On Wed, Sep 15, 2021 at 10:09:36AM -0700, Saravana Kannan wrote: > Intended for 5.15. > > Cc: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> > Cc: Andrew Lunn <andrew@xxxxxxx> > Cc: Vladimir Oltean <olteanv@xxxxxxxxx> > > v1->v2: > - Added a few Reviewed-by and Tested-by tags > - Addressed Geert's comments in patches 3 and 5 > - Dropped the fw_devlink.debug patch > - Added 2 more patches to the series to address other fw_devlink issues > > v2->v3: > - Split the logging/debug changes into a separate series I have taken this now into my tree. It fixes the real problem where drivers were making the wrong assumption that if they registered a device, it would be instantly bound to a driver. Drivers that did this were getting lucky, as this was never a guarantee of the driver core (think about if you enabled async probing, and the mess with the bus specific locks that should be preventing much of this) With this new flag, we can mark these drivers/busses that have this assumption and work to solve correctly over time. The issue with using a "generic vs. specific" driver is a bit related, I'm amazed that a subsystem actually implemented it this way, others of us have been avoiding this for a very long time due to the complexity involved when things are built as modules. thanks, greg k-h
