Multi-seat support

[Arisa-Snowbell]

Currently, seat-based hardware access is not strictly enforced; utilities such as loginctl merely offer hints, and support depends on implementation at the software level. Introducing robust multi-seat support in the compositor would be extremely valuable. It would enable multiple developers to effectively share a single machine—a practical solution given the rising cost of hardware and stagnant salary growth. Very beneficial for indie game developers. The implementation shouldn't be as hard.

[Silverdev2482]

This was previously discussed in #1731, some guy had some work done on it, though I am not sure if it would still be useful.

[Arisa-Snowbell]

5 likes @vaxerski

[vaxerski]

I realize.

[vaxerski]

I need some specifics. What part of multi-seat support do you need that's missing? Sure, we expose one seat (by design) - and I am generally against doing otherwise.

[Silverdev2482]

I think something like how we can configure input devices to be bound to a specific seat.

[Lassebq]

For a long time I thought the only form of multiseat on linux was logind's seat system. Until I learned that sway has something implemented to support it in a single compositor instance and as it turns out even the wayland spec mentions some kind of seats, separate from logind ones. There are still plenty of subsystems that would not support this I imagine, like running multiple user sessions on a single wayland socket. It's just broken by design imo. Unless it's intended to run as a single user.
Currently logind multiseat is the best option and has basically no compatibility issues on software side (assuming a proper DM implementation, like patched GDM), even if it works at the cost of one GPU per seat. (I have a long time experience using a multi-seat gaming machine, and ngl sharing storage and GPU resources is one of the best advantages)

Btw, you mentioned the rising cost of GPUs. You don't actually need a powerful GPU for each seat. You can run applications on secondary seats using "primary GPU" similarly how you would on laptop PCs. Either with DRI_PRIME env if it's a hardware supported by Mesa, or prime-run on Nvidia. I'm not actually certain though for Nvidia. The machines I tested it on had either intel iGPU & AMD dGPU or AMD iGPU & AMD dGPU. I also tried this on super cheap USB to VGA adapter, which despite what it does, can be used by logind as a separate seat. Although it has some problem with aforementioned way of using "primary GPU" which causes it to use software rendering.

In any case. In my opinion the real solution to wayland multiseat requires changes in logind multiseat or DRM subsystem. Not certain what or how exactly it needs to change, but from user's end who configures their machine for multiseat loginctl attach should also accept individual monitor connectors which should be respected by a wayland compositor when enumerating for displays which it takes ownership of.

I hope this info is useful for anyone who comes across this thread. Perhaps somebody even makes a PoC of my suggested solution.

[CatalinPuscoci]

I want to join this discussion to more clearly define a use case.

I would like to be able to use my computer normally, while there can be another hyprland instance, running off of the same GPU preferably, but with a headless output, a virtual(or not necessarily) cursor and keyboard (I suppose this is where they get bound to a different seat). Ideally, I would be able to do anything I can do in the first session if I am able to remote into it, or assign it a different set of peripherals. Most times I believe one would RDP into it, VNC into it, or sunshine/moonlight into it.

• Is the entire thing end-to-end possible?
• Is something in another component of the whole stack blocking Hyprland from achieving this?
• Is anything on the side of Hyprland preventing it from supporting such a use case?
• Assuming it can be implemented, what can we expect from a performance standpoint? For example, can two people play an FPS-capped game at the same time assuming reasonable hardware?

@Lassebq I imagine you can provide better information about potential blocks outside of Hyprland given your experience.

If the only blocking points are "nobody has bothered to do this", I would be willing to pay at least in part for the development time of such a feature.

[LeadProphet]

just chiming in as a multiseat user on dietpi (just a customized debian, really) - yes the performance is fine. My two kids run War Thunder at high frame rates from the same pc, which has 16gb RAM, a 3800XT CPU, and two RX 580 8gb GPUs.