experiences with images based on current layer heads

experiences with images based on current layer heads

[Andreas Müller]

Hi,

when migration of systemd to oe-core turned to unbuildable I decided
to freeze my images and stopped further tests. Now that many issues
were fixed and we are close to another release I built my standard
xfce-based image based on latest layer HEADs. With this I face some
issues which I would like to share:

1. X sessions crash with OOM. This is reproducible by opening
google-maps with midori and changing to satellite view. In
Xorg.0.log.old I see only

| (EE)
| (EE) Backtrace:
| (EE)
| (EE) Segmentation fault at address 0x9
| (EE) BUG: triggered 'if (inSignalContext)'
| (EE) BUG: log.c:502 in LogVMessageVerb()
| (EE) Warning: attempting to log data in a signal unsafe manner while
in signal context.
| Please update to check inSignalContext and/or use
LogMessageVerbSigSafe() or ErrorFSigSafe().
| The offending log format message is:
|
| Fatal server error:
|
| (EE)
| (EE) Backtrace:
| (EE)

With old images only the process causing OOM were stopped. I did not
fully cross check but I have the feeling that OOM appears much much
more often than with old images. Sometimes x- session is closed
without opening an additional application.

2. Something with udisks/gvfs/thunar-volman seems broken. When
auto-mounting, the partitions cannot be mounted by label-name. On
plugging a memory stick I see in kernel log

| VFS: could not find a valid V7 on sda.

Same kernel works properly with old images.

3. dhclient.service does not start (this has never worked):

root@overo:~# systemctl status dhclient.service
dhclient.service - Dynamic Host Configuration Protocol (DHCP)
          Loaded: loaded (/lib/systemd/system/dhclient.service; enabled)
          Active: failed (Result: resources) since Sat 2000-01-01
00:58:02 UTC; 7min ago
         Process: 177 ExecStart=/sbin/dhclient -cf
/etc/dhcp/dhclient.conf -q -lf /var/lib/dhcp/dhclient.leases
$INTERFACES (code=exited, status=0/SUCCESS)

Jan 01 00:58:01 overo systemd[1]: Starting Dynamic Host Configuration
Protocol (DHCP)...
Jan 01 00:58:02 overo systemd[1]: PID file /var/run/dhclient.pid not
readable (yet?) after start.

I wonder what happens to networkmanager once this autostart will work...

will test further...

Andreas

Re: experiences with images based on current layer heads

[Richard Purdie]

On Wed, 2013-04-03 at 16:02 +0200, Andreas Müller wrote:
> Hi,
> 
> when migration of systemd to oe-core turned to unbuildable I decided
> to freeze my images and stopped further tests. Now that many issues
> were fixed and we are close to another release I built my standard
> xfce-based image based on latest layer HEADs. With this I face some
> issues which I would like to share:
> 
> 1. X sessions crash with OOM. This is reproducible by opening
> google-maps with midori and changing to satellite view. In
> Xorg.0.log.old I see only
> 
> | (EE)
> | (EE) Backtrace:
> | (EE)
> | (EE) Segmentation fault at address 0x9
> | (EE) BUG: triggered 'if (inSignalContext)'
> | (EE) BUG: log.c:502 in LogVMessageVerb()
> | (EE) Warning: attempting to log data in a signal unsafe manner while
> in signal context.
> | Please update to check inSignalContext and/or use
> LogMessageVerbSigSafe() or ErrorFSigSafe().
> | The offending log format message is:
> |
> | Fatal server error:
> |
> | (EE)
> | (EE) Backtrace:
> | (EE)
> 
> With old images only the process causing OOM were stopped. I did not
> fully cross check but I have the feeling that OOM appears much much
> more often than with old images. Sometimes x- session is closed
> without opening an additional application.
> 
> 2. Something with udisks/gvfs/thunar-volman seems broken. When
> auto-mounting, the partitions cannot be mounted by label-name. On
> plugging a memory stick I see in kernel log
> 
> | VFS: could not find a valid V7 on sda.
> 
> Same kernel works properly with old images.
> 
> 3. dhclient.service does not start (this has never worked):
> 
> root@overo:~# systemctl status dhclient.service
> dhclient.service - Dynamic Host Configuration Protocol (DHCP)
>           Loaded: loaded (/lib/systemd/system/dhclient.service; enabled)
>           Active: failed (Result: resources) since Sat 2000-01-01
> 00:58:02 UTC; 7min ago
>          Process: 177 ExecStart=/sbin/dhclient -cf
> /etc/dhcp/dhclient.conf -q -lf /var/lib/dhcp/dhclient.leases
> $INTERFACES (code=exited, status=0/SUCCESS)
> 
> Jan 01 00:58:01 overo systemd[1]: Starting Dynamic Host Configuration
> Protocol (DHCP)...
> Jan 01 00:58:02 overo systemd[1]: PID file /var/run/dhclient.pid not
> readable (yet?) after start.
> 
> I wonder what happens to networkmanager once this autostart will work...
> 
> will test further...

