On Sat, 2022-02-05 at 12:20 +0000, Richard Purdie via lists.openembedded.org
wrote:
> On Fri, 2022-02-04 at 21:40 +0100, Alexander Kanavin wrote:
> > Some of the host distributions build the drivers in a way (RPATH/RUNPATH)
> > that tricks uninative loader into loading pieces of the host libc, if
> > the same pieces haven't been previously loaded by native binaries. Mixing
> > the two libc versions leads to failures.
> >
> > This change ensures that the correct (uninative) versions are always in use.
> >
> > Signed-off-by: Alexander Kanavin <a...@linutronix.de>
> > ---
> > meta/classes/qemuboot.bbclass | 2 +-
> > scripts/runqemu | 4 ++++
> > 2 files changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/meta/classes/qemuboot.bbclass b/meta/classes/qemuboot.bbclass
> > index cc1cbce69d..229bd88527 100644
> > --- a/meta/classes/qemuboot.bbclass
> > +++ b/meta/classes/qemuboot.bbclass
> > @@ -109,7 +109,7 @@ def qemuboot_vars(d):
> > build_vars = ['MACHINE', 'TUNE_ARCH', 'DEPLOY_DIR_IMAGE',
> > 'KERNEL_IMAGETYPE', 'IMAGE_NAME', 'IMAGE_LINK_NAME',
> > 'STAGING_DIR_NATIVE', 'STAGING_BINDIR_NATIVE',
> > - 'STAGING_DIR_HOST', 'SERIAL_CONSOLES']
> > + 'STAGING_DIR_HOST', 'SERIAL_CONSOLES', 'UNINATIVE_LOADER']
> > return build_vars + [k for k in d.keys() if k.startswith('QB_')]
> >
> > do_write_qemuboot_conf[vardeps] += "${@' '.join(qemuboot_vars(d))}"
> > diff --git a/scripts/runqemu b/scripts/runqemu
> > index 4e05c1bb15..7d1f85f357 100755
> > --- a/scripts/runqemu
> > +++ b/scripts/runqemu
> > @@ -460,6 +460,10 @@ class BaseConfig(object):
> > except subprocess.CalledProcessError as e:
> > raise RunQemuError("Could not determine the path to dri
> > drivers on the host via pkg-config.\nPlease install Mesa development files
> > (particularly, dri.pc) on the host machine.")
> > os.environ['LIBGL_DRIVERS_PATH'] = dripath.decode('utf-8').strip()
> > + uninative_path = os.path.dirname(self.get("UNINATIVE_LOADER"))
> > + preload_items = ['libdl.so.2', 'librt.so.1', 'libpthread.so.0']
> > + preload_paths = [os.path.join(uninative_path, i) for i in
> > preload_items]
> > + os.environ['LD_PRELOAD'] = " ".join(preload_paths)
> >
>
> This looks like a good way to work around this corner case and I don't see any
> other alternative. Can I ask that:
>
> a) we put some comment into the code so we can understand this in a couple of
> years time as it is a rather weird corner case
>
> b) we put some "if os.path.exists()" in there, maybe even just on the
> uninnative
> loader direction so that in the non-uninative case we don't throw warnings?
I've added a tweak doing this into master-next.
Cheers,
Richard
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#161394):
https://lists.openembedded.org/g/openembedded-core/message/161394
Mute This Topic: https://lists.openembedded.org/mt/88916306/21656
Group Owner: openembedded-core+ow...@lists.openembedded.org
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub
[arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
