[PATCH 2/3] can: rockchip: add RK3588 CAN-FD support

[Heiko Stübner]

Am Mittwoch, 1. Juli 2026, 14:29:22 Mitteleuropäische Sommerzeit schrieb 💫.220:
> Hi Heiko, Marc,
> 
> thanks for the review.
> 
> > please use a real name, not an alias.
> 
> I will use my real name in future revisions.
> 
> > Funnily enough, we seem to have worked on the same topic
> > at the same time :-)
> >
> > https://lore.kernel.org/lkml/20260630164336.3444550-1-heiko@sntech.de/
> 
> I missed Heiko's series before sending mine, sorry for the noise.
> (But it really is a very interesting coincidence.)
> Since the series overlap, I am happy to base further work on Heiko's
> series, or to let Heiko fold the RK3588 RX_FIFO_CNT bitfield change into
> his v2 if that is preferred.

I don't think we need two people working on this and you did the better
investigation into the differences, so you should get the credit :-)

So I guess my preference would be to:
- pick up the erratum 6 into your patch
- add my haikou patch to the series - that way we also get a user
- handle Krzysztof's comment from
  https://lore.kernel.org/linux-rockchip/20260701-sensible-cryptic-ocelot-58035b@quoll/

  as both our bindings have the same structure, so I guess it should be

    oneOf:
       - enum:
             rockchip,rk3568v2-canfd
             rockchip,rk3588v2-canfd
       - items:
           - const: rockchip,rk3568v3-canfd
           - const: rockchip,rk3568v2-canfd

And submit that as v2.


> For RX_FIFO_CNT, I found the bitfield difference by reading Rockchip's
> vendor kernel 6.1 driver and comparing the CAN support for RK3568 and
> RK3588. The vendor driver uses different RX FIFO count bitfields for the
> two SoCs, and my testing on RK3588v2 also indicates that bits 7:5 are
> needed. I can add a short note about this in the commit message or
> code comment.

It's already in the commit message, so that should be fine


> One more question about RKCANFD_QUIRK_CANFD_BROKEN: in my RK3588v2 test
> setup the two known CAN-FD trigger frames did not cause an Error
> Interrupt or Error-Warning. I also ran a 12 hour CAN-FD stress test with
> can0/can1 directly connected, 200 MHz CAN clock, 500 kbit/s nominal
> bitrate and 1 Mbit/s data bitrate. That test included periodic
> transmission of the two CANFD_BROKEN frames, variable DLC CAN-FD frames,
> CAN-FD+BRS+EFF load, and a canfdtest run with 19,417,129 frames without
> data mismatch.
> 
> Would it make sense to leave RKCANFD_QUIRK_CANFD_BROKEN disabled for
> RK3588v2, or have you seen this issue on RK3588 as well?

I was more going by the fact that even Rockchip removed every mention of
the -FD from all documentation, so was assuming it still being broken as
before. But if it actually works, then personally I'm more than fine with
enabling CAN-FD :-D .

I guess Marc might have more insight where the FD issue triggered on
the RK3568.


Heiko