[PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Darren Hart]

The root cause of the matchbox-panel segfault is the same as for the 
previous two bugs addressing the same issue:

o Bug 224 - [Netbook/emenlow] matchbox-panel segfault after X startup
o Bug 738 - matchbox-panel segfault after X startup

There apparently is a bug in gcc which results in the omission of some 
code leading to the segfault on core2. The following flags resolve the 
problem.

FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers -O2 
-ggdb -feliminate-unused-debug-types"

We can continue to patch each new machine's gtk+ recipe that is based on 
atom-pc.conf, or we can just add the above to tune-atom.conf. I've 
tested both on both the n450 and a Toshiba NB305 netbook, and each 
solution is effective.

Chaging it in tune-atom.conf likely fixes other bugs we either haven't 
solved yet or simply haven't hit. It is also much more invasive, and we 
may not be willing to accept that this close to 1.0. The patches follow, 
and we can include one or the other.

Preferences?

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

[PATCH A] n450: add optimization flags to gtk+ for n450 to avoid gcc core2 bug

[Darren Hart]

Fixes [YOCTO #853]

Without these added optimization flags, the matchbox-panel (and possibly other)
applications would segfault. This patch applies the change only to the gtk+
recipe for the n450 bsp.

Signed-off-by: Darren Hart <dvhart@linux.intel.com>
CC: Ke Yu <ke.yu@intel.com>
---
 meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
 create mode 100644 meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend

diff --git a/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend b/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
new file mode 100644
index 0000000..3af4187
--- /dev/null
+++ b/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
@@ -0,0 +1,2 @@
+#-fomit-frame-pointer in default FULL_OPTIMIZATION will cause matchbox-panel segfault on n450
+FULL_OPTIMIZATION_n450 = "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
-- 
1.7.1

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

[PATCH B] atom-pc: work around gcc bug for core2

[Darren Hart]

Fixes [YOCTO #853]

Without these added optimization flags, the matchbox-panel (and possibly other)
applications would segfault. This patch applies the changes to all machines
derived from atom-pc.conf.

Signed-off-by: Darren Hart <dvhart@linux.intel.com>
CC: Ke Yu <ke.yu@intel.com>
---
 meta/conf/machine/include/tune-atom.inc |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/meta/conf/machine/include/tune-atom.inc b/meta/conf/machine/include/tune-atom.inc
index a401856..db78b0d 100644
--- a/meta/conf/machine/include/tune-atom.inc
+++ b/meta/conf/machine/include/tune-atom.inc
@@ -3,3 +3,7 @@ TARGET_CC_ARCH = "-m32 -march=core2 -msse3 -mtune=generic -mfpmath=sse"
 #MOBLIN_CFLAGS = "-Os -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=4 -fasynchronous-unwind-tables"
 
 PACKAGE_EXTRA_ARCHS += "x86 i386 i486 i586 core2"
+
+# Work around a gcc bug for core2 which caused matchbox-panel to segfault
+# and possibly others.
+FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
-- 
1.7.1

Re: [PATCH A] n450: add optimization flags to gtk+ for n450 to avoid gcc core2 bug

[Darren Hart]

On 03/16/2011 02:49 PM, Darren Hart wrote:
>
> Fixes [YOCTO #853]
>
> Without these added optimization flags, the matchbox-panel (and possibly other)
> applications would segfault. This patch applies the change only to the gtk+
> recipe for the n450 bsp.
>
> Signed-off-by: Darren Hart<dvhart@linux.intel.com>
> CC: Ke Yu<ke.yu@intel.com>
> ---
>   meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend |    2 ++
>   1 files changed, 2 insertions(+), 0 deletions(-)
>   create mode 100644 meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
>
> diff --git a/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend b/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
> new file mode 100644
> index 0000000..3af4187
> --- /dev/null
> +++ b/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
> @@ -0,0 +1,2 @@
> +#-fomit-frame-pointer in default FULL_OPTIMIZATION will cause matchbox-panel segfault on n450
> +FULL_OPTIMIZATION_n450 = "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"


Note that this patch only addresses the problem for the n450, and it 
would remain on the atom-pc. If we don't want to take Patch B, then we 
would need another solution for atom-pc. The only thing that comes to 
mind would be to add the above to the core gtk+ recipe in the form:

#-fomit-frame-pointer in default FULL_OPTIMIZATION will cause 
matchbox-panel segfault on atom-pc
FULL_OPTIMIZATION_atom-pc = "-fexpensive-optimizations 
-frename-registers -O2 -ggdb -feliminate-unused-debug-types"


Thoughts?

