[OE-core] [PATCH 00/12] Enable accelerated OpenGL in qemu

Alexander Kanavin alex.kanavin at gmail.com
Fri Feb 8 14:45:41 UTC 2019

V2 changes: addressed feedback from the first review round
V3 changes:
- better fix for missing qemu X11 include, as discussed with upstream
- maintainers.inc entry for virglrenderer
- egl-headless support (see below for details)
- improvements to kmscube recipe
- fix to vte-native to allow building it with gcc 4.8
V4 changes
- address failures uncovered by yocto autobuilder
V5 changes
- remove patches that have been merged to master
- replace BBCLASSEXTEND_remove with PREFERRED_PROVIDER for mesa
recipe variants
V6 changes
- again rebase to master
- tweak oe-selftest to remove unnecessary re-builds of qemu-native

Why this? Why now?

1. I think we are heading towards a reality where some kind of working
GL is a 'must' for any kind of UI work.
2. Typically people build images and then transfer them to the NUC or
even target boards for testing - using qemu would be less awkward.
3. Current qemu configuration is basically useless here, as it only
provides a very slow software GL driver from mesa.

0. TLDR:

a) Gtk UI frontend

$ . oe-init-build-env build-virgl
$ bitbake core-image-sato-sdk
$ runqemu kvm gl
$$ run glxgears or any other GL-capable binary inside qemu

b) egl-headless

$ . oe-init-build-env build-virgl
$ bitbake core-image-sato-sdk
$ runqemu kvm egl-headless publicvnc
$ <connect with a vnc viewer to>
$$ run glxgears or any other GL-capable binary inside qemu

1. For the local UI, qemu is switched over to use gtk frontend. I simply 
couldn't get SDL frontend to work properly, it only displays a blank black 
window or doesn't start at all. Same thing happens with qemu binaries provided 
by Fedora and opensuse (Ubuntu's qemu lacks virgl support). Seems like SDL support 
has regressed.

2. What is egl-headless?

In this variant, Qemu does not open a UI window at all. Instead, it renders
all graphics, including GL, into a memory buffer, which can be seen with
a vnc or spice client (over a TCP socket). This has the following advantages:

- no need to be physically present at the host machine, output (including GL bits!) 
can be seen remotely
- no need for the host machine to run an X session

3. While the components are built against the most minimum necessary 
mesa-native set (gbm, egl, dri with no drivers), libepoxy loads and uses the 
host GL implementation (through a chrpath hack). This is both to save build 
times, and because the host is likely to have a set of drivers more 
appropriate for the physical machine (e.g. proprietary nvidia stack, 
although I didn't test that).

4. I tested this with
- glamor X server
- weston compositor with drm backend (build core-image-weston, then edit 
  weston.ini in the image to use drm backed instead of fbdev).
- kmscube
- glxgears
- https://www.geeks3d.com/gputest/

The latter two show FPS that is similar to running them directly on the host.

5. Some things I am not sure about:

 - how much of a 'default' should this be for running qemu. It works for me, 
but other people might find it less stable compared to the existing 
sdl-vmware-swrast setup.

- what other tests and demos could be run

- how could this be tested on the autobuilder. I wrote two oe-selftests for 
this (gtk, egl-headless), but one of them does require an X session, and both 
require ability to create opengl contexts on the host.

The following changes since commit 29099c8a49bf9d35514bbc1e77657655598a80cc:

  maintainers.inc: replace Changhyeok Bae's @lge email address with a personal one (2019-02-08 10:57:19 +0000)

are available in the Git repository at:

  git://git.yoctoproject.org/poky-contrib akanavin/virgl-gtk

Alexander Kanavin (12):
  virglrenderer: add a recipe
  qemu: enable virglrenderer and glx options for native/nativesdk builds
  local.conf.sample: adjust the qemu config to enable gtk+ instead of
  qemu: build target variant with gtk+, and nativesdk variant without
  qemu: remove support for building against host sdl
  qemu: add a gettext-native dependency to gtk option
  qemu: add a patch to avoid a missing definition error
  qemu: add environment variable wrappers to make qemu look good with
    gtk frontend
  qemu: add a backported patch to fix egl-headless support
  runqemu: add options for enabling virgl GL acceleration
  runqemu: do not check for GL libraries
  selftest: add tests for virgl GL acceleration

 meta-poky/conf/local.conf.sample              |  9 +--
 meta-selftest/lib/oeqa/runtime/cases/virgl.py | 28 ++++++++
 meta/conf/distro/include/maintainers.inc      |  1 +
 meta/lib/oeqa/selftest/cases/runtime_test.py  | 52 +++++++++++++++
 meta/recipes-devtools/qemu/qemu.inc           | 41 +++++++-----
 .../qemu/0001-Add-a-missing-X11-include.patch | 65 +++++++++++++++++++
 ...-egl-headless-add-egl_create_context.patch | 50 ++++++++++++++
 ...x-libcap-header-issue-on-some-distro.patch |  2 +-
 ...-messages-when-qemi_cpu_kick_thread-.patch |  2 +-
 meta/recipes-devtools/qemu/qemu_3.1.0.bb      |  2 +
 .../0001-vtest-add-missing-includes.patch     | 38 +++++++++++
 .../virglrenderer/virglrenderer_git.bb        | 19 ++++++
 scripts/runqemu                               | 58 +++++++----------
 13 files changed, 309 insertions(+), 58 deletions(-)
 create mode 100644 meta-selftest/lib/oeqa/runtime/cases/virgl.py
 create mode 100644 meta/recipes-devtools/qemu/qemu/0001-Add-a-missing-X11-include.patch
 create mode 100644 meta/recipes-devtools/qemu/qemu/0001-egl-headless-add-egl_create_context.patch
 create mode 100644 meta/recipes-graphics/virglrenderer/virglrenderer/0001-vtest-add-missing-includes.patch
 create mode 100644 meta/recipes-graphics/virglrenderer/virglrenderer_git.bb


More information about the Openembedded-core mailing list