Re: Inconsistent RC5 ir-keytable events

["Marko Mäkelä" <marko.makela@iki.fi>]

Hi Sean,

Tue, Feb 08, 2022 at 04:46:31PM +0000, Sean Young wrote:
>Hi Marko,
>
>On Sat, Jan 29, 2022 at 07:15:57PM +0200, Marko Mäkelä wrote:
>> Hi Sean,
>>
>> Did you have a chance to repeat my findings with a remote control unit that
>> uses the RC5 protocol?
>
>Yes, I've reproduced the problem now. It's very weird.
>
>After pressing button 1, waiting for a second or two, and then pressing button
>3, the device reports some old IR from button 1 before reporting the IR from
>button 3.
>
>The IR is not a copy of old data, it so presumably its IR that was not
>reported before. Now I don't know why this IR gets reported so late.

When I repeated the problem, the delay between subsequent button presses 
could have been even less than a second.

How did you determine that it is not a copy of old data? I assume that 
you can do that fairly easily for any kernel-side buffers, but what 
about the buffer on the USB device itself?

[snip]

>I have not seen this happen. You can enable debugging for this using:
>
>echo 'file rtl28xxu.c +p' > /sys/kernel/debug/dynamic_debug/control
>
>dmesg -w

Thank you. It might take me some time to return to this.

>I do not know exactly how the buffer works, only some experimentation 
>will help here. Unless there is a vendor driver/documentation for this 
>device.

I am just guessing here, since I do not have any experience with USB 
programming, and hardly any experience with the Linux kernel. Could it 
be that the IR samples are being stored in a ring buffer, and the device 
does not implicitly discard IR events as it is sending them to the USB 
host? A simple test could be to read the buffer twice in a row. Will it 
return the same contents, instead of returning nothing on the second 
read?

If my ring buffer hypothesis holds, perhaps the kernel driver should 
keep track on where the data ended on the previous read, re-read the 
device's entire ring buffer, and resume parsing from the previous end of 
the ring buffer? In that way, we should not see phantom old events.

Best regards,

	Marko