Add MIR_PIN_COMPOSITING_TO to allow pinning a certain provider for display#4958
Open
tarek-y-ismail wants to merge 1 commit into
Open
Add MIR_PIN_COMPOSITING_TO to allow pinning a certain provider for display#4958tarek-y-ismail wants to merge 1 commit into
MIR_PIN_COMPOSITING_TO to allow pinning a certain provider for display#4958tarek-y-ismail wants to merge 1 commit into
Conversation
tarek-y-ismail
force-pushed
the
composite-on-one-gpu
branch
from
8dbdd9a to
88debcf
Compare
Contributor
There was a problem hiding this comment.
Pull request overview
Adds an opt-in workaround for multi-GPU systems (notably Intel+iGPU + NVIDIA dGPU) by allowing Mir to pin compositor/provider selection to a single rendering provider across all displays via MIR_PIN_DISPLAY_TO, avoiding problematic cross-GPU DMA-BUF paths.
Changes:
- Introduces
GLRenderingProvider::driver()as a provider identifier and implements it across in-tree providers and test doubles. - Updates
DefaultDisplayBufferCompositorFactoryto honorMIR_PIN_DISPLAY_TOwhen choosing a rendering provider. - Adds driver identification logic for GBM-KMS (DRM driver name) and EGL-based providers (EGL vendor string).
Reviewed changes
Copilot reviewed 10 out of 10 changed files in this pull request and generated 6 comments.
Show a summary per file
| File | Description |
|---|---|
include/platform/mir/graphics/platform.h |
Adds GLRenderingProvider::driver() API for provider identification (platform interface change). |
src/server/compositor/default_display_buffer_compositor_factory.cpp |
Implements MIR_PIN_DISPLAY_TO handling when selecting a rendering provider per display. |
src/platforms/renderer-generic-egl/buffer_allocator.h |
Declares driver() override for the generic EGL provider. |
src/platforms/renderer-generic-egl/buffer_allocator.cpp |
Implements driver() (currently via EGL_VENDOR). |
src/platforms/gbm-kms/server/buffer_allocator.h |
Declares driver() override for the GBM-KMS provider. |
src/platforms/gbm-kms/server/buffer_allocator.cpp |
Implements driver() using drmGetVersion() to return the DRM driver name. |
src/platforms/eglstream-kms/server/buffer_allocator.h |
Declares driver() override for the EGLStream-KMS provider. |
src/platforms/eglstream-kms/server/buffer_allocator.cpp |
Implements driver() (currently via EGL_VENDOR). |
tests/include/mir/test/doubles/stub_gl_rendering_provider.h |
Updates stub GL rendering provider to implement driver(). |
tests/include/mir/test/doubles/mock_gl_rendering_provider.h |
Updates mock GL rendering provider to mock driver(). |
AlanGriffiths
changed the title
MIR_PIN_DISPLAY_TO to allow pinning a certain provider for displayMIR_PIN_COMPOSITING_TO to allow pinning a certain provider for display
tarek-y-ismail
force-pushed
the
composite-on-one-gpu
branch
from
88debcf to
803a83e
Compare
Contributor
Author
|
Tried another approach that doesn't require ABI breaks. It also allows us to choose a specific GPU instead of relying on drm/EGL names that might be shared for multiple cards. |
…compositing Works around warping on systems with intel + nvidia GPUs. When buffers are transferred over DMA, when rendering on an intel GPU and displaying on an nvidia GPU for example, nvidia GPUs assume the buffers to have a stride/alignment of 32 bytes, ignoring the given stride. This causes the output windows to appear warped. This sidesteps the problem by forcing a single GL provider to composite all displays. For compositors not on the pinned GPU, Mir falls back to the CPU copy path: the pinned GPU renders into an FBO, glReadPixels copies the result into a CPU-addressable buffer, and the display GPU scans that out directly. This avoids any cross-GPU DMA-BUF import, so the Nvidia driver never has a chance to misread the stride. The value of MIR_PIN_COMPOSITING_TO should be the DRM device node (e.g. /dev/dri/renderD128) of the GPU to pin to. Available nodes can be discovered with `ls /dev/dri/renderD*`. For deviceless platforms (e.g. mir:virtual), the module name is used as the identifier.
tarek-y-ismail
force-pushed
the
composite-on-one-gpu
branch
from
803a83e to
dab7781
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Supersedes #4640
What's new?
Works around display warping on systems with Intel + Nvidia GPUs.
In a multi-GPU setup, Mir normally picks the best rendering provider per display — Intel for Intel-connected
displays, Nvidia for Nvidia-connected displays. When buffers cross GPU boundaries via DMA-BUF for compositing,
Nvidia assumes a 32-byte stride/alignment, ignoring the actual stride. This causes output windows to appear warped.
Setting
MIR_PIN_COMPOSITING_TO=<devnode>(e.g./dev/dri/renderD128) forces all displays to be composited by one GLprovider. For displays not physically on that GPU, Mir falls back to a CPU copy path (glReadPixelsinto aCPU-addressable framebuffer), bypassing the DMA-BUF import entirely.
The device node is used as the identifier because it is unique per physical GPU, handling systems with multiple GPUs of the same driver type (e.g. two AMD cards). Available nodes can be discovered with ls
/dev/dri/renderD*How to test
On a system with Intel + Nvidia GPUs and a display on each:
Verify that previously warped windows on the Nvidia-connected display are now rendered correctly.
You can find reproduction steps in #4640