DRM refcounting, dma-buf heaps, and Blender UI debates

Kernel graphics lists carried several infrastructure patch series on dma_resv, system heaps, panels, and peer-to-peer DMA, while Blender threads examined sky rendering and Outliner defaults. Testing and virtio-gpu compute support also advanced.

dma_resv refcounting series reaches v3

Christian König posted version 3 of a series that replaces embedded dma_resv objects with dynamic allocation across DRM, TTM, i915, and many drivers. The change targets more consistent reference counting in the graphics memory path. Maintainers of DRM drivers should track it because the conversion spans core helpers and multiple hardware backends.

System heap 2MB chunks proposed

A three-patch series adds a 2MB order to the dma-buf system_heap so IOMMU superpages can be used. Christian König objected on design grounds and overhead during review. The outcome affects allocation efficiency for devices that benefit from large contiguous pages.

drm_panel embeds drm_bridge in v3

A 19-patch series at version 3 embeds a drm_bridge into every drm_panel, unifies the API, and deprecates panel_bridge helpers across many drivers. Discussion stayed focused on the structural cleanup. Display driver authors face a simpler but mandatory bridge model once the series lands.

PCI P2PDMA TLP-class routing at v6

Version 6 of an 18-patch series refines peer-to-peer DMA routing by TLP class and adds a dma-buf provider callback. Reviewers debated layering and whether dma-buf should issue PCI calls. The work matters for device-to-device transfers that avoid host memory copies.

Blender WIP ports Hillaire 2020 sky model

A community work-in-progress ports the Hillaire 2020 sky system plus celestials into Blender 5.3, supplies benchmarks, and asks for feedback. Thirteen participants exchanged dozens of technical messages. Rendering developers watching atmospheric quality and performance will find the thread relevant.

IGT cgroup dmem tests reach v6

A nine-patch IGT series adds an igt_cgroup dmem library and tests that exercise cgroup v2 device-memory limits and eviction on Xe and amdgpu. Three participants reviewed the coverage. The tests strengthen validation of GPU memory control under resource pressure.

Outliner focus and expand defaults debated

Blender developers and users discussed whether new Outliner Scroll-to-Active and Expand-on-Focus behavior should remain the default after complaints of distraction in complex scenes. Seventeen participants weighed in. The decision will set navigation defaults for large projects.

virtio-gpu userptr support at v8

Version 8 of a four-patch series adds userptr blobs and a ROCm capset to virtio-gpu for zero-copy guest compute workloads. Two participants handled review. Virtualized GPU compute users gain a path that avoids extra copies between guest and host.