Re: [PATCH 32/32] PCI: Whitelist Thunderbolt ports for runtime D3

[Mika Westerberg <mika.westerberg@linux.intel.com>]

On Fri, Jul 20, 2018 at 06:23:49PM +0300, Mika Westerberg wrote:
> On Wed, Jul 18, 2018 at 09:30:53PM +0200, Lukas Wunner wrote:
> > Hi Mika,
> > 
> > sorry for the delay!
> > 
> > On Thu, Jun 21, 2018 at 02:13:54PM +0300, Mika Westerberg wrote:
> > > On Sat, Jun 16, 2018 at 09:25:00PM +0200, Lukas Wunner wrote:
> > > > +		/* Even the oldest 2010 Thunderbolt controller supports D3. */
> > > > +		if (bridge->is_thunderbolt)
> > > > +			return true;
> > > 
> > > I have a small concern here. In PC world native PCIe hotplug used with
> > > Thunderbolt comes in two flavors:
> > > 
> > >   - Native PCI hotplug without runtime PM
> > >   - Native PCI hotplug with runtime PM
> > > 
> > > The former works so that even if it uses native PCIe hotplug, the power
> > > management is done so that the Thunderbolt host router is hot-removed
> > > when there is nothing connected (and that brings the power savings).
> > > With the above change we start putting all Thunderbolt PCIe hotplug
> > > ports to D3 runtime. While this probably works (and I tested it on the
> > > same Dell) I don't think Windows does it and it may lead to a path that
> > > has not been tested very thoroughly by OEMs.
> > > 
> > > So I wonder if it makes sense to restrict this particular check to Apple
> > > hardware at this point?
> > 
> > I'm not familiar at all with Windows, so this your call.  Normally I'd say,
> > if there's no known case which forces us to constrain this to Apple, we
> > probably shouldn't be overzealous.  OTOH I imagine you have to deal with
> > Windows and broken BIOSes on a daily basis and can anticipate if problems
> > are to be expected.
> > 
> > The "host router" is the NHI, right?  So IIUC all downstream ports and
> > the upstream port of the TB PCIe switch may go to D3hot, same for the
> > root port.  And your concern is that hotplug addition or removal of the
> > NHI might not work if those ports are in D3hot?
> 
> Host router is the PCIe switch + NHI + xHCI. My concern is that now we
> start putting the downstream hotplug ports of the host router (the ports
> where TBT PCIe daisy chain is to be extented) to D3hot even when the
> whole power management of the "Native PCI hotplug without runtime PM" is
> done pretty much by hot-removing the whole host router if there is
> nothing connected.

I see Bjorn already merged these so please ignore the above. We can
add the Apple check later if this turns out causing problems.