Skip to content

Improvements - #30

Draft
SaicharanKandukuri wants to merge 10 commits into
mainfrom
improvements/zink-requirements-check
Draft

SaicharanKandukuri wants to merge 10 commits into
mainfrom
improvements/zink-requirements-check

Conversation

@SaicharanKandukuri

@SaicharanKandukuri SaicharanKandukuri commented Oct 3, 2026 •

Copy link
Copy Markdown
Member
  • Add Zink requirements checks with readable results in Device compatibility and full probe output in Diagnostic logs.

  • Fix Venus native WSI format queries and bundle virglrenderer host 1.3.0-7.

  • Use native Venus presentation by default while keeping MESA_VK_WSI_DEBUG available as an environment override.

( This pr has too many things now ).

Bundle virglrenderer-android 1.3.0-7 with Android WSI query translation.

Use native Venus presentation by default while preserving per-distro WSI overrides.
Accept FD-backed pixmaps for GPU Present copies, add DRI3 DMA-BUF export and lifecycle diagnostics, and package the Mesa loader compatibility fixes. Include Perfetto profiling instructions and preserve the experimental source-profile boundary. GNOME interaction remains under investigation.
Repeated desktop and terminal diagnostics required slow, timing-sensitive UI interaction. Add a dev-only receiver protected by Android's DUMP permission that queues a fixed set of supervisor actions instead of exposing arbitrary guest command execution.

Provide a CLI for session state, bounded journal reads, and supervised start/stop actions. Add focused X11, input, damage, Present, and process-lifecycle probes so compositor faults can be correlated across the app, Lorie, and guest without changing release behavior.
Redirected Plasma windows and panels remain AHardwareBuffer-backed inside Lorie. BufferFromPixmap previously rejected those pixmaps, forcing Mesa's texture-from-pixmap path through XGetImage and producing damaged frames and long panel stalls.

Export a CPU-readable AHardwareBuffer allocation through the public NDK socket handle transport, validate the selected descriptor by size and DMA-BUF synchronization, and close every received descriptor on all paths. Preserve the existing raw-fd export and make the patch reversible between standard and experimental source profiles.
Apply the AHardwareBuffer and DRI3 transport contract to every embedded X11 build so standard and SPHAL/Zink sessions share the same compositor path. Keep raw import and lifecycle logging experimental, retain the GPU-copy disable patch only as an A/B rollback, and document the full patch inventory and qualification gates.
Add an opt-in About toggle, device and guest compatibility checks, vendor-tree bindings, and a shared terminal/desktop launch profile for a manually installed proot-sphal runtime. Keep the profile hidden and disabled by default until standalone packaging and broader device qualification are complete.
@SaicharanKandukuri
SaicharanKandukuri marked this pull request as draft October 19, 2026 19:07
Backport the Present ordering and buffer-lockability work from termux/termux-x11#1148, and add a repeatable Panfrost damage reproduction command for device investigation.

Co-authored-by: Twaik Yont <9674930+twaik@users.noreply.github.com>
@SaicharanKandukuri
SaicharanKandukuri force-pushed the improvements/zink-requirements-check branch from 934375c to a0bc3df Compare October 19, 2026 19:10
Synchronize nested EXA CPU access, preserve logical AHardwareBuffer dimensions, and gate GPU Present when CPU and GLES views are incoherent.\n\nKeep raw-FD probes opt-in, isolate the unsafe direct-flip experiment, remove redundant logging and GPU-disable patches, and make the production stack replay cleanly from the pinned Termux:X11 source.\n\nReferences: termux/termux-x11#1148
Give each patch a kernel-style subject and concise rationale, retain relevant upstream references, and document production, diagnostics, experiment, ordering, and validation boundaries.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant