[Skiboot] [PATCH 3/3] ... to prevent any potential poller recursions during lid load.
Ananth N Mavinakayanahalli
ananth at in.ibm.com
Tue Feb 17 17:42:04 AEDT 2015
On Tue, Feb 17, 2015 at 09:50:51AM +0530, Ananth N Mavinakayanahalli wrote:
> On Tue, Feb 17, 2015 at 08:18:57AM +1100, Benjamin Herrenschmidt wrote:
> > On Mon, 2015-02-16 at 20:45 +0530, Ananth N Mavinakayanahalli wrote:
> > > + /* Preload hostservices lids */
> > > + hservices_lid_preload();
> > > +
> > Definitely not the right place to call this.
> > My worry is I think we can get the OCC messages at any time after the
> > OPL, but the preload itself can take time operating on FSP and so must
> > be done after FSP is ready.
> > Can you check with the FSP people when exactly we can start getting the
> > startOCC messages ? Is it after we set the HV state ? We need to stick
> > the preload just before that.
> Sure, will check.
The FSP apparently can service a FETCH_SP_DATA after HV functional --
end of OPL. We can stick the preload in there... but... the FSP folks
aren't sure about the earliest time we can get a OCC load; I will write
to the OCC folks to get that info.
End of OPL is significantly early wrt loading other lids... so we will
not hit the concurrency seen now.
More information about the Skiboot