On 02/07/2011 03:45 PM, Benjamin Herrenschmidt wrote:
In my previous reply I said that "it is not so much as a need as it is a
potential simplification." After further reflection, I don't think that
is completely true. As we get into AMP systems with higher core counts,
then implementing this functionality using the existing
"protected-sources" implementation versus the new "pic-no-reset" work is
going to be harder to maintain.
I'm not arguing that your approach isn't more suitable for AMP systems,
I just want to leave the existing protected-sources mechanism alone. I'm
not opposing adding a new, better, mechanism for newer platforms.
Is the mechanism mentioned earlier of having "protected-sources" as a
synonym for "pic-no-reset" not suitable? Or would you like the current
protected sources implementation left completely intact?
However, I'd name it differently. "pic-no-reset" doesn't carry enough
meaning in that case. What we want to point out here is that the PIC
has been pre-initialized.
Another option, which may be cleaner, is to stick to "no-reset" (no need
for pic- prefix) and make it do just that (prevent the reset), and then
It originally was "no-reset", but that was considered too broad. [1] :)
use a positive variant of "protected-sources", call it
"allowed-sources". Maybe even make it a series of ranges. Then have the
MPIC only access these.
That would work, but I still don't like having to mention this
information twice in the device tree. All the sources encoded in the
various "interrupts" properties _are_ the allowed sources, right?
I think this is more robust as it would also prevent "accidental" use of
the wrong sources (bad device-tree, drivers that let you muck around
with irq numbers, etc...).
That would be nice. All though, it may not be as helpful as it sounds.
There is as much of a risk that someone will botch the
"allowed-sources" property as there is they will botch the "interrupts"
property. We could perhaps still preform these checks without the extra
property: if a source is not mentioned in an interrupts property, then
it is not allowed.
Cheers,
Ben.
[1] http://lists.ozlabs.org/pipermail/linuxppc-dev/2011-February/088244.html
--
Meador Inge | meador_inge AT mentor.com
Mentor Embedded | http://www.mentor.com/embedded-software
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev
