[PATCH v6] unstable/drm-lease: DRM lease protocol support

[Pekka Paalanen]

On Tue, 30 Jul 2019 08:53:31 -0400
Drew DeVault <sir at cmpwn.com> wrote:

> From: Marius Vlad <marius.vlad at collabora.com>
> 
> DRM leasing is a feature which allows the DRM master to "lease" a subset
> of its DRM resources to another DRM master via drmModeCreateLease, which
> returns a file descriptor for the new DRM master. We use this protocol
> to negotiate the terms of the lease and transfer this file descriptor to
> clients.
> 
> In less DRM-specific terms: this protocol allows Wayland compositors to
> give over their GPU resources (like displays) to a Wayland client to
> exclusively control.
> 
> The primary use-case for this is Virtual Reality headsets, which via the
> non-desktop DRM property are generally not used as desktop displays by
> Wayland compositors, and for latency reasons (among others) are most
> useful to games et al if they have direct control over the DRM resources
> associated with it. Basically, these are peripherals which are of no use
> to the compositor and may be of use to a client, but since they are tied
> up in DRM we need to use DRM leasing to get them into client's hands.
> 
> Signed-off-by: Marius Vlad <marius.vlad at collabora.com>
> Signed-off-by: Drew DeVault <sir at cmpwn.com>

Hi Drew,

I seem to recall that you didn't want to add multi-DRM-device support
here just yet and go first with just one implied DRM device. That is
ok, but would be nice to have a TODO note somewhere near the top in the
XML file saying that this will be re-designed to support multiple DRM
devices at some point.

> ---
> When implementing this for Xwayland, we realized that we would really
> like to be able to obtain a lease with zero connectors. The kernel does
> not support this today, but adding support shouldn't be especially
> difficult. v6 changes the protocol accordingly to allow for leases with
> zero connectors, though on today's kernels this will fail vaugely with
> the finished event.

Can you point to the discussion or elaborate here on when a zero
connector lease would be useful?

I checked the Xwayland discussion and didn't see it there. I remember
some old talk about giving out a no-resources lease first for
discovering DRM resources to lease, but that didn't work because
non-leased resources would not be visible either.

> 
>  Makefile.am                                  |   1 +
>  unstable/drm-lease/README                    |   5 +
>  unstable/drm-lease/drm-lease-unstable-v1.xml | 246 +++++++++++++++++++
>  3 files changed, 252 insertions(+)
>  create mode 100644 unstable/drm-lease/README
>  create mode 100644 unstable/drm-lease/drm-lease-unstable-v1.xml
> 
> diff --git a/Makefile.am b/Makefile.am
> index 345ae6a..d9fff89 100644
> --- a/Makefile.am
> +++ b/Makefile.am
> @@ -4,6 +4,7 @@ unstable_protocols =								\
>  	unstable/pointer-gestures/pointer-gestures-unstable-v1.xml		\
>  	unstable/fullscreen-shell/fullscreen-shell-unstable-v1.xml		\
>  	unstable/linux-dmabuf/linux-dmabuf-unstable-v1.xml			\
> +	unstable/drm-lease/drm-lease-unstable-v1.xml				\
>  	unstable/text-input/text-input-unstable-v1.xml				\
>  	unstable/text-input/text-input-unstable-v3.xml				\
>  	unstable/input-method/input-method-unstable-v1.xml			\
> diff --git a/unstable/drm-lease/README b/unstable/drm-lease/README
> new file mode 100644
> index 0000000..16f8551
> --- /dev/null
> +++ b/unstable/drm-lease/README
> @@ -0,0 +1,5 @@
> +Linux DRM lease
> +
> +Maintainers:
> +Drew DeVault <sir at cmpwn.com>
> +Marius Vlad <marius-cristian.vlad at nxp.com>
> diff --git a/unstable/drm-lease/drm-lease-unstable-v1.xml b/unstable/drm-lease/drm-lease-unstable-v1.xml
> new file mode 100644
> index 0000000..083d004
> --- /dev/null
> +++ b/unstable/drm-lease/drm-lease-unstable-v1.xml
> @@ -0,0 +1,246 @@
> +<?xml version="1.0" encoding="UTF-8"?>
> +<protocol name="drm_lease_unstable_v1">
> +  <copyright>
> +    Copyright © 2018 NXP
> +    Copyright © 2019 Status Research & Development GmbH.
> +
> +    Permission is hereby granted, free of charge, to any person obtaining a
> +    copy of this software and associated documentation files (the "Software"),
> +    to deal in the Software without restriction, including without limitation
> +    the rights to use, copy, modify, merge, publish, distribute, sublicense,
> +    and/or sell copies of the Software, and to permit persons to whom the
> +    Software is furnished to do so, subject to the following conditions:
> +
> +    The above copyright notice and this permission notice (including the next
> +    paragraph) shall be included in all copies or substantial portions of the
> +    Software.
> +
> +    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> +    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> +    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
> +    THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> +    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> +    FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
> +    DEALINGS IN THE SOFTWARE.
> +  </copyright>
> +
> +  <interface name="zwp_drm_lease_manager_v1" version="1">
> +    <description summary="lease manager">
> +      This protocol is used by Wayland compositors which act as Direct
> +      Renderering Manager (DRM) masters to lease DRM resources to Wayland
> +      clients. Once leased, the compositor will not use the leased resources
> +      until the lease is revoked or the client closes the file descriptor.
> +
> +      The lease manager is used to advertise connectors which are available for
> +      leasing, and by the client to negotiate a lease request.
> +
> +      Warning! The protocol described in this file is experimental and
> +      backward incompatible changes may be made. Backward compatible changes
> +      may be added together with the corresponding interface version bump.
> +      Backward incompatible changes are done by bumping the version number in
> +      the protocol and interface names and resetting the interface version.
> +      Once the protocol is to be declared stable, the 'z' prefix and the
> +      version number in the protocol and interface names are removed and the
> +      interface version number is reset.
> +    </description>
> +
> +    <enum name="error">
> +      <entry name="stopped_manager" value="0"
> +        summary="request sent to a manager which has been stopped"/>
> +    </enum>
> +
> +    <request name="create_lease_request">
> +      <description summary="create a lease request object">
> +        Creates a lease request object.
> +
> +        See the documentation for zwp_drm_lease_request_v1 for details.
> +      </description>
> +      <arg name="id" type="new_id" interface="zwp_drm_lease_request_v1" />
> +    </request>
> +
> +    <request name="stop">
> +      <description summary="stop sending events">
> +        Indicates the client no longer wishes to receive connector events. The
> +        compositor may still send connector events until it sends the finish
> +        event, however.
> +
> +        The client must not send any requests after this one.
> +      </description>
> +    </request>
> +
> +    <event name="connector">
> +      <description summary="advertise connectors available for leases">
> +        The compositor may choose to advertise 0 or more connectors which may be
> +        leased to clients, and will use this event to do so. This object may be
> +        passed into a lease request to lease that connector. See
> +        zwp_drm_lease_request_v1.add_connector for details.
> +
> +        When this global is bound, the compositor will send all connectors
> +        available for lease, but may send additional connectors at any time.
> +      </description>
> +      <arg name="id" type="new_id" interface="zwp_drm_lease_connector_v1" />
> +    </event>
> +
> +    <event name="finished">
> +      <description summary="the compositor has finished using the manager">
> +        This event indicates that the compositor is done sending connector
> +        events. The compositor will destroy this object immediately after
> +        sending this event, and it will become invalid. The client should
> +        release any resources associated with this manager after receiving this
> +        event.
> +      </description>

I see that this event is used instead of a destructor request. This is
to ensure that no client can get the destruction sequence wrong, they
are forced to use "stop" request, right?

Very good, the tear-down looks fine to me.

