cosmic-comp crashes with "Failed to destroy old mode property blob" on display hotplug (KVM switch) — reproducible on multiple machines

[directlx]

Affected Systems:

Machine 1 — MSI Katana GF76 11UE

• OS: Pop!_OS 24.04 LTS
• cosmic-comp: 0.1177938082624.04~6ebe2a1
• GPU: Intel TigerLake-H UHD (iGPU) + NVIDIA RTX 3060 (hybrid graphics)
• Display connection: AGON AG493UG7R4 49" ultrawide (3840×1080) via HDMI → HDMI-A-1 on Intel GPU

Machine 2 — System76 Oryx Pro

• OS: Pop!_OS 24.04 LTS
• GPU: NVIDIA (hybrid graphics mode required for USB-C DP Alt Mode)
• Display connection: AGON AG493UG7R4 49" ultrawide (3840×1080) via DisplayPort (USB-C DP Alt Mode)

Shared setup:

• Both machines connected to the same KVM: Hearvo S7240PH2 (HDMI 2.1 + DP 1.4, EDID Adaptive)
• The AGON is connected to KVM via both HDMI (from Machine 1) and DP (from Machine 2)

---

Description:

When switching a KVM between two machines, the active display connection experiences a hotplug event (disconnect → reconnect). On both machines running COSMIC DE, this causes cosmic-comp to crash with:

Failed to destroy old mode property blob

The compositor hangs or terminates, taking all Wayland client windows with it. The external monitor shows a black screen. The COSMIC
session manager restarts cosmic-comp, but the external display is not recovered automatically.

This is reproducible on both HDMI and DisplayPort connections, and on both Intel iGPU and NVIDIA GPU paths.

Steps to reproduce:

1. Connect an external display via a KVM switch (HDMI or DisplayPort)
2. Switch the KVM to another PC (display signal drops)
3. Switch the KVM back (display signal restores / HPD event fires)
4. cosmic-comp crashes with "Failed to destroy old mode property blob"
5. External display shows black screen until cosmic-comp is manually restarted via pkill -x cosmic-comp

Additional notes:

• The bug also triggers when running cosmic-randr disable or cosmic-randr enable — any command that causes a mode property change
• i915 enable_dc=0 reduces freeze frequency on the Intel GPU machine but does not prevent the crash
• KVM has EDID Adaptive enabled, but HPD still fires on switch

Workaround (with significant side effect):

Running pkill -x cosmic-comp restarts the compositor and recovers the external display. However, because cosmic-comp is the Wayland
compositor, killing it terminates all open Wayland client windows — including every terminal, browser, and application. This makes the
workaround highly disruptive in a working environment, particularly when background tasks or terminal sessions are running.

A terminal multiplexer (tmux) can be used to preserve terminal sessions through the restart, but this should not be necessary for
something as routine as switching a KVM.

Expected behaviour:
cosmic-comp survives a display hotplug event and redetects the output without crashing.

Actual behaviour:
cosmic-comp crashes with "Failed to destroy old mode property blob" on every hotplug event caused by KVM switching, across both HDMI and
DisplayPort, on multiple machines.

[directlx]

Additional finding: DP unaffected — isolates bug to HDMI hotplug

After further testing with the same setup, I found that a second machine connected to the same KVM via DisplayPort (USB-C DP Alt Mode) is completely unaffected — KVM switching works without any issues on that machine.

Setup:

• Machine 1 (MSI Katana GF76, Intel i915, HDMI) → crashes every KVM switch
• Machine 2 (System76 Oryx Pro, NVIDIA, DP via USB-C) → works perfectly

Same KVM (Hearvo S7240PH2, EDID Adaptive), same monitor (AGON AG493UG7R4).

Two possible explanations:

1. The KVM's EDID Adaptive maintains HPD on the DP port but drops it on HDMI — so the Oryx Pro never sees a hotplug event while the MSI does
2. The mode blob bug specifically affects the i915 driver's HDMI hotplug path — the NVIDIA driver handles DP hotplug cleanly

This strongly suggests the bug is in HDMI hotplug handling specifically, not a general display hotplug issue in cosmic-com