On Fri, Sep 27, 2024 at 10:04 AM Nayeemahmed Badebade <nayeemahmed.badebade@xxxxxxxx> wrote: > > On Thu, Sep 26, 2024 at 02:34:26PM +0200, Krzysztof Kozlowski wrote: > > On 26/09/2024 13:07, Nayeemahmed Badebade wrote: > > >> > > >>>> not required at the time of booting. > > >>>> Example: drivers/pci/controller/dwc/pci-imx6.c > > >> > > >> Just this one? I don't see anything obvious that can't turn that into a > > >> module, have you tried? What went wrong? > > >> > > > > > > Yes we have tried building it as a module. > > > This driver currently registers abort handler for pci using function > > > hook_fault_code() on arm. This function is not exported and marked with __init > > > tag. So we can't use it post boot. > > > > Then this is something to fix. > > > Thank you for the suggestion. > As per discussion in below link, its mentioned that hooks should be static and > should not change during runtime due to locking support not being there. > So its not safe to export this function to use in modules as per the comments > there. > We would appreciate any suggestions you might have on any possible > alternatives. > > > Also from past attempt made to modularize this driver in community, we saw the > > > hook is not safe to be used post boot. > > > Reference: > > > https://lore.kernel.org/linux-arm-kernel/1454889644-27830-2-git-send-email-paul.gortmaker@xxxxxxxxxxxxx/T/#m8995c6dcc40c54baef0665a7ee16d4209cb59655 The hook implementations have no interaction with the drivers other than being installed by the driver. So move them out of the drivers and the handler can be built-in with the driver as a module. For example, see arch/arm/mach-bcm/bcm_5301x.c. Could possibly combine some implementations. I haven't figured out why imx6 checks 2 different instructions while keystone only handles one. But imx6's implementation being a superset should work for keystone perhaps. Rob
