Re: [PATCH v2 06/10] can: netlink: add CAN_CTRLMODE_XL_ERR_SIGNAL

[Oliver Hartkopp <socketcan@hartkopp.net>]

On 09.11.25 15:54, Vincent Mailhol wrote:
> On 06/11/2025 at 09:50, Oliver Hartkopp wrote:
>> On 21.10.25 17:47, Vincent Mailhol wrote:
>>> Classical CAN and CAN FD must generate error frames on the CAN bus
>>> when detecting a protocol violation.
>>>
>>> CAN XL's error signaling is different and works as follows:
>>>
>>>     - In interoperability mode (both FD and XL), error signaling must be
>>>       on.
>>>
>>>     - When operating a CAN controller in CAN XL only mode but with TMS
>>>       off, the user can decide whether the error signalling is enabled
>>>       or disabled.
>>>
>>>     - On the contrary, when using TMS, error signalling must be off.
>>>
>>> Introduce the new CAN_CTRLMODE_XL_ERR_SIGNAL control mode. This new
>>> option is only made available for CAN XL, so despite the error
>>> signalling being always on for Classical CAN and CAN FD, forbid the
>>> use of this flag when CAN XL is off.
>>>
>>> If the user provides the error signalling flag, check its validity. If
>>> the flag is omitted, activate error signalling by default whenever
>>> possible. This is summarized in below table:
>>>
>>>              CAN_CTRLMODE_XL_ERR_SIGNAL
>>>      -------------------------------------------
>>>      CC/FD        option not available
>>>      CC/FD/XL    on
>>
>> Yes. This is the 'mixed-mode'
>> I would propose to use the 'mixed-mode' expression in the patch description.
> 
> Ack!
> 
>>>      XL TMS off    configurable (default on)
>>
>> Good default.
>>
>>>      XL TMS on    off
>>>
>>> Suggested-by: Oliver Hartkopp <socketcan@hartkopp.net>
>>> Link: https://lore.kernel.org/linux-can/20250527195625.65252-9-
>>> socketcan@hartkopp.net/
>>> Signed-off-by: Vincent Mailhol <mailhol@kernel.org>
>>> ---
>>>    drivers/net/can/dev/dev.c        |  2 ++
>>>    drivers/net/can/dev/netlink.c    | 29 +++++++++++++++++++++++++++--
>>>    include/uapi/linux/can/netlink.h |  1 +
>>>    3 files changed, 30 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/can/dev/dev.c b/drivers/net/can/dev/dev.c
>>> index 1de5babcc4f3..0c16d0174f7f 100644
>>> --- a/drivers/net/can/dev/dev.c
>>> +++ b/drivers/net/can/dev/dev.c
>>> @@ -125,6 +125,8 @@ const char *can_get_ctrlmode_str(u32 ctrlmode)
>>>            return "xl-tdc-manual";
>>>        case CAN_CTRLMODE_XL_TMS:
>>>            return "xl-tms";
>>> +    case CAN_CTRLMODE_XL_ERR_SIGNAL:
>>> +        return "xl-error-signalling";
>>>        default:
>>>            return "<unknown>";
>>>        }
>>> diff --git a/drivers/net/can/dev/netlink.c b/drivers/net/can/dev/netlink.c
>>> index 8afd2baa03cf..6126b191fea0 100644
>>> --- a/drivers/net/can/dev/netlink.c
>>> +++ b/drivers/net/can/dev/netlink.c
>>> @@ -191,7 +191,8 @@ static int can_validate_xl_flags(struct netlink_ext_ack
>>> *extack,
>>>            }
>>>            if (masked_flags & CAN_CTRLMODE_XL_TMS) {
>>>                const u32 tms_conflicts_mask = CAN_CTRLMODE_FD |
>>> -                CAN_CTRLMODE_XL_TDC_MASK;
>>> +                CAN_CTRLMODE_XL_TDC_MASK |
>>> +                CAN_CTRLMODE_XL_ERR_SIGNAL;
>>>                u32 tms_conflicts = masked_flags & tms_conflicts_mask;
>>>                  if (tms_conflicts) {
>>> @@ -201,11 +202,23 @@ static int can_validate_xl_flags(struct netlink_ext_ack
>>> *extack,
>>>                    return -EOPNOTSUPP;
>>>                }
>>>            }
>>> +        if ((masked_flags & CAN_CTRLMODE_FD) &&
>>> +            (mask & CAN_CTRLMODE_XL_ERR_SIGNAL) &&
>>> +            !(masked_flags & CAN_CTRLMODE_XL_ERR_SIGNAL)) {
>>> +            NL_SET_ERR_MSG(extack,
>>> +                       "When using both CAN FD and XL, error signalling must
>>> be on");
> 
> I changed that error message to:
> 
> 	NL_SET_ERR_MSG(extack, "Mixed mode requires error signalling");
> 

Good!

>> This implicitly tells us that mixed-mode is CC/FD/XL ;-)
> 
> I was under the assumption that Classical CAN was always allowed, even under
> TMS. The arbitration still uses the nominal bittiming anyway, so I still have
> some issue understanding why an XL nodes operating under TMS wouldn't be able to
> send a classical CAN frame.

TMS is XL-Only and has error-signalling off. Therefore no CC/FD traffic.

> The restriction seems rather arbitrary to me. I would be curious to understand
> what the issue would be to allow Classical CAN under TMS.

I tried to clarify this in the other answer.

Best regards,
Oliver