Re: [PATCH] gpio: Return EPROBE_DEFER if gc->to_irq is NULL

From: Shreeya Patel
Date: Mon Oct 18 2021 - 18:14:57 EST



On 14/10/21 10:21 pm, Linus Walleij wrote:
On Thu, Oct 14, 2021 at 1:05 PM Shreeya Patel
<shreeya.patel@xxxxxxxxxxxxx> wrote:

We are racing the registering of .to_irq when probing the
i2c driver. This results in random failure of touchscreen
devices.

Following errors could be seen in dmesg logs when gc->to_irq is NULL

[2.101857] i2c_hid i2c-FTS3528:00: HID over i2c has not been provided an Int IRQ
[2.101953] i2c_hid: probe of i2c-FTS3528:00 failed with error -22

To avoid this situation, defer probing until to_irq is registered.

This issue has been reported many times in past and people have been
using workarounds like changing the pinctrl_amd to built-in instead
of loading it as a module or by adding a softdep for pinctrl_amd into
the config file.

References :-
https://bugzilla.kernel.org/show_bug.cgi?id=209413
https://github.com/Syniurge/i2c-amd-mp2/issues/3

Signed-off-by: Shreeya Patel <shreeya.patel@xxxxxxxxxxxxx>
I understand the issue.

There is one problem.

@@ -3084,7 +3084,7 @@ int gpiod_to_irq(const struct gpio_desc *desc)

return retirq;
}
- return -ENXIO;
+ return -EPROBE_DEFER;
If you after five minutes plug in a USB FTDI or similar UART thing
with a GPIO expander, and someone request an IRQ from
one of those lines (they do not support interrupts), why should
it return -EPROBE_DEFER?

The point is that I think this will in certain circumstances return
a bogus error.

I was worried about the same but didn't really know under what scenario this could occur.
Thanks for pointing this out.


We cannot merge this other than with a fat comment above:

/*
* This is semantically WRONG because the -EPROBE_DEFER
* is really just applicable during system bring-up.
*/
return -EPROBE_DEFER;

Can we use some kind of late_initcall() to just switch this over
to -ENXIO after a while?


I have sent a v2 which tries to fix this in an easy way. Let me know what do you
think about that approach or else we could also think about using late_initcall().



Thanks,
Shreeya Patel


Yours,
Linus Walleij