Hi,
On 01-12-17 09:38, Heikki Krogerus wrote:
Hi,
Thanks for taking a look at this..
On Sun, Nov 26, 2017 at 12:23:31PM +0100, Hans de Goede wrote:
Hi Heiko,
On 28-09-17 13:35, Heikki Krogerus wrote:
Hi guys,
The bus allows SVID specific communication with the partners to be
handled in separate drivers for each alternate mode.
Alternate mode handling happens with two separate logical devices:
1. Partner alternate mode devices which represent the alternate modes
on the partner. The driver for them will handle the alternate mode
specific communication with the partner using VDMs.
2. Port alternate mode devices which represent connections from the
USB Type-C port to devices on the platform.
The drivers will be bind to the partner alternate modes. The alternate
mode drivers will need to deliver the result of the negotiated pin
configurations to the rest of the platform (towards the port alternate
mode devices). This series includes API for that, however, not the
final implementation yet.
The connections to the other devices on the platform the ports have
can be described by using the remote endpoint concept [1][2] on ACPI
and DT platforms, but I have no solution for the "platform data" case
where we have neither DT nor ACPI to describe the connections for us.
Sorry about the slow reply, I've been a bit swamped with other stuff,
but now I would like to get back to this.
I've been trying to wrap my head around what you're proposing here and
I see how this can help with implementing display-port alternate mode
support, but I don't see how it is going to help with regular superspeed
USB support / the mux problem.
The problems I see / questions I have are:
1) This seems to be driven by having a bus using svid-s as match functions,
but the standard USB function does not have any svid, or at least currently
does not show as such under e.g. /sys/class/typec/port0/port0-partner
USB is the "normal" mode, not alt-mode. We don't need any specific
driver for the USB mode. In alternate modes, we have to communicate
with the partner using SVID specific messages, and that is what we
need the drivers for.
Ack, but we do still need to control the mux in USB-mode I was under
the impression that the alt-mode drivers would be responsible for
switching the mux to the right component in the graph, if (which
may not be true) the alt-mode drivers indeed will be the ones controlling
the mux, then we need a dummy alt-mode for USB mode and a dummy
alt-mode driver for that.
2) The alt-mode drivers you are suggesting seem to be about 2 things:
a) Alt-mode specific PD communication
b) Telling other components about pin-configs, e.g. telling the i915 driver how
much display port lanes are configured
What this seems to miss a mechanism to control the mux between the "superspeed"
data-pairs on the port and the dp-port pins on the SoC / the superspeed USB
pins on the SoC. Even leaving display-port out of the picture for now we still
need to control the port -> SoC superspeed pins routing which need to be
one of: tristated (default) / normal / upside-down routing.
We will need to deliver the orientation to the GPU/DP drivers and I'm
not supporting that yet in this draft.
That depends, at least with the PI3USB30532 USB switch, the switch/mux
should take care of upside-downness, but I believe I remember some other
hw where the displayport lanes get swapped when upside down, so I guess
that we need to add info like if the upside-downness is handled inside
the mux to the graph info.
I'm preparing a more complete
version of these, and I'll propose something for the orientation as
well, though it is a little bit out side the scope of this series.
That information comes from the device drivers or tcpm, not the
alternate mode drivers.
The USB MUX question is a separate topic, however, the idea of
describing the connections a Type-C port (or any usb port) has to
other components in ACPI and DT, including the USB mux, should be done
using device graph (remote-endpoints) in the future. For the existing
boards we need to figure out somekind of a lookup method to do the
same, and that is the biggest missing thing we still have IMO.
If you can provide proof-of-concept code for the graph stuff +
code to driver the mux from there, etc. perhaps even tested on
a device which has the graph in dt so that it actually works :)
Then I can try to implement a lookup method for my Cherry Trail +
Whiskey Cove PMIC + FUSB302 USB Type-C Controller + PI3USB30532
USB switch boards.
---------------
| Type-C port |
---------------
|
-------
/ MUX \
-----------
/ \
/ \
/ \
---------- ---------
| GPU | / USB MUX \
---------- -------------
/ \
/ \
/ \
------------------- ---------------------
| Host controller | | Device controller |
------------------- ---------------------
Yes this is pretty much what my HW looks like :)
### end Type-c discussion ###
### Related USB device/host mode switch discussion ###
Another problem for USB is even once the right pins on the port are routed to
the right pins on the SoC then the SoC may have an internal mux to route
all the USB pins (both USB-2 and superspeed) to either the host or device
USB controller.
My previous patches for this tried to use the mux framework for this, but that
was nacked because of a misunderstanding how the current mux framemork works.
The current mux framework is based on the notion of there being a shared
bus between e.g. the SoC and various devices, where the devices are not
directly addressable from the bus, but there is a mux in between which
connects the SoC to the device it wants to talk to. So the SoC can reach
all connected devices, but only one add at a time. To make this clear
a driver talking to one of the devices needs to switch the mux to a
specific device and then release the mux when it is done. If another
driver tries to switch the mux before it is released its request will
block until the mux gets released.
If the framework was designed for only such specific cases, I'm not
sure it should be called "mux framework" at all.
I think it was mainly designed to model (ARM + devicetree) boards
with i2c muxes on them.
With USB OTG / Type-C things are different, there is only one device
connected, and depending on which type of device is detected as being
connected we need to connect the OTG / Type-C port pins to the
corresponding controller inside the SoC. There is no "time-shared"
access to multiple devices at the same time. Thus the current mux
subsys is a poor match for the needs for USB OTG / Type-C.
One option I've been considering is to introduce a new usb_mux subsys
when I get around to this (hopefully sometime the coming month) which
will be modelled after the mux subsys, but without the "time-sharing"
concepts and thus without the need to release a mux setting before
the mux can be switched to route the data else where.
Wasn't that basically what Baolu proposed [1] ?
More or less, that seems to lack anyway to lookup a mux device though,
typically you will want to have one driver for the mux and another
which decides what the role should be, that second driver needs a way
to lookup the mux, so that it can then actually call portmux_switch()
(in Baolu's code) on the mux to switch its role.
I really hoped we had a generic mux framework instead of usb specific
one. How about if you call it "demux framework" :-)
I like Baolu's portmux name, and I like the over structure from the
existing mux framework, Peter Rosin the mux framework maintainer does
not really want to extend the existing mux framework for the Type-C
use-case it seems and perhaps more importantly the DT bindings for
the existing framework are a poor match for Type-C I think, so
maybe as part of the graph work you can do a new portmux; or
graphmux core (this can be quite small) which the Type-C code can
use and which ties into the graph DT bindings ?
In any case, we still need a way to describe the connections. I don't
think we should propose yet an other framework specific lookup method.
There really should be a generic "device graph for board files"
solution that we could use also here. Well, ideally.
Ack, for me this is a sparetime project and I don't think having
2 chefs in the kitchen is going to help here. So I'm just going to
wait and see what you come up with. If you've something which is
testable even if I still need to hack together the lookup stuff
let me know and I will try to get it going for my boards / use-case.
Regards,
Hans
My primary target here is using this on devices with a micro-usb
connector, traditionally on ARM systems the mux in these devices has
been hidden inside the phy-driver. But on x86 systems:
a) The phy is setup from firmware and we don't have a phy driver; and
b) Often the mux is not part of the phy at all, instead we have a
separate mux (sometimes in separate IC outside the SoC) driven by
gpios or i2c.
I think we could extend this usb_mux subsys easily to also handle the
mux parts of type-c (specifically controlling mux ICs such as the
PI3USB30532 USB switch).
[1] https://lkml.org/lkml/2016/6/1/980
Thanks,
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