The data is certainly interesting, thanks for sharing. It would
obviously be good to dig into and fix these issues, at least tracking
the ones from OE-Core in the bugzilla might help so people are aware of
the issue and of any progress that is made by anyone looking into it...

Cheers,

Richard

Re: experiences with images based on current layer heads

[Andreas Müller]

On Wed, Apr 3, 2013 at 4:02 PM, Andreas Müller
<schnitzeltony@googlemail.com> wrote:
> Hi,
>
> when migration of systemd to oe-core turned to unbuildable I decided
> to freeze my images and stopped further tests. Now that many issues
> were fixed and we are close to another release I built my standard
> xfce-based image based on latest layer HEADs. With this I face some
> issues which I would like to share:
>
> 1. X sessions crash with OOM. This is reproducible by opening
> google-maps with midori and changing to satellite view. In
> Xorg.0.log.old I see only
>
> | (EE)
> | (EE) Backtrace:
> | (EE)
> | (EE) Segmentation fault at address 0x9
> | (EE) BUG: triggered 'if (inSignalContext)'
> | (EE) BUG: log.c:502 in LogVMessageVerb()
> | (EE) Warning: attempting to log data in a signal unsafe manner while
> in signal context.
> | Please update to check inSignalContext and/or use
> LogMessageVerbSigSafe() or ErrorFSigSafe().
> | The offending log format message is:
> |
> | Fatal server error:
> |
> | (EE)
> | (EE) Backtrace:
> | (EE)
>
seems we are not alone [1]

[1] http://www.raspberrypi.org/phpBB3/viewtopic.php?t=38568&p=321298

Andreas

Re: experiences with images based on current layer heads

[Mark Hatle]

If you are unable to look into this yourself, can you please file a bug on the 
bugzilla.yoctoproject.org website?  This issue is likely to get more attention 
if there is an active bug with reproducer steps.  (What you have below may be 
enough for someone who understands X and xfce to reproduce it, but I'm not sure.)

(I've very lightly used X under oe-core master, but I've not done anything more 
then xterms myself.)

It would of course be even better if you could figure out a proper test case to 
reproduce this, but that's not required...

--Mark

On 4/4/13 5:29 PM, Andreas Müller wrote:
> On Wed, Apr 3, 2013 at 4:02 PM, Andreas Müller
> <schnitzeltony@googlemail.com> wrote:
>> Hi,
>>
>> when migration of systemd to oe-core turned to unbuildable I decided
>> to freeze my images and stopped further tests. Now that many issues
>> were fixed and we are close to another release I built my standard
>> xfce-based image based on latest layer HEADs. With this I face some
>> issues which I would like to share:
>>
>> 1. X sessions crash with OOM. This is reproducible by opening
>> google-maps with midori and changing to satellite view. In
>> Xorg.0.log.old I see only
>>
>> | (EE)
>> | (EE) Backtrace:
>> | (EE)
>> | (EE) Segmentation fault at address 0x9
>> | (EE) BUG: triggered 'if (inSignalContext)'
>> | (EE) BUG: log.c:502 in LogVMessageVerb()
>> | (EE) Warning: attempting to log data in a signal unsafe manner while
>> in signal context.
>> | Please update to check inSignalContext and/or use
>> LogMessageVerbSigSafe() or ErrorFSigSafe().
>> | The offending log format message is:
>> |
>> | Fatal server error:
>> |
>> | (EE)
>> | (EE) Backtrace:
>> | (EE)
>>
> seems we are not alone [1]
>
> [1] http://www.raspberrypi.org/phpBB3/viewtopic.php?t=38568&p=321298
>
> Andreas
>
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core
>

Re: experiences with images based on current layer heads

[Andreas Müller]

