On 8/27/26 19:08, Tzung-Bi Shih wrote:
> On Thu, Aug 27, 2026 at 06:54:03AM -0700, Guenter Roeck wrote:
>> On Thu, Aug 27, 2026 at 05:26:12PM +0800, Wanming Gao wrote:
>>> The MediaTek watchdog pretimeout interrupt is level-triggered and does
>>> not have a separate acknowledge register. The interrupt is cleared by
>>> changing WDT_MODE_IRQ_LEVEL_EN and then restoring it to its original
>>> state. Without this transition, the interrupt may remain asserted and
>>> cause an interrupt storm.
>>>
>>> After changing WDT_MODE_IRQ_LEVEL_EN, wait 70 us before restoring it.
>>> This is longer than two 32 kHz watchdog clock cycles, allowing the
>>> level change to propagate across the clock domain.
>>>
>>> WDT_MODE is also updated by the watchdog start, stop, and pretimeout
>>> operations. Protect its read-modify-write sequences and the complete
>>> IRQ acknowledge sequence with the watchdog spinlock so that concurrent
>>> updates cannot overwrite the temporary IRQ level state.
>>>
>>> Signed-off-by: Wanming Gao <wanming.gao@mediatek.com>
>>
>> Applied.
> 
> Might be a bit late, but tested on a MT8196 board:
> Tested-by: Tzung-Bi Shih <tzungbi@kernel.org>
> Reviewed-by: Tzung-Bi Shih <tzungbi@kernel.org>

Never too late, really, unless I already sent a pull request.
Also, I have not pushed anything out, waiting for v7.3-rc1.

Thanks,
Guenter