Re: Thunderbolt: One missing DisplayPort?

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

Hi,

On Tue, May 31, 2022 at 09:45:55PM +0200, Stefan Hoffmeister wrote:
> What could be reasons that the second tunnel is not established on the
> Dell? I read somewhere that Intel hands off the firmware to vendors
> (Dell) who then customize it for their systems? Could the vendor have
> made bad customizations / configurations of that package while
> integrating it?

Probably not a firmware issue.

> I would imagine that plugging in a DisplayPort cable makes the dock
> (firmware) signal something to the notebook (TB firmware) and a
> negotiation will take place. That negotiation fails, otherwise the
> tunnel would be established, and remain established? Is there a means
> to trace the negotiation?

It is all done in firmware but when you plug in DisplayPort cable to the
dock, it generates a hotplug event for that DP OUT adapter and this will
then be handled by the firmware connection manager by establishing a DP
tunnel (if it finds resources).

> FWIW, I have read the phrase "insufficent provision of GPU Interfaces
> to the TB port" (sic, on Reddit), and a lengthy related post at
> https://www.dell.com/community/XPS/Understanding-Thunderbolt-docks-GPU-bandwidth-and-GPU-interfaces/td-p/7678776
> which I will not pretend to understand.
> 
> What I wonder about is whether the "GPU interfaces" situation would be
> reliably discoverable by inspecting ... something ... anything?
> 
> Anyway, my impression, from a layering point of view, is that on the
> stack (my imagination!)
> 
> * notebook hardware
> * firmware (BIOS, Thunderbolt firmware / connection manager, ...)
> * Linux thunderbolt driver
> * Linux graphics drivers: drm / kms (i915 / nvidia / nouveau)
> 
> the graphics drivers are not involved when it comes to building /
> maintaining the Thunderbolt(!) tunnel?

Correct.

> I am also reading "Thunderbolt Alternate Mode encapsulates DisplayPort
> Alternate Mode". To my ears this sounds like "wrap the raw DisplayPort
> Alternate Mode bitstream", just with more bandwidth. Pure "DisplayPort
> Alternate Mode" I can force with success by way of disabling
> Thunderbolt in the BIOS (at the expense of bandwidth -> bad refresh
> rate). And "DisplayPort Alternate Mode" gives me _both_ screens,
> apparently very much scraping along at the max protocol bandwidth,
> with the 4K screen going black (out of sync?) every once in a while.
> 
> Sorry for my rambling, this is an area where I have no expertise.
> 
> Anyway, if those graphics drivers are involved for _Thunderbolt_,
> please do tell me, and I'll venture over to dri-devel.

In case of firmware based connection manager, the Thunderbolt driver
does not do much. Pretty much just the PCIe tunnel authorization and
power management things (and P2P).

IIRC this non-working system had a discrete (NVIDIA?) GPU? It may be
that routing it to the DP IN adapters in the Thunderbolt host router
requires something we don't implement in Linux side yet.

> And given what I see above, is that still "Thunderbolt 4 Certified"
> ("Two 4K displays") in the case of the Dell Inspiron 7610?

This I don't know I would expect Dell testing this, at least with their
own dock.