On 22.12.2022 07:44:59, Stefan Althöfer wrote:
> I'm using the MCP2518FD CAN controller with Rasperry PI 4 and are
> suffering from wrong data reception.
> 
> My hardware is a Model 4 B with MCP2581 Click board(s) attached with a
> prototype (hand-made) interface PCB, as well as baseboards for a CM4
> with a single MCP2518 on it. The error occurs on both of them.

Do you get the same error if the test is only running on a single
mcp2518fd?

Do you get the same error if only a single mcp2518fd is attached to the
raspi?

> I have reduced my test case to a simple single thread self-receipt test:
>     * TX two messages
>     * Wait for RX and send out a new message on every receipt
>     * TX for messages in total

Can you share the config/scripts/... to set up the interfaces?
Can you share the test code?

> Refer to the attached PDF for some error cases.

The text in the PDF is rendered down to paths only. So not select- nor
searchable, please send plain text files for further logs.

> Last send frames are at the top of the logs. You can see that wrong
> messages appear in the RX queue, which have been successfully
> transmitted in previous test loop. The data that is actually sent out
> is correct however (checked with an external logger for some cases).

Looking at the first page, there are 2 identical lines marked in red.
The next page show 4 red lines.

So under certain load situations you see that old messages are received
twice. If I remember correctly I heard something similar from someone
else before. It occurred with a modified driver. For various reasons
that has never debugged. I think I've fixed the root cause of the
problem during the v5.18 development cycle.

> The error appeared very often (every 1..10 minutes) when I was using
> kernel raspberrypi 5.15.32-v7l+ and reduced to once in several hours
> when I switched to raspberrypi 6.0.12-v7l+. BTW I modified the
> loopback to external loopback To check for correct transmission. No
> other changes to the driver as taken from the Raspbian repository.

Can you share your kernel, including the modifications?

> I see infrequent mcp251xfd CRC read errors. I think those are due to
> the 2518 SPI errata. However they don't occur at the time when the
> wrong messages are received (refer to the PDF).
> 
> Questions:
> - Is this a known problem?

No - it should not happen on any mainline kernel.

> - What did change from 5.15.32 to 6.0.12 to reduce the error rate?
>   Maybe this is just because of a change in execution timing?

Quite a lot: Next to probably unrelated changes (e.g. PLL support), the
workarounds for CRC errors were updated, in v5.18 multiple RX FIFOs are
used and IRQ coalescing support has been added.

> - Any suggestion how I can step further in fixing this issue.

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde           |
Embedded Linux                   | https://www.pengutronix.de  |
Vertretung West/Dortmund         | Phone: +49-231-2826-924     |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-5555 |