[Skiboot] [PATCH v2 2/2] prd: PRD framework
benh at au1.ibm.com
Thu Feb 12 17:29:58 AEDT 2015
On Thu, 2015-02-12 at 14:15 +0800, Jeremy Kerr wrote:
> Hi Ben,
> > Jeremy, that changes things a bit for us... We can keep the code simpler
> > still I think, but we probably do need some kind of queue. The code is
> > simpler because userspace will only process one at a time and we
> > can bake that in the design on our side.
> We do queue in the Linux layer - we'll keep a list of "unread" messages
> (via the chardev interface). Is there any issue with it doing it in the
> kernel rather than firmware?
Well, we sort-of have to in fw no ?
IE, we get the attn irq, we *have* to mask it or it's going to come back
again and again and hang that cpu. So we also need to process it and
create a message. Now sending a message to linux is an asynchronous
process, we set an OPAL event and eventually Linux will read the
message, so we might already have one pending, we need to queue up
another one. I don't think we can get away without this.
More information about the Skiboot