> +    </event>
> +  </interface>
> +
> +  <interface name="zwp_drm_lease_connector_v1" version="1">
> +    <description summary="a leasable DRM connector">
> +      Represents a DRM connector which is available for lease. These objects are
> +      created via zwp_drm_lease_manager_v1.connector, and should be passed into
> +      lease requests via zwp_drm_lease_request_v1.add_connector.
> +    </description>
> +
> +    <event name="name">
> +      <description summary="name">
> +        The compositor sends this event once the connector is created to
> +        indicate the name of this connector. This will not change for the
> +        duration of the Wayland session, but is not guaranteed to be consistent
> +        between sessions.
> +
> +        If the compositor also supports zxdg_output_manager_v1 and this
> +        connector corresponds to a zxdg_output_v1, this name will match the
> +        name of this zxdg_output_v1 object.
> +      </description>
> +      <arg name="name" type="string" summary="connector name" />

Ok.

FYI, there have been talks about creating persistent names for
DRM connectors, scoped inside the specific DRM device instance which
could again be named persistently by devpath or such. It's possible
that at some point we might be able to use more reliable names.

> +    </event>
> +
> +    <event name="description">
> +      <description summary="description">
> +        The compositor sends this event once the connector is created to provide
> +        a human-readable description for this connector, which may be presented
> +        to the user.
> +      </description>
> +      <arg name="description" type="string" summary="connector description" />
> +    </event>
> +
> +    <event name="connector_id">
> +      <description summary="connector_id">
> +        The compositor will send this event to indicate the DRM ID which
> +        represents the underlying connector which is being offered. Note that
> +        the final lease may include additional object IDs, such as CRTCs and
> +        planes.
> +      </description>
> +      <arg name="connector_id" type="int" summary="DRM Connector ID" />

Right. This ID is scoped to the DRM device (instance), and cannot be used
to reliably determine which DRM device it originated from.

Is the ID useful for anything? The lease requesting interface uses
Wayland objects, so that does not need it.

> +    </event>
> +
> +    <event name="edid">
> +      <description summary="edid">
> +        The compositor may send this event once the connector is created to
> +        provide a file descriptor which may be memory-mapped to read the
> +        connector's EDID, to assist in selecting the correct connectors
> +        for lease. The fd must be mapped with MAP_PRIVATE by the recipient.

Should this be more clear about mmapping read-only?

> +
> +        Note that not all displays have an EDID, and this event will not be
> +        sent in such cases.
> +      </description>
> +      <arg name="edid" type="fd" summary="EDID file descriptor" />
> +      <arg name="size" type="uint" summary="EDID size, in bytes"/>

Yes, EDID size is a risk.

But even with this, I wonder if the initial data burst from binding to
the global interface could too easily grow too big... then again, I
don't have any better ideas either.

> +    </event>
> +
> +    <event name="withdrawn">
> +      <description summary="lease offer withdrawn">
> +        Sent to indicate that the compositor will no longer honor requests for
> +        DRM leases which include this connector. The client may still issue a
> +        lease request including this connector, but the compositor will send
> +        zwp_drm_lease_v1.finished without issuing a lease fd.
> +      </description>
> +    </event>
> +
> +    <request name="destroy" type="destructor">
> +      <description summary="destroy connector">
> +        The client may send this request to indicate that it will not issue a
> +        lease request for this connector. Clients are encouraged to send this
> +        after receiving the "withdrawn" request so that the server can release
> +        the resources associated with this connector offer.
> +      </description>
> +    </request>
> +  </interface>

The advertising interface looks ok, but I wonder if it is sufficient. It
lacks information about a connector's connectedness. If EDID is
provided, then the connector likely is connected, but the opposite is
not true.

This is a quote from #openhmd on Freenode from today:

< pq> ISTR there are VR/XR headsets that do not show up as connected on
the HDMI/DP/whatever display connector until they are turned on with
some USB commands, is that correct?

< pq> How useful is it to be able to see a video connector changing to
connected as you poke the USB commands, e.g. to see which display
device it is?

< pq> I'm asking because I intent to take look at the proposed Wayland
DRM leasing protocol extension, and want to know about use cases a bit.

