On Thu, Nov 10, 2016 at 10:13:02AM +0100, Maxime Ripard wrote: > On Wed, Nov 09, 2016 at 08:02:08PM -0800, Dmitry Torokhov wrote: > > On Wed, Nov 09, 2016 at 09:02:35AM +0100, Maxime Ripard wrote: > > > Hello Dmitry, > > > > > > On Tue, Nov 08, 2016 at 04:04:00PM -0800, Dmitry Torokhov wrote: > > > > On Mon, Nov 07, 2016 at 03:40:24PM +0100, Maxime Ripard wrote: > > > > > The TCA8418 interrupt has a level trigger, not a edge one. > > > > > > > > > > Signed-off-by: Maxime Ripard <maxime.rip...@free-electrons.com> > > > > > > > > Hmm, maybe we could rely on OF data for trigger type? > > > > > > We might, even though the i2c core doesn't change the trigger type > > > when it retrieves the interrupt from the DT. > > > > i2c core itself does not, and should not, but irq code does: > > > > of_irq_get() -> irq_create_of_mapping() -> irq_create_fwspec_mapping() > > -> irqd_set_trigger_type(). > > Ah, indeed, I overlooked that. I wonder why platform_get_irq does it > then. > > > > > > > > > However, I'm a bit worried about the other probing mechanims (ACPI, > > > board files) that should be supported as well, and removing the > > > trigger type from the flags might break those. There's no board files > > > using it though in the tree, but I don't know about ACPI systems. > > > > The driver is not enabled for ACPI systems, at least not in mainline. > > Ok. Good. > > > By the way, this is what TCA8418 binding dochas to say: > > > > "- interrupts: IRQ line number, should trigger on falling edge" > > > > so it seems there was at least one system that needed falling edge and > > not level interrupt. > > I don't know, looking at the datasheet, it really looks like it's > level triggered to me, and we were actually seeing issues when set in > edge. > > http://www.ti.com/lit/ds/symlink/tca8418.pdf > Especially page 20 and 33. > > My understanding is that in input, the chip will trigger on edges, but > the line coming from that device to the SoC will be level triggered.
The issue is that it is not necessarily connected directly to the SoC/AP. It could be on a daughterboard with any number of converters, possibly inverting polarity, or doing level->edge, etc. So the only sane solution is to leave it to device tree to describe the setup to the driver. Thanks. -- Dmitry