On 1/27/23 11:41, Krzysztof Kozlowski wrote: > On 27/01/2023 17:22, Sean Anderson wrote: >> On 1/27/23 11:15, Krzysztof Kozlowski wrote: >>> On 27/01/2023 17:11, Sean Anderson wrote: >>>>>>>> >>>>>>>> .../boot/dts/freescale/fsl-ls1046a-rdb.dts | 112 ++++++++++++++++++ >>>>>>>> drivers/phy/freescale/Kconfig | 1 + >>>>>>> >>>>>>> The phy driver Kconfig change shouldn't be part of this patch. >>>>>> >>>>>> I put it here for bisectability, since this is the point where we need >>>>>> to enable it. But I can do this in a separate patch if you want. >>>>> >>>>> From DT ABI perspective, it's already broken anyway if you need to change >>>>> kernel and DT atomically. >>>> >>>> AIUI new kernels must work with old device trees, but new device trees need not >>>> work with old kernels. So a change like this is fine, since the kernel won't >>>> touch the serdes if it isn't supplied. >>> >>> You used the argument "bisectability". If the patchset is not >>> bisectable, the ABI is broken. >> >> Well, because Shawn wants it in a separate patch I am just going to enable >> the driver by default on Layerscape before adding the device nodes. That way we have >> >> 1. Base state, driver not enabled and node is disabled >> 2. Driver enabled but not used because the node is disabled >> 3. Driver enabled and bound to node >> >> So there is never a case where the node is bound but the driver isn't enabled >> (which would cause the ethernet drivers to fail to probe). > > Then there is no bisectability issues and the Kconfig patch should have > been squashed here... Mentioning bisectability and that squash just > confuses. The Kconfig is currently squashed, but I am going to split it off as requested. --Sean
