Re: [libgpiod] [PATCH 11/19] API: add support for SET_CONFIG

[Kent Gibson <warthog618@xxxxxxxxx>]

On Wed, Nov 20, 2019 at 04:18:24PM +0100, Bartosz Golaszewski wrote:
> śr., 20 lis 2019 o 15:36 Kent Gibson <warthog618@xxxxxxxxx> napisał(a):
> >
> > On Wed, Nov 20, 2019 at 03:18:36PM +0100, Bartosz Golaszewski wrote:
> > > śr., 20 lis 2019 o 15:13 Kent Gibson <warthog618@xxxxxxxxx> napisał(a):
> > > >
> > > > On Wed, Nov 20, 2019 at 03:08:57PM +0100, Bartosz Golaszewski wrote:
> > > > > śr., 20 lis 2019 o 14:59 Kent Gibson <warthog618@xxxxxxxxx> napisał(a):
> > > > > >
> > > > > > On Wed, Nov 20, 2019 at 12:00:45PM +0100, Bartosz Golaszewski wrote:
> > > > > > > wt., 19 lis 2019 o 16:53 Kent Gibson <warthog618@xxxxxxxxx> napisał(a):
> > > > > > > >
> > > > > > > > On Mon, Nov 18, 2019 at 10:48:25PM +0800, Kent Gibson wrote:
> > > > > > > > > On Mon, Nov 18, 2019 at 02:52:04PM +0100, Bartosz Golaszewski
> > > > > > > > > > > +
> > > > > > > > > > > +int gpiod_line_set_flags_bulk(struct gpiod_line_bulk *bulk, int flags)
> > > > > > > > > > > +{
> > > > > > > > > > > +       struct gpiod_line *line;
> > > > > > > > > > > +       int values[GPIOD_LINE_BULK_MAX_LINES];
> > > > > > > > > > > +       unsigned int i;
> > > > > > > > > > > +       int direction;
> > > > > > > > > > > +
> > > > > > > > > > > +       line = gpiod_line_bulk_get_line(bulk, 0);
> > > > > > > > > > > +       if (line->as_is) {
> > > > > > > > > >
> > > > > > > > > > Can you explain the purpose of this as_is field? I'm not sure this is
> > > > > > > > > > really needed.
> > > > > > > > > >
> > > > > > > > >
> > > > > > > > > It is there for gpiod_set_flags, which has to populate the direction
> > > > > > > > > flags in the SET_CONFIG ioctl. The existing line->direction is
> > > > > > > > > either input or output.  It is drawn from GPIOLINE_FLAG_IS_OUT, so
> > > > > > > > > as-is is gets mapped to input.
> > > > > > > > > I didn't want to change the existing line->direction, and adding the
> > > > > > > > > as-is seemed clearer than adding another flavour of direction that
> > > > > > > > > contained all three.
> > > > > > > > >
> > > > > > > >
> > > > > > > > Hmmm, I think I see what you were getting at - the line->direction is the
> > > > > > > > direction from the kernel, so it doesn't hurt to use that value to set the
> > > > > > > > corresponding request flags - even if the original request was as-is??
> > > > > > > >
> > > > > > > > If that is the case then the line->as_is can be dropped throughout.
> > > > > > > >
> > > > > > > > Kent.
> > > > > > > >
> > > > > > >
> > > > > > > Yes, this is what I was thinking. Just need to make sure the value
> > > > > > > from the kernel is up-to-date.
> > > > > > >
> > > > > >
> > > > > > So fail if needs_update?
> > > > > >
> > > > > > Kent.
> > > > >
> > > > > I'd say: do an implicit update before setting config.
> > > > >
> > > >
> > > > So gpiod_line_update if needs_update, and fail if that fails?
> > > >
> > > > Kent.
> > >
> > > Without the if - needs_update is only set if an implicit update fails
> > > in line_maybe_update(). But in this case we need to be sure, so do it
> > > unconditionally.
> > >
> >
> > Given that line_maybe_update is called at the end of request creation, and
> > whenever set_config is called, how can line->direction be inconsistent
> > with the kernel state - as long as needs_update is false?
> >
> 
> I don't think we should call line_maybe_update() on set_config() - in
> this case we should call gpiod_line_update() and fail in set_config()
> if it fails.
> 
> I hope that's clearer.
> 

Not really.  I was already shaky on the needs_update and I'm getting more
confused about the overall needs_update handling policy by the minute.

Perhaps it will be more productive to go to the code.
I'll send out what I have as v2 and we can go from there.

Cheers,
Kent.