On Wednesday, June 9th, 2021 at 17:03, Daniel Vetter <daniel@xxxxxxxx> wrote: > On Wed, Jun 09, 2021 at 10:39:03AM +0000, Simon Ser wrote: > > When a uevent only updates a single connector, add a CONNECTOR property > > to the uevent. This allows user-space to ignore other connectors when > > handling the uevent. This is purely an optimization, drivers can still > > send a uevent without the CONNECTOR property. > > > > The CONNECTOR property is already set when sending HDCP property update > > uevents, see drm_sysfs_connector_status_event. > > > > This has been tested with a wlroots patch [1]. > > > > amdgpu has been updated to use these new fine-grained uevents. > > That's a bit thin for motivating roll-out ... My intention was to add a few other drivers as well if the idea is good. amdgpu just happens to be a codebase I know and can work with hardware I own. > Also my idea was that with the per-connector epoch counter we could make > this happen for anyone using the probe helpers. Yeah, I remember something about this, but then nothing materialized. > I also thought that the dp hotplug code was a bit more unified by now to > not need the duplication across all drivers. But there's only three for > that. > > Ofc drivers that hand-roll their hpd code need to be adjusted, but I've > hoped that with the epoch counter we could just make the current hotplug > event code dtrt thing here automatically maybe? Sorry, not signing up for that. It looked easy enough to adjust existing calls, but anything more complicated and I'm out of time. If anything, this patch is mostly a signal that user-space would like to receive CONNECTOR properties. User-space is typed up and ready if any kernel dev wants to implement this. I'm completely fine with abandonning this patch.
