[BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[Rob Landley]

"make headers_install" copies include/mtd/ubi_user.h to userspace, where
things like busybox link against it.  With 3.0, building defconfig
busybox gets this error:

In file included from miscutils/ubi_tools.c:63:
/home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:329:
error: conflicting types for '__packed'
/home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:313:
error: previous declaration of '__packed' was here

This is due to the Linux 3.0 commit in the title:

    UBI: use __packed instead of __attribute__((packed))

This relies on __packed being defined in:

  include/linux/compiler-gcc.h:
  #define __packed			__attribute__((packed))

Which is not exported to userspace.

I.E. this "cleanup" broke the ability for userspace to use this header,
even though the header is exported to userspace by headers_install.

Rob

Re: [BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[Stefan Richter]

On Aug 14 Rob Landley wrote:
> "make headers_install" copies include/mtd/ubi_user.h to userspace, where
> things like busybox link against it.  With 3.0, building defconfig
> busybox gets this error:
> 
> In file included from miscutils/ubi_tools.c:63:
> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:329:
> error: conflicting types for '__packed'
> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:313:
> error: previous declaration of '__packed' was here
> 
> This is due to the Linux 3.0 commit in the title:
> 
>     UBI: use __packed instead of __attribute__((packed))
> 
> This relies on __packed being defined in:
> 
>   include/linux/compiler-gcc.h:
>   #define __packed			__attribute__((packed))
> 
> Which is not exported to userspace.
> 
> I.E. this "cleanup" broke the ability for userspace to use this header,
> even though the header is exported to userspace by headers_install.

As far as I can tell, the __packed or __attribute__((packed))
qualifications are superfluous in include/mtd/ubi-user.h anyway.
-- 
Stefan Richter
-=====-==-== =--- -===-
http://arcgraph.de/sr/

Re: [BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[Rob Landley]

On 08/14/2011 01:03 PM, Stefan Richter wrote:
> On Aug 14 Rob Landley wrote:
>> "make headers_install" copies include/mtd/ubi_user.h to userspace, where
>> things like busybox link against it.  With 3.0, building defconfig
>> busybox gets this error:
>>
>> In file included from miscutils/ubi_tools.c:63:
>> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:329:
>> error: conflicting types for '__packed'
>> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:313:
>> error: previous declaration of '__packed' was here
>>
>> This is due to the Linux 3.0 commit in the title:
>>
>>     UBI: use __packed instead of __attribute__((packed))
>>
>> This relies on __packed being defined in:
>>
>>   include/linux/compiler-gcc.h:
>>   #define __packed			__attribute__((packed))
>>
>> Which is not exported to userspace.
>>
>> I.E. this "cleanup" broke the ability for userspace to use this header,
>> even though the header is exported to userspace by headers_install.
> 
> As far as I can tell, the __packed or __attribute__((packed))
> qualifications are superfluous in include/mtd/ubi-user.h anyway.

And the email address he signed-off-by the broken commit from bounced,
trying again with the one out of MAINTAINERS.

Rob

Re: [BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[Rob Landley]

On 08/14/2011 01:03 PM, Stefan Richter wrote:
> On Aug 14 Rob Landley wrote:
>> "make headers_install" copies include/mtd/ubi_user.h to userspace, where
>> things like busybox link against it.  With 3.0, building defconfig
>> busybox gets this error:
>>
>> In file included from miscutils/ubi_tools.c:63:
>> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:329:
>> error: conflicting types for '__packed'
>> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:313:
>> error: previous declaration of '__packed' was here
>>
>> This is due to the Linux 3.0 commit in the title:
>>
>>     UBI: use __packed instead of __attribute__((packed))
>>
>> This relies on __packed being defined in:
>>
>>   include/linux/compiler-gcc.h:
>>   #define __packed			__attribute__((packed))
>>
>> Which is not exported to userspace.
>>
>> I.E. this "cleanup" broke the ability for userspace to use this header,
>> even though the header is exported to userspace by headers_install.
> 
> As far as I can tell, the __packed or __attribute__((packed))
> qualifications are superfluous in include/mtd/ubi-user.h anyway.

And Adrian Hunter <adrian.hunter@nokia.com> (the other listed UBI
maintainer) also bounces.  Did the developer exodus after Nokia decided
to cozy up to The Black Widow take 'em both out?

