Re: [RFC PATCH v3 8/9] can: slcan: add support to set bit time register (btr)

[Oliver Hartkopp]

On 27.07.22 20:24, Marc Kleine-Budde wrote:
> On 27.07.2022 19:28:39, Max Staudt wrote:
> > On Wed, 27 Jul 2022 13:30:54 +0200
> > Marc Kleine-Budde <mkl@xxxxxxxxxxxxxx> wrote:
> >
> > > As far as I understand, setting the btr is an alternative way to set the
> > > bitrate, right? I don't like the idea of poking arbitrary values into a
> > > hardware from user space.
> >
> > I agree with Marc here.
> >
> > This is a modification across the whole stack, specific to a single
> > device, when there are ways around.
> >
> > If I understand correctly, the CAN232 "S" command sets one of the fixed
> > bitrates, whereas "s" sets the two BTR registers. Now the question is,
> > what do BTR0/BTR1 mean, and what are they? If they are merely a divider
> > in a CAN controller's master clock, like in ELM327, then you could
> >
> >    a) Calculate the BTR values from the bitrate userspace requests, or
>
> Most of the other CAN drivers write the BTR values into the register of
> the hardware. How are these BTR values transported into the driver?
>
> There are 2 ways:
>
> 1) - user space configures a bitrate
>     - the kernel calculates with the "struct can_bittiming_const" [1] given
>       by driver and the CAN clock rate the low level timing parameters.
>
>       [1] https://elixir.bootlin.com/linux/v5.18/source/include/uapi/linux/can/netlink.h#L47
>
> 2) - user space configures low level bit timing parameter
>       (Sample point in one-tenth of a percent, Time quanta (TQ) in
>        nanoseconds, Propagation segment in TQs, Phase buffer segment 1 in
>        TQs, Phase buffer segment 2 in TQs, Synchronisation jump width in
>        TQs)
>      - the kernel calculates the Bit-rate prescaler from the given TQ and
>        CAN clock rate
>
> Both ways result in a fully calculated "struct can_bittiming" [2]. The
> driver translates this into the hardware specific BTR values and writes
> the into the registers.
>
> If you know the CAN clock and the bit timing const parameters of the
> slcan's BTR register you can make use of the automatic BTR calculation,
> too. Maybe the framework needs some tweaking if the driver supports both
> fixed CAN bit rate _and_ "struct can_bittiming_const".
>
> [2] https://elixir.bootlin.com/linux/v5.18/source/include/uapi/linux/can/netlink.h#L31
>
> >    b) pre-calculate a huge table of possible bitrates and present them
> >       all to userspace. Sounds horrible, but that's what I did in can327,
> >       haha. Maybe I should have reigned them in a little, to the most
> >       useful values.
>
> If your adapter only supports fixed values, then that's the only way to
> go.
>
> >    c) just limit the bitrates to whatever seems most useful (like the
> >       "S" command's table), and let users complain if they really need
> >       something else. In the meantime, they are still free to slcand or
> >       minicom to their heart's content before attaching slcan, thanks to
> >       your backwards compatibility efforts.
>
> In the early stages of the non-mainline CAN framework we had tables for
> BTR values for some fixed bit rates, but that turned out to be not
> scaleable.

The Microchip CAN BUS Analyzer Tool is another example for fixed bitrates:

https://elixir.bootlin.com/linux/v5.18.14/source/drivers/net/can/usb/mcba_usb.c#L156

So this might be the way to go here too ...

Best regards,
Oliver

> > In short, to me, this isn't a deal breaker for your patch series.
>
> Marc