MFC of r180753: ABI problems?

[M. Warner Losh]

In message: <200808230742.10902.jhb at freebsd.org>
            John Baldwin <jhb at freebsd.org> writes:
: On Saturday 23 August 2008 02:42:09 am M. Warner Losh wrote:
: > In message: <200808212351.13464.max at love2party.net>
: >
: >             Max Laier <max at love2party.net> writes:
: > : Hi,
: > :
: > : I'm wondering how to merge r180753 to stable/7 as luoqi@ has indicated
: > : that he doesn't have time to take care of it right now.
: > :
: > : It seems that changing the size of pcicfgregs (aka struct pcicfg) which
: > : is part of struct pci_devinfo is out of the question, right?  Ideas where
: > : to store the HT related state or how to avoid storing the state are
: > : welcome.
: > :
: > : The merge result is attached for reference.  This fix is essential for
: > : many nforce based boards from ASUS which are rather common, I'm afraid. 
: > : So it would be good to have this in 7.1/6.4, I think.
: >
: > I think this is OK.
: >
: > pcicfgregs is an internal to pci implementation detail.  You've added
: > it at the end, so any leakage of the offsets won't matter.  All
: > subclasses of pci would be affected.  Internal to the kernel isn't all
: > that interesting, since they are all compiled at the same time.  This
: > would only matter for modules.  Cardbus and acpi would be the only
: > modules affected.  That would mean you couldn't boot a 7.0 kernel with
: > a 7.1 set of modules or vice versa.  I'm not sure that is actually
: > going to work anyway...
: 
: ACPI (and OFW's) PCI bus code isn't going to care, and I doubt cardbus is 
: either.  Hmm, actually, cardbus doesn't, but ACPI actually does (acpi_pci 

CardBus' does because it creates a slightly larger pcicfgreg per device...

: uses its own extended ivars for PCI devices to cache ACPI handles).  That 
: said, this particular ABI was actually broken earlier by MSI (though I didn't 
: realize it at the time. :( ).

Warner