On 02.11.2021 16:03:33, Matthias Weißer wrote:
> we recently had a case here where one member of a CAN bus couldn't
> receive frames with data content of only zeros:
> 
> $ cansend can0 123#0000000000000000
> 
> After some investigation we found the root cause to be a slight
> difference (about 1%) in actual bitrates of the two members. The one
> with showed the RX errors had a sjw value of 1 and a lot of time
> quanta (40) due to the 40MHz CAN clock.
> 
> This leads to a build up of phase error (as sjw is not able to
> compensate for enough of the bitrate difference) which at some point
> leads to a framing error due to missing a stuff bit. Playing around
> with the sample point can improve or worsen the behavior.
> 
> We can fix this quite easily by specifying a higher sjw value.

We have seen this behavior, too. :(

> Question is now:
> Wouldn't it make sense to increase sjw in can_calc_bittiming() to
> something like 5% of the total time quanta?

You mean of the total bit time?

The problem is that the SJW is set to 1 time quanta by default. For a
given bit rate the time quanta is not a constant value, but depends on
the oscillator frequency and configured bit rate prescaler. As you
suggested, I agree a better solution would be to set the SJW to a
certain percentage of the bit time. But, what is a proper default value
of a percentage based SJW?

> This may increase the reliability of the CAN network when there are
> differences in the bitrates of the single members. Are there any
> arguments against such an arbitrary selection of sjw?
> 
> If you agree with such a change I can come up with a proper patch. If
> I wrote totally nonsense please tell me :-)

I'm interested in a patch. There are some requirements and thoughts:
- introduce a define for the default SJW percentage value
- use the default SJW percentage value if the user has not provided a
  SJW
- the "struct can_bittiming" cannot be modified

In a later patch we can introduce a configuration interface to set SJW
percentage value from user space.

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 |