Table 25.2.2 It is now availableFor the average user, this, which is a minor update, and the rest, which can only be medium or major, seem to be of no great importance, but everything changes if you intend to play on Linux.
If you work or play on Linux, *BSD, or Android with compatibility layers, you might be interested in: This release addresses real-world bugs found in games and synthetic tests. (DOOM: The Dark Ages, Dragon Age: Veilguard, 3DMark Solar Bay), fixes regressions, and improves critical video paths and synchronization. The release was managed by Eric Engestrom and published on September 3, 2025, as a maintenance update.
What Table 25.2.2 really brings
It is a correction version (bug fix release) on 25.2.1, with no notable "new features" block in this iteration. Still, the patch set has practical impact: video decoding/encoding quality, compiler robustness, cache coherency, barriers and synchronization, and several GPU-specific fixes.
On the OpenGL side, Mesa implements the 4.6 API, but The version returned by the context depends on the driver and the context typeTo have 4.6, you must explicitly request it when creating the context; in compatibility contexts, some drivers will report a lower version if a requirement is missing.
In Vulkan, Mesa declares support for the 1.4 API; however, The value reported in VkPhysicalDeviceProperties.apiVersion will vary depending on the specific driver. (radv, nvk, anv, panvk, lavapipe), as each enables capabilities at different rates and for different chips.
Highlighted fixes visible to the user in Mesa 25.2.2
During these two weeks, arrangements have been integrated that, on a day-to-day basis, are noticeable in games, benchmarks and video playback:
- GL_EXT_shader_clock fixed in the OpenGL path, avoiding inconsistencies with the shader clock in applications that query it for timing.
- Video on Radeon (VCN and Vulkan Video): H.264/H.265 adjustments, compatibility with older firmware and VP9. Disables the 8x8 transform when CABAC is not active, fixes loop filter and segmentation parameters in VP9, and fixes HEVC details such as cu_qp_delta and transform_skip with older firmware.
- NVK (NVIDIA on Vulkan): support for larger QMDs and other improvements for recent hardware, fine-tuning compatibility and stability on new generation GPUs.
- Intel ANV and i965/brw: Various fixes to L3 cache, barriers, utrace, and broadcast opcode, as well as revert problematic changes and temporarily disable extensions to prevent regressions.
- Zink (OpenGL on Vulkan): 14 patches with improvements in synchronization, clears and layouts, including sparse resource jam fixes and multiple stencil state adjustments.
- DRIRC: Added workaround to deal with broken behaviors with ANGLE In some scenarios, reducing problems in apps that use that GL->D3D/Vulkan translator.
List of fixed crawler bugs
In addition to the low-level changes, Mesa 25.2.2 closes issues reported by users and CIs. These are the tracker issues that are marked as fixed in this release.:
- radv: Artifacts when viewing VP9 video with hardware decoding on Vulkan.
- CI zink-venus-lavapipe: Errors in the integration and testing of the Zink/Venus/Lavapipe stack.
- ANV in DG2/BMG: Incorrect rendering in 3DMark Solar Bay.
- lavapipe: Non-optimal export selection when multiple types were requested; dma-buf is prioritized when feasible.
- year: : Hang on Intel GPU with simple compute shader due to excessive loop unrolling.
- INTEL_DEBUG=spill_fs: Fixed regression in debug path and spill in fragment shader.
- NIR: Failed validation after nir_lower_io with DOOM: The Dark Ages; fixes also in ACO for undefined operands in p_parallelcopy.
- Dragon Age: Veilguard: : Fixed distortions in the skill wheel (targeting).
- AMD OpenGL: glTexturePageCommitmentEXT no longer fails when working with mip levels greater than 0.
- Zinc: : Lack of synchronization for scattered resources fixed.
- RX 9070 (XT): System crashes on VR startup mitigated with critical path fixes.
Table 25.2.2 Technical and Compatibility Notes
In OpenGL, for applications that rely on fine timing or synchronization, the GL_EXT_shader_clock fix Reduces discrepancies in internal measurements. In Vulkan, changes to ANV and RADV ensure cache and barrier coherence, especially when moving data between queues or pre-rasterization stages.
Users who work with hardware decoding/encoding On Radeon users will notice less artifacts in VP9 and more predictable behavior in H.264/H.265 when certain compression tools (such as CABAC) are not active or when using older firmware.
Synchronization and memory: dma-buf, sparse, and WSI
The management of external memory and dispersed resources receives special attention: lavapipe prioritizes dma-buf whenever possible, which improves interoperability with other multimedia stacks; Zink fixes waits and barriers on resource deallocations and one-off blits to prevent data loss.
In Android and panvk the WSI layer is reinforced: memory linked from offset 0 and covered AHB cases, reducing surprises with swap buffers and device-dependent formats.
Next steps of the project
While delivering 25.2.2, The next feature release will be Mesa 25.3, expected around November if the schedule holds. Development is already underway on mainline, with the usual mix of new extensions, optimizations, and support for emerging hardware.
This 25.2.2 meets what is expected in a point release: stability, reliability and regression correction No nasty surprises. If you're coming from 25.2.1 or the initial 25.2, it's worth upgrading for the Radeon video improvements, Zink timing fixes, critical path cleanup in ANV/Intel, and NVK boost for newer GPUs—all while maintaining OpenGL 4.6 and Vulkan 1.4 support, with the usual nuances depending on the driver and context.
