Skip to contentSkip to conversation

State of stable warrior in Core/bitbake/YP layers

DISCUSSION akuster808 3 participants 4 messages Oct 27, 2019

lore
Share & tools
lore

akuster808

author
#0 lore raw

Hello,

Of the past month the warrior stable branch has been plagued with a build failure on FedoraCore30 regarding qemu virgl tests. This has caused the slip of the next dot release to occur after Thud's.

https://bugzilla.yoctoproject.org/show_bug.cgi?id=13543

The error message was this:

runqemu - ERROR - Failed to run qemu: libEGL warning: MESA-LOADER:
failed to open swrast (search paths /usr/lib64/dri) libEGL warning: MESA-LOADER: failed to open swrast (search paths /usr/lib64/dri)

qemu-system-i386: egl: eglInitialize failed
qemu-system-i386: OpenGL is not supported by the display

What has fixed this error is updating libdrm to 2.4.99 from 2.4.97. What I can not tell is what commit(s) in those updates fixed the above error

https://lists.freedesktop.org/archives/dri-devel/2019-July/225069.html
https://lists.x.org/archives/xorg-announce/2019-April/002992.html

Has anyone seen the above error and might have an idea what commits would solve this issue?

Anyone familiar with the libdrm update policy? is the .99 a bugfix only, it does not look like it from the commit messages.

The current Stable policy indicates newer hosts are supposed to be supported and if not updating to the .99 release is an option, I foresee it taking additional effort to figure which commits will fix the error.

Any clues or feedback is appreciated.

Regards,
Armin

#1 lore raw
Quoted 16 lines · On Sun, 27 Oct 2019 at 09:13, akuster808 <[email protected]> wrote:

Of the past month the warrior stable branch has been plagued with a build failure on FedoraCore30 regarding qemu virgl tests. This has caused the slip of the next dot release to occur after Thud's.

https://bugzilla.yoctoproject.org/show_bug.cgi?id=13543

The error message was this:

runqemu - ERROR - Failed to run qemu: libEGL warning: MESA-LOADER:
failed to open swrast (search paths /usr/lib64/dri) libEGL warning: MESA-LOADER: failed to open swrast (search paths /usr/lib64/dri)

qemu-system-i386: egl: eglInitialize failed
qemu-system-i386: OpenGL is not supported by the display

Quoted 9 lines

What has fixed this error is updating libdrm to 2.4.99 from 2.4.97. What I can not tell is what commit(s) in those updates fixed the above error

https://lists.freedesktop.org/archives/dri-devel/2019-July/225069.html
https://lists.x.org/archives/xorg-announce/2019-April/002992.html

Has anyone seen the above error and might have an idea what commits would solve this issue?

The DRI modules loaded from the host (swrast particularly) may try to resolve a symbol in libdrm-native that doesn't exist.

I remember debugging a similar problem when TLS got accidentally switched off in mesa-native, what helped was tracing symbol
resolution with the help of environment variables described in 'man ld.so' (can't remember off the top of my head what was the exact one).

Alex

author
#2 lore raw
Quoted 27 lines · Alexander Kanavin

On Sun, 27 Oct 2019 at 09:13, akuster808 <[email protected] <mailto:[email protected]>> wrote:

Of the past month the warrior stable branch has been plagued with a build failure on FedoraCore30 regarding qemu virgl tests. This has caused the slip of the next dot release to occur after Thud's.

https://bugzilla.yoctoproject.org/show_bug.cgi?id=13543

The error message was this:

runqemu - ERROR - Failed to run qemu: libEGL warning: MESA-LOADER:
failed to open swrast (search paths /usr/lib64/dri) libEGL warning: MESA-LOADER: failed to open swrast (search paths /usr/lib64/dri)

qemu-system-i386: egl: eglInitialize failed
qemu-system-i386: OpenGL is not supported by the display

What has fixed this error is updating libdrm to 2.4.99 from 2.4.97. What
I can not tell is what commit(s) in those updates fixed the above error

https://lists.freedesktop.org/archives/dri-devel/2019-July/225069.html
https://lists.x.org/archives/xorg-announce/2019-April/002992.html

Quoted 6 lines

Has anyone seen the above error and might have an idea what commits would solve this issue?

The DRI modules loaded from the host (swrast particularly) may try to resolve a symbol in libdrm-native that doesn't exist.

What I justed noticed on the host is that his has both 5.2 and 5.3 kernels. FC 30 was tested in the 2.7.1 so it worked at one point in time. I wonder if the kernel got updated to 5.3 and now we need updated headers for libdrm? That might support the missing symbols you mentioned above.

Thanks for the hint.

- Armin

Quoted 7 lines

I remember debugging a similar problem when TLS got accidentally switched off in mesa-native, what helped was tracing symbol resolution with the help of environment variables described in 'man ld.so' (can't remember off the top of my head what was the exact one).

Alex

#3 lore raw
Quoted 38 lines · On Sun, Oct 27, 2019 at 4:06 PM akuster808 <[email protected]> wrote:

On 10/27/19 4:03 AM, Alexander Kanavin wrote:

Quoted 17 lines · On Sun, 27 Oct 2019 at 09:13, akuster808 <[email protected]> wrote:

Of the past month the warrior stable branch has been plagued with a build failure on FedoraCore30 regarding qemu virgl tests. This has caused the slip of the next dot release to occur after Thud's.

https://bugzilla.yoctoproject.org/show_bug.cgi?id=13543

The error message was this:

runqemu - ERROR - Failed to run qemu: libEGL warning: MESA-LOADER:
failed to open swrast (search paths /usr/lib64/dri) libEGL warning: MESA-LOADER: failed to open swrast (search paths /usr/lib64/dri)

qemu-system-i386: egl: eglInitialize failed
qemu-system-i386: OpenGL is not supported by the display

Quoted 8 lines

What has fixed this error is updating libdrm to 2.4.99 from 2.4.97. What I can not tell is what commit(s) in those updates fixed the above error

https://lists.freedesktop.org/archives/dri-devel/2019-July/225069.html
https://lists.x.org/archives/xorg-announce/2019-April/002992.html

Has anyone seen the above error and might have an idea what commits would solve this issue?

The DRI modules loaded from the host (swrast particularly) may try to resolve a symbol in libdrm-native that doesn't exist.

What I justed noticed on the host is that his has both 5.2 and 5.3 kernels. FC 30 was tested in the 2.7.1 so it worked at one point in time. I wonder if the kernel got updated to 5.3 and now we need updated headers for libdrm? That might support the missing symbols you mentioned above.

Quoted 8 lines

Thanks for the hint.

- Armin

I remember debugging a similar problem when TLS got accidentally switched off in mesa-native, what helped was tracing symbol resolution with the help of environment variables described in 'man ld.so' (can't remember off the top of my head what was the exact one).

https://lists.x.org/archives/xorg-announce/2019-April/002992.html

does have header fixes which might be interesting to this issue

Quoted 8 lines

Alex

--
_______________________________________________
Openembedded-core mailing list
[email protected]
http://lists.openembedded.org/mailman/listinfo/openembedded-core