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: [PATCH A] n450: add optimization flags to gtk+ for n450 to avoid gcc core2 bug

[Darren Hart]

On 03/16/2011 03:09 PM, Darren Hart wrote:
> On 03/16/2011 02:49 PM, Darren Hart wrote:
>>
>> Fixes [YOCTO #853]
>>
>> Without these added optimization flags, the matchbox-panel (and
>> possibly other)
>> applications would segfault. This patch applies the change only to the
>> gtk+
>> recipe for the n450 bsp.
>>
>> Signed-off-by: Darren Hart<dvhart@linux.intel.com>
>> CC: Ke Yu<ke.yu@intel.com>
>> ---
>> meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend | 2 ++
>> 1 files changed, 2 insertions(+), 0 deletions(-)
>> create mode 100644 meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
>>
>> diff --git a/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
>> b/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
>> new file mode 100644
>> index 0000000..3af4187
>> --- /dev/null
>> +++ b/meta-n450/recipes-gnome/gtk+/gtk+_2.22.1.bbappend
>> @@ -0,0 +1,2 @@
>> +#-fomit-frame-pointer in default FULL_OPTIMIZATION will cause
>> matchbox-panel segfault on n450
>> +FULL_OPTIMIZATION_n450 = "-fexpensive-optimizations
>> -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
>
>
> Note that this patch only addresses the problem for the n450, and it
> would remain on the atom-pc. If we don't want to take Patch B, then we
> would need another solution for atom-pc. The only thing that comes to
> mind would be to add the above to the core gtk+ recipe in the form:
>
> #-fomit-frame-pointer in default FULL_OPTIMIZATION will cause
> matchbox-panel segfault on atom-pc
> FULL_OPTIMIZATION_atom-pc = "-fexpensive-optimizations
> -frename-registers -O2 -ggdb -feliminate-unused-debug-types"

Guess I should have looked at the recipe first :-) It already does this. 
I'll push the n450 patch per Saul's preference of minimal change.

--
Darren

>
>
> Thoughts?
>

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: [PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Richard Purdie]

On Wed, 2011-03-16 at 14:47 -0700, Darren Hart wrote:
> The root cause of the matchbox-panel segfault is the same as for the 
> previous two bugs addressing the same issue:
> 
> o Bug 224 - [Netbook/emenlow] matchbox-panel segfault after X startup
> o Bug 738 - matchbox-panel segfault after X startup
> 
> There apparently is a bug in gcc which results in the omission of some 
> code leading to the segfault on core2. The following flags resolve the 
> problem.
> 
> FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers -O2 
> -ggdb -feliminate-unused-debug-types"
> 
> We can continue to patch each new machine's gtk+ recipe that is based on 
> atom-pc.conf, or we can just add the above to tune-atom.conf. I've 
> tested both on both the n450 and a Toshiba NB305 netbook, and each 
> solution is effective.
> 
> Chaging it in tune-atom.conf likely fixes other bugs we either haven't 
> solved yet or simply haven't hit. It is also much more invasive, and we 
> may not be willing to accept that this close to 1.0. The patches follow, 
> and we can include one or the other.
> 
> Preferences?

I'm in favour of C which is to add this to tune-atom.inc:

FULL_OPTIMIZATION_pn-gtk+ = "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"

:)

Cheers,

Richard

Re: [PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Stewart, David C]

Great job finding this, Darren!

