On Wed, 24 Sep 2014, Yasuaki Ishimatsu wrote:

I cannot imagine why the feature makes x86 irq architecture much more clear
and more easy to maintain in this description. Of course, I have read the
following threads:

https://lkml.org/lkml/2014/9/11/101

Here is the long version of the idea:

     https://lkml.org/lkml/2014/8/26/707

Short version is:

Separate the irq handling entities in a clear layered way instead of
having vector/[io]apic specific knowledge/data in places like msi,
remap etc.

Thanks,

	tglx