[PATCH]: PCI Error Recovery Implementation

John Rose johnrose at austin.ibm.com
Wed Jun 1 08:13:11 EST 2005


Hi Linas-

> What would be the correct way of forcing the rpaphp.o module to get
> loaded, 

The burden is on the distro/user to either compile-in or load the PCI
Hotplug module.  This is the case with or without your patch, and
regardless of the ultimate solution to the EEH problem.

> and what would be the correct way of invoking functions in that
> module?

Arch/ppc64/kernel/eeh.c could have function pointers for enable/disable
slot.  Something like:
int (*hp_disable_slot)(struct pci_bus *bus) = NULL;
int (*hp_enable_slot)(struct pci_bus *bus) = NULL;

These could either be exported, or be static and accompanied by small
accessor functions.  The RPA hotplug module could set these pointers at
module init, either directly or through accessors.  If these aren't set
at runtime, the module isn't loaded.

This puts the eeh footprint on rpaphp at 4 lines, and leaves the EEH
implementation in an eeh file.

Thoughts?
John




More information about the Linuxppc64-dev mailing list