On Fri, Apr 5, 2013 at 5:54 PM, Mark Hatle <mark.hatle@windriver.com> wrote:
> If you are unable to look into this yourself, can you please file a bug on
> the bugzilla.yoctoproject.org website?  This issue is likely to get more
> attention if there is an active bug with reproducer steps.  (What you have
> below may be enough for someone who understands X and xfce to reproduce it,
> but I'm not sure.)
>
> (I've very lightly used X under oe-core master, but I've not done anything
> more then xterms myself.)
>
> It would of course be even better if you could figure out a proper test case
> to reproduce this, but that's not required...
>
> --Mark
>
>
> On 4/4/13 5:29 PM, Andreas Müller wrote:
>>
>> On Wed, Apr 3, 2013 at 4:02 PM, Andreas Müller
>> <schnitzeltony@googlemail.com> wrote:
>>>
>>> Hi,
>>>
>>> when migration of systemd to oe-core turned to unbuildable I decided
>>> to freeze my images and stopped further tests. Now that many issues
>>> were fixed and we are close to another release I built my standard
>>> xfce-based image based on latest layer HEADs. With this I face some
>>> issues which I would like to share:
>>>
>>> 1. X sessions crash with OOM. This is reproducible by opening
>>> google-maps with midori and changing to satellite view. In
>>> Xorg.0.log.old I see only
>>>
>>> | (EE)
>>> | (EE) Backtrace:
>>> | (EE)
>>> | (EE) Segmentation fault at address 0x9
>>> | (EE) BUG: triggered 'if (inSignalContext)'
>>> | (EE) BUG: log.c:502 in LogVMessageVerb()
>>> | (EE) Warning: attempting to log data in a signal unsafe manner while
>>> in signal context.
>>> | Please update to check inSignalContext and/or use
>>> LogMessageVerbSigSafe() or ErrorFSigSafe().
>>> | The offending log format message is:
>>> |
>>> | Fatal server error:
>>> |
>>> | (EE)
>>> | (EE) Backtrace:
>>> | (EE)
>>>
>> seems we are not alone [1]
>>
>> [1] http://www.raspberrypi.org/phpBB3/viewtopic.php?t=38568&p=321298
>>
>> Andreas
I am not a friend of bugzillas because notification is missing as long
as I don't add a comment. Anyway I will add a description for xorg
problem.

Meanwhile I found another issue with libpng 1.6.0 - see
https://bugzilla.yoctoproject.org/show_bug.cgi?id=4060

Andreas

Re: experiences with images based on current layer heads

[Andreas Müller]

On Wed, Apr 3, 2013 at 4:02 PM, Andreas Müller
<schnitzeltony@googlemail.com> wrote:
> Hi,
>
> when migration of systemd to oe-core turned to unbuildable I decided
> to freeze my images and stopped further tests. Now that many issues
> were fixed and we are close to another release I built my standard
> xfce-based image based on latest layer HEADs. With this I face some
> issues which I would like to share:
>
> 1. X sessions crash with OOM. This is reproducible by opening
> google-maps with midori and changing to satellite view. In
> Xorg.0.log.old I see only
>
> | (EE)
> | (EE) Backtrace:
> | (EE)
> | (EE) Segmentation fault at address 0x9
> | (EE) BUG: triggered 'if (inSignalContext)'
> | (EE) BUG: log.c:502 in LogVMessageVerb()
> | (EE) Warning: attempting to log data in a signal unsafe manner while
> in signal context.
> | Please update to check inSignalContext and/or use
> LogMessageVerbSigSafe() or ErrorFSigSafe().
> | The offending log format message is:
> |
> | Fatal server error:
> |
> | (EE)
> | (EE) Backtrace:
> | (EE)

Seems I found this one: it is caused by undefined signedness for char.
ARM uses unsigned by default but in the code we find

if (pClient->smart_priority < 0)

After final tests and sending upstream I'll send a patch here.

Andreas

Re: experiences with images based on current layer heads

[Richard Purdie]

On Tue, 2013-04-16 at 14:17 +0200, Andreas Müller wrote:
> On Wed, Apr 3, 2013 at 4:02 PM, Andreas Müller
> <schnitzeltony@googlemail.com> wrote:
> > Hi,
> >
> > when migration of systemd to oe-core turned to unbuildable I decided
> > to freeze my images and stopped further tests. Now that many issues
> > were fixed and we are close to another release I built my standard
> > xfce-based image based on latest layer HEADs. With this I face some
> > issues which I would like to share:
> >
> > 1. X sessions crash with OOM. This is reproducible by opening
> > google-maps with midori and changing to satellite view. In
> > Xorg.0.log.old I see only
> >
> > | (EE)
> > | (EE) Backtrace:
> > | (EE)
> > | (EE) Segmentation fault at address 0x9
> > | (EE) BUG: triggered 'if (inSignalContext)'
> > | (EE) BUG: log.c:502 in LogVMessageVerb()
> > | (EE) Warning: attempting to log data in a signal unsafe manner while
> > in signal context.
> > | Please update to check inSignalContext and/or use
> > LogMessageVerbSigSafe() or ErrorFSigSafe().
> > | The offending log format message is:
> > |
> > | Fatal server error:
> > |
> > | (EE)
> > | (EE) Backtrace:
> > | (EE)
> 
> Seems I found this one: it is caused by undefined signedness for char.
> ARM uses unsigned by default but in the code we find
> 
> if (pClient->smart_priority < 0)
> 
> After final tests and sending upstream I'll send a patch here.

Nice work in tracking that down! :)

Cheers,

Richard