Re: [PATCH v5 1/7] Input: matrix_keypad - use fsleep for variable delay duration

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Mon, Feb 24, 2025 at 10:40:55PM -0800, Dmitry Torokhov wrote:
> On Wed, Feb 19, 2025 at 05:34:49PM +0100, Manuel Traut wrote:
> > On Fri, Jan 10, 2025 at 06:49:00AM +0100, Markus Burri wrote:
> > > The delay is retrieved from a device-tree property, so the duration is
> > > variable. fsleep guesses the best delay function based on duration.
> > > 
> > > see Documentation/timers/delay_sleep_functions.rst
> > > 
> > > Signed-off-by: Markus Burri <markus.burri@mt.com>
> > 
> > Reviewed-by: Manuel Traut <manuel.traut@mt.com> 
> 
> As I mentioned in other review activate_col() may be called in atomic
> context where we can not sleep:
> 
> "activate_col() may be called in atomic context, and if fsleep() turns
> into usleep_range() or msleep() we are going to have a bad time.
> 
> We should either stop using request_any_context_irq() or figure out if
> interrupt handler can sleep or not and adjust behavior properly."
> 
> Unfortunately this was completely ignored.

My apologies, it looks like it only is called from work handler, so my
comment was wrong.

Thanks.

-- 
Dmitry