< pq> particularly how you pick the right connector/monitor

< pH5> pq: yes, I think that was Rift CV1 (it has a display power bit
in its power register), there might be others.

< pH5> although in this special case, the EDID and the USB device both
have the same serial number, so we could match based on that.

< pH5> the HTC Vive doesn't have the serial number set in EDID data, so
in case of two connected vives appearance times might be useful.

----

Therefore there could be some use for dynamically updating a
connector's connectedness status to a Wayland client. When the client
pokes USB devices, the change in a DRM connector status could indicate
which connector to lease.

Or, is there some underlying assumption that compositors would offer
only connected connectors and withdraw on disconnect?

> +
> +  <interface name="zwp_drm_lease_request_v1" version="1">
> +    <description summary="DRM lease request">
> +      A client that wishes to lease DRM resources will attach the list of
> +      connectors advertised with zwp_drm_lease_manager_v1.connector that they
> +      wish to lease, then use zwp_drm_lease_request_v1.submit to submit the
> +      request.
> +    </description>
> +
> +    <enum name="error">
> +      <entry name="submitted_lease" value="0"
> +        summary="attempted to reuse a submitted lease"/>
> +    </enum>
> +
> +    <request name="destroy" type="destructor">
> +      <description summary="destroys the lease request object">
> +        Indicates that the client will no longer use this lease request.
> +      </description>
> +    </request>
> +
> +    <request name="request_connector">
> +       <description summary="request a connector for this lease">
> +         Indicates that the client would like to lease the given connector.
> +         This is only used as a suggestion, the compositor may choose to
> +         include any resources in the lease it issues, or change the set of
> +         leased resources at any time.
> +       </description>
> +       <arg name="connector" type="object"
> +         interface="zwp_drm_lease_connector_v1" />
> +    </request>
> +
> +    <request name="submit">
> +       <description summary="submit the lease request">
> +         Submits the lease request and creates a new zwp_drm_lease_v1 object.
> +         After calling submit, issuing any other request than destroy is a
> +         protocol error.
> +       </description>
> +       <arg name="id" type="new_id" interface="zwp_drm_lease_v1" />
> +    </request>
> +  </interface>
> +
> +  <interface name="zwp_drm_lease_v1" version="1">
> +    <description summary="a DRM lease">
> +      A DRM lease object is used to transfer the DRM file descriptor to the
> +      client and manage the lifetime of the lease.
> +    </description>
> +
> +    <event name="lease_fd">
> +      <description summary="shares the DRM file descriptor">
> +        This event returns a file descriptor suitable for use with DRM-related
> +        ioctls. The client should use drmModeGetLease to enumerate the DRM
> +        objects which have been leased to them. If the compositor cannot or

drmModeGetLease? Wouldn't drmModeGetResources() list only the leased
resources?

I suspect drmModeGetLease() would be for the lessor to check what's
leased out or something.

> +        will not grant a lease for the requested connectors, it will not send
> +        this event, instead sending the finished event immediately.
> +
> +        It is a protocol error for the compositor to send this event more than
> +        once for a given lease.
> +      </description>
> +      <arg name="leased_fd" type="fd" summary="leased DRM file descriptor" />
> +    </event>
> +
> +    <event name="finished">
> +      <description summary="sent when the lease has been revoked">
> +        When the compositor revokes the lease, it will issue this event to
> +        notify clients of the change. If the client requires a new lease, they
> +        should destroy this object and submit a new lease request. The
> +        compositor will send no further events for this object after sending
> +        the finish event.
> +      </description>
> +    </event>
> +
> +    <request name="destroy" type="destructor">
> +      <description summary="destroys the lease object">
> +        The client should send this to indicate that it no longer wishes to use
> +        this lease. The compositor should use drmModeRevokeLease on the
> +        appropriate file descriptor, if necessary, then release this object.

What does "then release this object" mean?

> +      </description>
> +    </request>
> +  </interface>
> +</protocol>


Thanks,
pq
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/wayland-devel/attachments/20190902/07d0a036/attachment.sig>