Re: [PATCH] PCI: Support SRIOV on Legacy EndPoint device

[Bjorn Helgaas]

On Thu, Feb 04, 2016 at 04:21:08PM +0000, Zytaruk, Kelly wrote:
> > -----Original Message-----
> > From: Bjorn Helgaas [mailto:helgaas@xxxxxxxxxx]
> > Sent: Thursday, February 04, 2016 10:14 AM
> > To: Zytaruk, Kelly
> > Cc: bhelgaas@xxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx; linux-
> > kernel@xxxxxxxxxxxxxxx; Alex Williamson; Yu Zhao
> > Subject: Re: [PATCH] PCI: Support SRIOV on Legacy EndPoint device
> > 
> > [+cc Alex, Yu (participants in previous discussion)]
> > 
> > Hi Kelly,
> > 
> > On Thu, Feb 04, 2016 at 09:48:26AM -0500, Kelly Zytaruk wrote:
> > > Some AMD GPUs have hardware support for grapics SRIOV.
> > > If the GPU has a display output then the GPU needs to support Legacy VGA
> > operation.
> > > If CLASS_CODE = VGA then the device should have a Port Type = Legacy
> > EndPoint.
> > > Therefore in order to enable SRIOV on a GPU with a display output
> > LEGACY_END_POINT is supported as a valid Port Type.
> > >
> > > Signed-off-by: Kelly Zytaruk <kelly.zytaruk@xxxxxxx>
> > 
> > We had an interesting discussion about this two years ago:
> > http://lkml.kernel.org/r/B756807489D6244883AC0B799A6EEC15EAB2E8@store
> > xdag02.amd.com
> 
> Unfortunately 2 years ago I couldn't complete your request as it
> would have disclosed Information about an unannounced technology
> that we were working on.  We have recently made that technology
> public and I can now send you the requested information.

> > Please include a reference to that discussion in your changelog.  In that
> > discussion, I also asked for some details (dmesg and lspci info) that motivate the
> > change, so please collect and add a reference to them as well.
> 
> The information that you ask for is included below.  I have
> abbreviated it so that this does not become a huge email.  I can
> send full logs if you want them.

Can you open a bugzilla at http://bugzilla.kernel.org and attach the full
logs there?  Then we can include the URL in your patch changelog.

> > It's not clear to me why we check the device type at all.  If
> > there's no spec restriction on the types of devices that can have
> > an SR-IOV capability, we should consider removing the test
> > altogether (Alex mentioned this possiblity in the earlier
> > discussion).
> 
> I am as well not clear why the check is in there.  I would be just
> as happy either adding TYPE_LEG_END or removing the check all
> together.  I don't know what the side effects of removing the check
> would be. I don't have any sriov devices other than a graphics card
> to test with so I wouldn't be able to test other scenarios.

If we don't have a reason to do the check, I think we should just
remove it altogether.

Bjorn