Re: [PATCH 2/3] can: esd_usb: Improved behavior on esd CAN_ERROR_EXT event (2)

[Marc Kleine-Budde]

On 23.01.2023 15:47:22, Frank Jungclaus wrote:
> On Thu, 2022-12-22 at 11:21 +0900, Vincent MAILHOL wrote:
> > On Thu. 22 Dec. 2022 at 03:42, Frank Jungclaus <Frank.Jungclaus@xxxxxx> wrote:
> > > On Tue, 2022-12-20 at 14:49 +0900, Vincent MAILHOL wrote:
> > > > On Tue. 20 Dec. 2022 at 06:29, Frank Jungclaus <frank.jungclaus@xxxxxx> wrote:
> > > > > Started a rework initiated by Vincents remarks "You should not report
> > > > > the greatest of txerr and rxerr but the one which actually increased."
> > > > > [1]
> > > > 
> > > > I do not see this comment being addressed. You are still assigning the
> > > > flags depending on the highest value, not the one which actually
> > > > changed.
> > > 
> > > 
> > > Yes, I'm assigning depending on the highest value, but from my point of
> > > view doing so is analogue to what is done by can_change_state().
> > 
> > On the surface, it may look similar. But if you look into details,
> > can_change_state() is only called when there is a change on enum
> > can_state. enum can_state is the global state and does not
> > differentiate the RX and TX.
> > 
> > I will give an example. Imagine that:
> > 
> >   - txerr is 128 (ERROR_PASSIVE)
> >   - rxerr is 95 (ERROR_ACTIVE)
> > 
> > Imagine that rxerr then increases to 96. If you call
> > can_change_state() under this condition, the old state:
> > can_priv->state is still equal to the new one: max(tx_state, rx_state)
> > and you would get the oops message:
> > 
> >   https://elixir.bootlin.com/linux/latest/source/drivers/net/can/dev/dev.c#L100
> > 
> > So can_change_state() is indeed correct because it excludes the case
> > when the smallest err counter changed.
> > 
> > > And
> > > it should be fine, because e.g. my "case ESD_BUSSTATE_WARN:" is reached
> > > exactly once while the transition from ERROR_ACTIVE to
> > > ERROR_WARN. Than one of rec or tec is responsible for this
> > > transition.
> > > There is no second pass for "case ESD_BUSSTATE_WARN:"
> > > when e.g. rec is already on WARN (or above) and now tec also reaches
> > > WARN.
> > > Man, this is even difficult to explain in German language ;)
> > 
> > OK. This is new information. I agree that it should work. But I am
> > still puzzled because the code doesn't make this limitation apparent.
> > 
> > Also, as long as you have the rxerr and txerr value, you should still
> > be able to set the correct flag by comparing the err counters instead
> > of relying on your device events.
> > 
> 
> I agree, this would be an option. But I dislike the fact that then
> - beside the USB firmware - there is a second instance which decides on
> the bus state. I'll send a reworked patch which makes use of
                      ^^^^^^^^^^^^^^^^^^^^^
> can_change_state(). Hopefully that will address your concerns ;)
> This also will fix the imperfection, that our current code e.g. does
> an error_warning++ when going back in direction of ERROR_ACTIVE ...

Not taking this series, waiting for the reworked version.

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 |

Attachment: signature.asc
Description: PGP signature