Rob

Re: [BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[Artem Bityutskiy]

On Sun, 2011-08-14 at 14:16 -0500, Rob Landley wrote:
> On 08/14/2011 01:03 PM, Stefan Richter wrote:
> > On Aug 14 Rob Landley wrote:
> >> "make headers_install" copies include/mtd/ubi_user.h to userspace, where
> >> things like busybox link against it.  With 3.0, building defconfig
> >> busybox gets this error:
> >>
> >> In file included from miscutils/ubi_tools.c:63:
> >> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:329:
> >> error: conflicting types for '__packed'
> >> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:313:
> >> error: previous declaration of '__packed' was here
> >>
> >> This is due to the Linux 3.0 commit in the title:
> >>
> >>     UBI: use __packed instead of __attribute__((packed))
> >>
> >> This relies on __packed being defined in:
> >>
> >>   include/linux/compiler-gcc.h:
> >>   #define __packed			__attribute__((packed))
> >>
> >> Which is not exported to userspace.
> >>
> >> I.E. this "cleanup" broke the ability for userspace to use this header,
> >> even though the header is exported to userspace by headers_install.
> > 
> > As far as I can tell, the __packed or __attribute__((packed))
> > qualifications are superfluous in include/mtd/ubi-user.h anyway.
> 
> And Adrian Hunter <adrian.hunter@nokia.com> (the other listed UBI
> maintainer) also bounces.  Did the developer exodus after Nokia decided
> to cozy up to The Black Widow take 'em both out?

Yes, both of us have left Nokia recently. Adrian should fix the
"MAINTAINERS" file.

WRT to the breakage - yes, it was my mistake to follow the new trend in
the Linux kernel - the commit message has a reference to the other
commit which inspired the check, and you can find out that some other
projects made a similar change. Moreover, checkpatch.pl prints the
following warning "WARN("__packed is preferred over
__attribute__((packed))". So you hardly can blame :-)

Anyways, this problem was discussed in the "__packed vs.
__attribute__((packed)) in kernel headers" thread in LKML.

Then a fix has been sent, see the "{PATCH] fix __packed in exported
kernel headers" thread in LKML. This fix was accepted by "Michal Marek
<mmarek@suse.cz>", whom I put to CC.

-- 
Best Regards,
Artem Bityutskiy (Битюцкий Артём)

Re: [BUG] Linux 3.0 commit 3627924acf70a broke include/mtd/ubi_user.h.

[Artem Bityutskiy]

On Mon, 2011-08-15 at 06:39 +0300, Artem Bityutskiy wrote:
> On Sun, 2011-08-14 at 14:16 -0500, Rob Landley wrote:
> > On 08/14/2011 01:03 PM, Stefan Richter wrote:
> > > On Aug 14 Rob Landley wrote:
> > >> "make headers_install" copies include/mtd/ubi_user.h to userspace, where
> > >> things like busybox link against it.  With 3.0, building defconfig
> > >> busybox gets this error:
> > >>
> > >> In file included from miscutils/ubi_tools.c:63:
> > >> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:329:
> > >> error: conflicting types for '__packed'
> > >> /home/landley/aboriginal/aboriginal/build/simple-cross-compiler-i686/bin/../include/mtd/ubi-user.h:313:
> > >> error: previous declaration of '__packed' was here
> > >>
> > >> This is due to the Linux 3.0 commit in the title:
> > >>
> > >>     UBI: use __packed instead of __attribute__((packed))
> > >>
> > >> This relies on __packed being defined in:
> > >>
> > >>   include/linux/compiler-gcc.h:
> > >>   #define __packed			__attribute__((packed))
> > >>
> > >> Which is not exported to userspace.
> > >>
> > >> I.E. this "cleanup" broke the ability for userspace to use this header,
> > >> even though the header is exported to userspace by headers_install.
> > > 
> > > As far as I can tell, the __packed or __attribute__((packed))
> > > qualifications are superfluous in include/mtd/ubi-user.h anyway.
> > 
> > And Adrian Hunter <adrian.hunter@nokia.com> (the other listed UBI
> > maintainer) also bounces.  Did the developer exodus after Nokia decided
> > to cozy up to The Black Widow take 'em both out?
> 
> Yes, both of us have left Nokia recently. Adrian should fix the
> "MAINTAINERS" file.
> 
> WRT to the breakage - yes, it was my mistake to follow the new trend in
> the Linux kernel - the commit message has a reference to the other
> commit which inspired the check, and you can find out that some other
> projects made a similar change. Moreover, checkpatch.pl prints the
> following warning "WARN("__packed is preferred over
> __attribute__((packed))". So you hardly can blame :-)
> 
> Anyways, this problem was discussed in the "__packed vs.
> __attribute__((packed)) in kernel headers" thread in LKML.
> 
> Then a fix has been sent, see the "{PATCH] fix __packed in exported
> kernel headers" thread in LKML. This fix was accepted by "Michal Marek
> <mmarek@suse.cz>", whom I put to CC.

And the fix is upstream, see:

commit f210735fe2f17a6225432ee3d1239bcf23a8659c
Author: Markus Trippelsdorf <markus@trippelsdorf.de>
Date:   Fri Jun 24 15:51:00 2011 +0200

Probably someone can ping the -stable team and ask them to take it to
linux-3.0-stable?

-- 
Best Regards,
Artem Bityutskiy (Битюцкий Артём)