On 06.09.2021 23:17:40, Vincent MAILHOL wrote:
> > > To prevent this from happening, we do a local copy of can_priv, work
> > > on it, an copy it at the very end of the function (i.e. only if all
> > > previous checks succeeded).
> >
> > I don't like the optimization of using a static priv. If it's too big to
> > be allocated on the stack, allocate it on the heap, i.e. using
> > kmemdup()/kfree().
> 
> The static declaration is only an issue of coding style, correct?

I don't know (but I haven't checked) if the coding style doc says
anything about that.

> Or is there an actual risk of doing so?

As you pointed out, this relies on the serialization of the changelink
callback by the networking stack. There's no sane way in C to track this
requirement in the networking stack, so I don't want to have any
roadblocks and/or potential bugs in the CAN code. Marking a variable as
static places it in the BSS section, right? This mean, the memory is
always "used", even if not setting the bitrate.

> This is for my understanding, I will remove the static
> declaration regardless of your answer.

tnx

> On my x86_64 machine, sizeof(priv) is 448 and if I declare priv on the stack:
> | $ objdump -d drivers/net/can/dev/netlink.o | ./scripts/checkstack.pl
> | 0x00000000000002100 can_changelink []:            1200
> 
> So I will allocate it on the heap.

Sounds reasonable.

> N.B. In above figures CONFIG_CAN_LEDS is *off* because that driver
> was tagged as broken in:
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=30f3b42147ba6f29bc95c1bba34468740762d91b

ok - BTW: I think we can remove LEDs support now, it's marked as broken
for more than 3 years.

> > > Once this done, there is no more need to have a temporary variable for
> > > a specific parameter. As such, the bittiming and data bittiming (bt
> > > and dbt) are directly written to the temporary priv variable.
> > >
> > > Finally, function can_calc_tdco() was retrieving can_priv from the
> > > net_device and directly modifying it. We changed the prototype so that
> > > it instead writes its changes into our temporary priv variable.
> >
> > Is it possible to split this into a separate patch, so that the part
> > without the tdco can be backported more easily to older kernels not
> > having tdco? The patch fixing the tdco would be the 2nd patch...
> 
> ACK. I will send a v3 with that split.

Thanks for helping taking care of the LTS kernels!

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 |