>-----Original Message-----
>From: poky-bounces@yoctoproject.org [mailto:poky-
>bounces@yoctoproject.org] On Behalf Of Richard Purdie
>Sent: Wednesday, March 16, 2011 5:14 PM
>To: Darren Hart
>Cc: poky@yoctoproject.org
>Subject: Re: [poky] [PATCH A or B] Fix for bug #853 - matchbox-panel
>segfaults, X becomes unusable
>
>On Wed, 2011-03-16 at 14:47 -0700, Darren Hart wrote:
>> The root cause of the matchbox-panel segfault is the same as for the
>> previous two bugs addressing the same issue:
>>
>> o Bug 224 - [Netbook/emenlow] matchbox-panel segfault after X startup
>> o Bug 738 - matchbox-panel segfault after X startup
>>
>> There apparently is a bug in gcc which results in the omission of some
>> code leading to the segfault on core2. The following flags resolve the
>> problem.
>>
>> FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers -O2
>> -ggdb -feliminate-unused-debug-types"
>>
>> We can continue to patch each new machine's gtk+ recipe that is based
>on
>> atom-pc.conf, or we can just add the above to tune-atom.conf. I've
>> tested both on both the n450 and a Toshiba NB305 netbook, and each
>> solution is effective.
>>
>> Chaging it in tune-atom.conf likely fixes other bugs we either haven't
>> solved yet or simply haven't hit. It is also much more invasive, and
>we
>> may not be willing to accept that this close to 1.0. The patches
>follow,
>> and we can include one or the other.
>>
>> Preferences?
>
>I'm in favour of C which is to add this to tune-atom.inc:
>
>FULL_OPTIMIZATION_pn-gtk+ = "-fexpensive-optimizations -frename-
>registers -O2 -ggdb -feliminate-unused-debug-types"
>
>:)
>
>Cheers,
>
>Richard
>
>
>_______________________________________________
>poky mailing list
>poky@yoctoproject.org
>https://lists.yoctoproject.org/listinfo/poky

Re: [PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Darren Hart]

On 03/16/2011 05:15 PM, Stewart, David C wrote:
> Great job finding this, Darren!

Thanks, but credit also goes to Ke who confirmed that this was the issue 
after I dismissed it because my build didn't reflect the change! Go Team.

:)

--
Darren

>
>> -----Original Message-----
>> From: poky-bounces@yoctoproject.org [mailto:poky-
>> bounces@yoctoproject.org] On Behalf Of Richard Purdie
>> Sent: Wednesday, March 16, 2011 5:14 PM
>> To: Darren Hart
>> Cc: poky@yoctoproject.org
>> Subject: Re: [poky] [PATCH A or B] Fix for bug #853 - matchbox-panel
>> segfaults, X becomes unusable
>>
>> On Wed, 2011-03-16 at 14:47 -0700, Darren Hart wrote:
>>> The root cause of the matchbox-panel segfault is the same as for the
>>> previous two bugs addressing the same issue:
>>>
>>> o Bug 224 - [Netbook/emenlow] matchbox-panel segfault after X startup
>>> o Bug 738 - matchbox-panel segfault after X startup
>>>
>>> There apparently is a bug in gcc which results in the omission of some
>>> code leading to the segfault on core2. The following flags resolve the
>>> problem.
>>>
>>> FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers -O2
>>> -ggdb -feliminate-unused-debug-types"
>>>
>>> We can continue to patch each new machine's gtk+ recipe that is based
>> on
>>> atom-pc.conf, or we can just add the above to tune-atom.conf. I've
>>> tested both on both the n450 and a Toshiba NB305 netbook, and each
>>> solution is effective.
>>>
>>> Chaging it in tune-atom.conf likely fixes other bugs we either haven't
>>> solved yet or simply haven't hit. It is also much more invasive, and
>> we
>>> may not be willing to accept that this close to 1.0. The patches
>> follow,
>>> and we can include one or the other.
>>>
>>> Preferences?
>>
>> I'm in favour of C which is to add this to tune-atom.inc:
>>
>> FULL_OPTIMIZATION_pn-gtk+ = "-fexpensive-optimizations -frename-
>> registers -O2 -ggdb -feliminate-unused-debug-types"
>>
>> :)
>>
>> Cheers,
>>
>> Richard
>>
>>
>> _______________________________________________
>> poky mailing list
>> poky@yoctoproject.org
>> https://lists.yoctoproject.org/listinfo/poky

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: [PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Yu Ke]

On Mar 17, 00:13, Richard Purdie wrote:
> On Wed, 2011-03-16 at 14:47 -0700, Darren Hart wrote:
> > The root cause of the matchbox-panel segfault is the same as for the 
> > previous two bugs addressing the same issue:
> > 
> > o Bug 224 - [Netbook/emenlow] matchbox-panel segfault after X startup
> > o Bug 738 - matchbox-panel segfault after X startup
> > 
> > There apparently is a bug in gcc which results in the omission of some 
> > code leading to the segfault on core2. The following flags resolve the 
> > problem.
> > 
> > FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers -O2 
> > -ggdb -feliminate-unused-debug-types"
> > 
> > We can continue to patch each new machine's gtk+ recipe that is based on 
> > atom-pc.conf, or we can just add the above to tune-atom.conf. I've 
> > tested both on both the n450 and a Toshiba NB305 netbook, and each 
> > solution is effective.
> > 
> > Chaging it in tune-atom.conf likely fixes other bugs we either haven't 
> > solved yet or simply haven't hit. It is also much more invasive, and we 
> > may not be willing to accept that this close to 1.0. The patches follow, 
> > and we can include one or the other.
> > 
> > Preferences?
> 
> I'm in favour of C which is to add this to tune-atom.inc:
> 
> FULL_OPTIMIZATION_pn-gtk+ = "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
> 
> :)
> 
> Cheers,
> 
> Richard

