The Linux Media Summit 2025, held in Nice alongside Embedded Recipes and co-sponsored by Cisco and Collabora, gathered around 20 media subsystem developers. Key discussions covered a hybrid stateless/stateful video encoder approach using compound controls and eBPF-extensible rate control, the complexity of camera infrastructure with 50+ driver nodes, memory accounting fixes for DRM/V4L2 to prevent unbounded allocations and cgroup bypasses, the fate of long-stagnant staging drivers, PipeWire hardware decoding for sandboxed apps, and progress on a multi-committer model using merge requests for better transparency and CI integration.
Table of contents
Stateless Video EncodersCamera SupportStaging DriversMemory AccountingMulti-committer modelLooking aheadQuestions this post answers
What is the proposed approach for stateless video encoding in the Linux kernel?
The proposed approach uses a hybrid model: stateless encoding via compound controls and requests, paired with a stateful rate controller implemented in the kernel. A shared C implementation would be used across drivers, with the option to extend it using eBPF in the future, while always providing a fixed per-frame control to applications. Linux kernel media developers tracking encoder work follow these discussions on daily.dev.
Why can memory allocations in V4L2 and DRM drivers bypass cgroup limits and how is it being fixed?
Memory allocations in V4L2 and DRM drivers are not properly tracked, meaning cgroup constraints can be bypassed and there is effectively no real limit on allocation. Maxime Ripard developed an implementation that centralizes memory accounting through the DMA API, mirroring the approach used in DRM, to improve driver robustness and sandboxing safety. Developers working on V4L2 sandboxing and driver security can track these kernel changes on daily.dev.