On Wednesday, August 17, 2022 8:35 AM, Jia He wrote: > Previous, there is only one edac can be registering in the EDAC core. After > ghes_edac is modularized, the registering of ghes devices may be deferred > when ghes_edac is loaded. Prevent the chipset-specific edac drivers from > loading after ghes_edac is registered, which allow the edac drivers to > get selected in e.g. HPE platform. ... > @@ -1382,6 +1395,7 @@ static int ghes_probe(struct platform_device > *ghes_dev) > platform_set_drvdata(ghes_dev, ghes); > > ghes->dev = &ghes_dev->dev; > + set_ghes_devs_registered(false); This does not look right to me. The condition of using ghes_edac is (ghes-present) && (ghes-preferred), where: - ghes-present is latched on ghes_probe() - ghes-preferred is true if the platform-check passes. ghes_get_device() introduced in the previous patch works as the ghes-preferred check. We cannot use ghes_edac registration as the ghes-present check in this scheme since it is deferred to module_init(). I'd suggest the following changes: - Update ghes_get_device() to check a flag latched on ghes_probe(). - Remove 'force' argument from ghes_get_device(). ghes_edac_init() is too late to set this flag. Also, other edac drivers should not be able to control this flag. I think this force_load kernel option needs to belong to ghes with this scheme. - ghes_get_device() exposes 'ghes_devs' to chipset-specific edac drivers, which should be internal to ghes. ghes_get_device() may be renamed to something like ghes_edac_preferred() which returns true / false. Toshi
