Hi Stefan, > Hi Andrew, > > Am 16.04.25 um 23:58 schrieb Andrew Lunn: > >>> - eth_switch: switch@800f8000 { > >>> - reg = <0x800f8000 0x8000>; > >>> + eth_switch: switch@800f0000 { > >>> + compatible = "nxp,imx28-mtip-switch"; > >>> + reg = <0x800f0000 0x20000>; > >>> + interrupts = <100>, <101>, <102>; > >>> + clocks = <&clks 57>, <&clks 57>, <&clks > >>> 64>, <&clks 35>; > >>> + clock-names = "ipg", "ahb", "enet_out", > >>> "ptp"; status = "disabled"; > >> from my understanding of device tree this file should describe the > >> hardware, not the software implementation. After this change the > >> switch memory region overlaps the existing mac0 and mac1 nodes. > >> > >> Definition in the i.MX28 reference manual: > >> ENET MAC0 ENET 0x800F0000 - 0x800F3FFF 16KB > >> ENET MAC1 ENET 0x800F4000 - 0x800F7FFF 16KB > >> ENT Switch SWITCH 0x800F8000 - 0x800FFFFF 32KB > >> > >> I'm not the expert how to solve this properly. Maybe two node > >> references to mac0 and mac1 under eth_switch in order to allocate > >> the memory regions separately. > > I get what you are saying about describing the hardware, but... > > > > The hardware can be used in two different ways. > > > > 1) Two FEC devices, and the switch it left unused. > > > > For this, it makes sense that each FEC has its own memory range, > > there are two entries, and each has a compatible, since there are > > two devices. > > > > 2) A switch and MAC conglomerate device, which makes use of all > > three blocks in a single driver. > > > > The three hardware blocks have to be used as one consistent whole, > > by a single driver. There is one compatible for the whole. Given the > > ranges are contiguous, it makes little sense to map them > > individually, it would just make the driver needlessly more complex. > > > > It should also be noted that 1) and 2) are mutually exclusive, so i > > don't think it matters the address ranges overlap. Bad things are > > going to happen independent of this if you enable both at once. > > > > Andrew > i wasn't aware how critical possible overlapping memory regions are. > I was just surprised because it wasn't mention in the commit message. > As long as everyone is fine with this approach, please ignore my last > comment. > Just for the record - there was an attempt to "re-use" FEC enet driver in switch [1], but this approach has been rejected as one not very robust and "clear by design" (i.e. similar to cpsw_new.c driver). And I do agree with Andrew - that approach presented in this patch set is the correct one. Links: [1] - https://source.denx.de/linux/linux-imx28-l2switch/-/commits/imx28-v5.12-L2-upstream-switchdev-RFC_v1 > Regards > Best regards, Lukasz Majewski -- DENX Software Engineering GmbH, Managing Director: Erika Unter HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de