[Skiboot] [PATCH 1/9] npu2: Split out common helper functions into separate file
alistair at popple.id.au
Mon Jan 8 16:18:49 AEDT 2018
On Monday, 8 January 2018 1:39:59 PM AEDT Stewart Smith wrote:
> Alistair Popple <alistair at popple.id.au> writes:
> > On Friday, 22 December 2017 3:22:32 PM AEDT Stewart Smith wrote:
> >> Andrew Donnellan <andrew.donnellan at au1.ibm.com> writes:
> >> > On 19/12/17 11:37, Alistair Popple wrote:
> >> >>> +bool is_p9dd1(void)
> >> >>
> >> >> Does OpenCAPI still need this? If not we should just drop it and DD1 support
> >> >> along with it as it's very unlikely anything still works there.
> >> >
> >> > OpenCAPI has no DD1 support whatsoever. If you want to drop DD1 support
> >> > for NVLink then I have no objections!
> >> I have no objection to this.
> >> Anyone currently attempting NVLink on DD1 gets all the checkstops they deserve.
> > Yep. I'm going to pull it out. Although it may just lead to xstops on boot for
> > DD1 ... is that something we care about or should we still detect DD1 and
> > disable NVLink (which would prevent any checkstops)?
> Detect and disable sounds good. just skip all nvlink init on
> dd1... although perhaps we'd have to rip some stuff out of the device
> (or not add it)
Not adding the NPU2 to the device-tree would be sufficient. Removing
old_witherspoon_probe() would achieve this as I doubt there is any Hostboot that
supports DD1 and passes valid HDAT information for NPU device-node creation.
More information about the Skiboot