On Mon, Sep 20, 2021 at 08:49:21PM +0100, Russell King (Oracle) wrote: > On Mon, Sep 20, 2021 at 09:41:47PM +0200, Andreas Schwab wrote: > > On Sep 20 2021, Russell King (Oracle) wrote: > > > Therefore, this change breaks module autoloading. > > Reverting this change breaks module autoloading. > No. > Module autoloading worked before. Then someone probably noticed a > problem, and thought they'd fix it by changing how the module alias > strings SPI provides are produced. In fixing it, they broke existing > setups that have worked for years. To be clear Russell is absolutely right here. These changes have broken module autoloading for spi-nor on device tree based systems which don't list the jedec,spi-nor fallback compatible, and quite likely for some other drivers/systems that were also relying on the fallback compatible mechanism in a similar way. They will also have fixed systems where we weren't autoloading based on DT compatibles but the broken systems are still broken and regardless of the quality of the DTs that those systems have DT is an ABI so they have to continue to work. Ideally we'll be able to keep both sets of drivers working (and I think we probably should just get all the compatibles listed in the spi-nor driver for the sake of robustness if nothing else), unfortunately this wasn't noticed until after v5.14 was released so we might now have systems relying on the new behaviour too which complicates things a bit. Still, if we can't get it fixed reasonably promptly (this week say, for the next -rc) I expect I'll revert as Russell has suggested.
Attachment:
signature.asc
Description: PGP signature
