multiple MADT issue & acpi_get_table()

[Len Brown <lenb@kernel.org>]

Alexey, Bob,

There are boxes with multiple MADT's in the RSDT,
and proof that Linux' current policy of choosing the 1st one
does not work around this BIOS bug properly.

http://bugzilla.kernel.org/show_bug.cgi?id=7465

Before the ACPICA table update, I had a patch in 7465
that made this situation verbose, and added a boot param so that
we could switch to using the last MADT and have a knob if
that didn't always work.

That patch is no good in 2.6.21 as ACPICA's acpi_get_table()
now owns finding the table instance.

Also, I see code in processor_core.c that is uses acpi_get_table()
to find the MADT -- so my patch in the bug report above probably
was incomplete anyway, as in some cases the table code would look at
one instance and the processor driver still looked a the 1st instance...

What do you suggest we do with acpi_map_table() to handle
the fact that when there are multiple MADT's, we probably always
want the last one?  Further, we need to issue a warning when this happens,
and we need a hook for a boot param in case it doesn't work for everybody.

Note that I've only seen this issue with the MADT.

thanks,
-Len