I have an option D :)

i.e. in gtk+_2.22.1.bb, add:
FULL_OPTIMIZATION_core2= "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"

because this issue is caused by gcc optimization for core2 arch, so in theory, all machine with core2 arch setting will be affected, no only limited to atom machine

Regards
Ke

Re: [PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Darren Hart]

On 03/16/2011 07:13 PM, Yu Ke wrote:
> On Mar 17, 00:13, Richard Purdie wrote:
>> On Wed, 2011-03-16 at 14:47 -0700, Darren Hart wrote:
>>> The root cause of the matchbox-panel segfault is the same as for
>>> the previous two bugs addressing the same issue:
>>>
>>> o Bug 224 - [Netbook/emenlow] matchbox-panel segfault after X
>>> startup o Bug 738 - matchbox-panel segfault after X startup
>>>
>>> There apparently is a bug in gcc which results in the omission of
>>> some code leading to the segfault on core2. The following flags
>>> resolve the problem.
>>>
>>> FULL_OPTIMIZATION = "-fexpensive-optimizations -frename-registers
>>> -O2 -ggdb -feliminate-unused-debug-types"
>>>
>>> We can continue to patch each new machine's gtk+ recipe that is
>>> based on atom-pc.conf, or we can just add the above to
>>> tune-atom.conf. I've tested both on both the n450 and a Toshiba
>>> NB305 netbook, and each solution is effective.
>>>
>>> Chaging it in tune-atom.conf likely fixes other bugs we either
>>> haven't solved yet or simply haven't hit. It is also much more
>>> invasive, and we may not be willing to accept that this close to
>>> 1.0. The patches follow, and we can include one or the other.
>>>
>>> Preferences?
>>
>> I'm in favour of C which is to add this to tune-atom.inc:
>>
>> FULL_OPTIMIZATION_pn-gtk+ = "-fexpensive-optimizations
>> -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
>>
>> :)
>>
>> Cheers,
>>
>> Richard
>
> I have an option D :)
>
> i.e. in gtk+_2.22.1.bb, add: FULL_OPTIMIZATION_core2=
> "-fexpensive-optimizations -frename-registers -O2 -ggdb
> -feliminate-unused-debug-types"
>
> because this issue is caused by gcc optimization for core2 arch, so
> in theory, all machine with core2 arch setting will be affected, no
> only limited to atom machine

There are apparently at least 4 ways to do this :-) RP has merged his 
change (version C) to bernard and I've tested it on atom-pc. I've 
removed the existing changes from the original gtk+ recipe and the 
various bbappends in meta-intel. I'm testing on n450 now. If that works 
I'll send the patch against gtk+ and commit those to bernard and master 
for meta-intel.

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel

Re: [PATCH A or B] Fix for bug #853 - matchbox-panel segfaults, X becomes unusable

[Richard Purdie]

On Thu, 2011-03-17 at 10:13 +0800, Yu Ke wrote:
> On Mar 17, 00:13, Richard Purdie wrote:
> > I'm in favour of C which is to add this to tune-atom.inc:
> > 
> > FULL_OPTIMIZATION_pn-gtk+ = "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
>
>
> I have an option D :)
> 
> i.e. in gtk+_2.22.1.bb, add:
> FULL_OPTIMIZATION_core2= "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"
> 
> because this issue is caused by gcc optimization for core2 arch, so in theory, all machine with core2 arch setting will be affected, no only limited to atom machine

This wouldn't work as "core2" is PACKAGE_ARCH and PACKAGE_ARCH is not in
OVERRIDES. I did briefly discuss this over jabber and you could do:

FULL_OPTIMIZATION_i586= "-fexpensive-optimizations -frename-registers -O2 -ggdb -feliminate-unused-debug-types"

since TARGET_ARCH (i586) is in OVERRIDES but this would then affect
machines like qemux86 and I decided we didn't want to do that at this
point in the release cycle.

Cheers,

Richard