Re: RFC remove CAN_CTRLMODE_XL_ERR_SIGNAL

[Vincent Mailhol <mailhol@kernel.org>]

Hi Oliver,

On 09/11/2025 at 22:07, Oliver Hartkopp wrote:
> Hi Vincent,
> 
> I've been playing with the PEAK CAN XL bitrate tool again and it gave me a new
> idea:
> 
> There were settings for different bitrates and the belonging/resulting flags for
> TMS and ES (error-signalling).
> 
> Staring at the options we have only three options for the CAN XL support aka "xl
> on":
> 
> 1. Providing CC/FD/XL bitrates => ES on, TMS off (mixed-mode)
> 
> 2. Providing only CC/XL bitrates => ES off, TMS off (CANXL-only)
> 
> 3. Providing only CC/XL bitrates => ES off, TMS on (CANXL-only)

Why isn't

  4. Providing only CC/XL bitrates => ES on, TMS off

a valid option?

> Therefore we only need "tms" as an additional option when xl is on.
> 
> The error signalling "on" automatically results from the availability of "fd on"
> and the FD bitrate.
> 
> Examples:
> 
> 1. Providing CC/FD/XL bitrates => ES on, TMS off
> 
> ip link set can0 type can bitrate 500000 fd on dbitrate 2000000 xl on xbitrate
> 4000000
> 
> 2. Providing only CC/XL bitrates => ES off, TMS off
> 
> ip link set can0 type can bitrate 500000 xl on xbitrate 4000000
> 
> 3. Providing only CC/XL bitrates => ES off, TMS on
> 
> ip link set can0 type can bitrate 500000 xl on xbitrate 10000000 tms on
> 
> That's simple and provides only the needed switches, which makes
> CAN_CTRLMODE_XL_ERR_SIGNAL obsolete in the netlink API.
> 
> What do you think about this approach?

What really bothers me here is that the ISO standard explicitly state that error
signaling is a configurable option. Making this an implicit option would result
in a somehow non-compliant implementation.

I appreciate that for most of the use cases the error signaling can be inferred
from the other values, and this is what I tried to implement (c.f. the table of
default values which I put in my patch).

But I want to leave room so that people who wants to push the standard to its
limits can.

I see this a bit like the can_frame->len8_dlc thing. Should you use DLCs greater
than 8 in production code? Hell no! But I still want the implementation to give
me this option so that I can do my weird tests.


Yours sincerely,
Vincent Mailhol