Hello Greg and Andre,
> From: Andre Werner <andre.werner@xxxxxxxxxxxxxxxxxxxxx>
>
> Dear Greg:
>
> On Thu, 19 Dec 2024, Greg KH wrote:
>
> > On Thu, Dec 19, 2024 at 10:22:40AM +0100, Andre Werner wrote:
> > > Dear Greg,
> > > On Thu, 19 Dec 2024, Greg KH wrote:
> > >
> > > > On Thu, Dec 19, 2024 at 09:46:38AM +0100, Andre Werner wrote:
> > > > > Fall back to polling mode if no interrupt is configured because not
> > > > > possible. If "interrupts" property is missing in devicetree the driver
> > > > > uses a delayed worker to pull state of interrupt status registers.
> > > > >
> > > > > Signed-off-by: Andre Werner <andre.werner@xxxxxxxxxxxxxxxxxxxxx>
> > > > > ---
> > > > > This driver was tested on Linux 5.10. We had a custom board that was not
> > > > > able to connect the interrupt port. Only I2C was available.
> > > >
> > > > Could you not test this on the latest tree? 5.10 is _VERY_ old now.
> > >
> > > I will try it on devboard with a 6.1 Kernel. Is that okay for you?
> >
> > 6.1 was released in December of 2022, 2 full years and hundreds of
> > thousands of changes ago. Please work off of Linus's latest tree, we
> > can't go back in time :)
I agree with Greg that such a new feature only belongs in the current development.
> > > > > @@ -1537,7 +1564,13 @@ int sc16is7xx_probe(struct device *dev, const
> struct sc16is7xx_devtype *devtype,
> > > > >
> > > > > /* Always ask for fixed clock rate from a property. */
> > > > > device_property_read_u32(dev, "clock-frequency", &uartclk);
> > > > > + s->polling = !device_property_present(dev, "interrupts");
> > > > >
> > > > > + if (s->polling) {
> > > > > + dev_warn(dev,
> > > > > + "No interrupt definition found. Falling back to
> polling mode.\n");
> > > >
> > > > What is a user supposed to do with this message? And why would a device
> > > > NOT have any interrupts? This feels like it is just going to pound on
> > > > the device and cause a lot of power drain for just a simple little uart.
> > >
> > > I thought it could be interesting to know that the device has missing
> > > interrupt support.
> >
> > Maybe, but as you are now warning a user about this, what are they
> > supposed to do to fix it?
Again, I agree with Greg, this message is appropriate to the developer only.
I suggest to change the log level.
> > > > Why can't your system provide a valid irq line?
Here, I am more with Andre. This just happens sometimes, esp. in embedded
solutions.
> > > In our case we have only an I2C available in a connection cable and the
> > > GPIOs are linked through a two way level shifter.
> > > It was a very special situation in our case because target platform and
> > > sensor platform are provided.
> > > The IRQ from the sensor war not able to drive the two way level shifter low so
> > > we always detect outgoing traffic and the IRQ signal but at the target
> > > board after the level shifter the signal remains stable. So
> > > communication failed with a timeout. So we need to force polling the
> > > interrupt status register because
> > > both HW solution should not be changed in any way.
> >
> > Again, you are burning a TON of power just for a simple little uart,
> > with your system never being able to go to sleep, are you sure this is
> > something that you want others to emulate and support?
But the alternative is even worse: a non-functioning uart. Something that
works but consumes too much power is better than something that doesn't
even work at all.
> I got your point and I'm fully with. This caused me to print a warning
> in Kernel log because it should not be the general working method.
> In our special case we do not have any other option because the sensor
> module using the SC16IS7xx and the hardware with the MCU running Linux OS
> are fixed. We had no possibilities to move any GPIO or such. This was
> the only chance to support the dedicated sensor platform and I may be
> the case that someone else faces the same problems. I thought that
> someone else may benefit from this workaround too. But as I got your
> point I'm also fine if it is not merged into main Linux Kernel sources.
I vote this patch to go in (after modification).
Kind regards,
Maarten
