Skip to contentSkip to conversation

BUG in Linux 2.6.16/2.6.16.1 (compilation failure of third party software)

PATCH Thierry Godefroy 3 participants 4 messages Mar 28, 2006 – Mar 29, 2006 Locking primitives

lore
Share & tools
lore
#0 lore raw

Paragon_NTFS_3.x.v5.1 fails to compile (with gcc v3.4.3) with the following errors:

gcc -DMODULE -fshort-wchar -DUFSD_NTFS=1 -DUFSD_NTFS_SECURITY -DUFSD_NTFS_OBJECTID -DUFSD_NTFS_WRITE_ENABLE -DNDEBUG -DUFSD_DEVICE=ufsd -DUFSD_READONLY=0 -DKBUILD_BASENAME=_ufsdvfs_ -include fs_conf.h -pipe -Wall -fno-exceptions -Wno-multichar -Wstrict-prototypes -Wno-unknown-pragmas -mpreferred-stack-boundary=2 -march=k8 -I /usr/src/linux/include/asm/mach-default -g0 -O2 -fomit-frame-pointer -fno-strict-aliasing -fno-common -c -I /usr/src/linux/include -I "/include" -I/include ufsdvfs.c -o objfre/vfs/ufsdvfs.o
In file included from /usr/src/linux/include/linux/rwsem.h:27,

                 from /usr/src/linux/include/asm/semaphore.h:42,
                 from /usr/src/linux/include/linux/sched.h:20,
                 from /usr/src/linux/include/linux/module.h:10,
                 from ufsdvfs.c:30:

/usr/src/linux/include/asm/rwsem.h: In function `__down_read':
/usr/src/linux/include/asm/rwsem.h:105: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/rwsem.h: In function `__down_write':
/usr/src/linux/include/asm/rwsem.h:157: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/rwsem.h: In function `__up_read':
/usr/src/linux/include/asm/rwsem.h:194: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/rwsem.h:188: warning: unused variable tmp' /usr/src/linux/include/asm/rwsem.h: In function __up_write':
/usr/src/linux/include/asm/rwsem.h:220: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/rwsem.h: In function `__downgrade_write':
/usr/src/linux/include/asm/rwsem.h:245: error: syntax error before "_ufsdvfs_" In file included from /usr/src/linux/include/linux/sched.h:20,

                 from /usr/src/linux/include/linux/module.h:10,
                 from ufsdvfs.c:30:

/usr/src/linux/include/asm/semaphore.h: In function `down':
/usr/src/linux/include/asm/semaphore.h:105: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/semaphore.h: In function `down_interruptible':
/usr/src/linux/include/asm/semaphore.h:130: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/semaphore.h: In function `down_trylock':
/usr/src/linux/include/asm/semaphore.h:155: error: syntax error before "_ufsdvfs_" /usr/src/linux/include/asm/semaphore.h: In function `up':
/usr/src/linux/include/asm/semaphore.h:179: error: syntax error before "_ufsdvfs_" ufsdvfs.c: In function `ufsd_readdir':
ufsdvfs.c:864: warning: implicit declaration of function `update_atime' ufsdvfs.c: At top level:
ufsdvfs.c:2219: warning: initialization from incompatible pointer type
make: *** [objfre/vfs/ufsdvfs.o] Erreur 1

------

I tracked down the problem to a patch of include/linux/spinlock.h which was introduced somewhere in 2.6.16:
diff -urN linux-2.6.15/include/linux/spinlock.h linux-2.6.16.1/include/linux/spinlock.h gives:

include/linux/spinlock.h 2006-03-24 19:55:28.000000000 +0100 renamed from include/linux/spinlock.h 2006-03-28 19:52:29.000000000 +0200 +1 -2
@@ -59,8 +59,7 @@
 /*
  * Must define these before including other files, inline functions need them
  */
-#define LOCK_SECTION_NAME                       \
-        ".text.lock." __stringify(KBUILD_BASENAME)
+#define LOCK_SECTION_NAME ".text.lock."KBUILD_BASENAME
 
 #define LOCK_SECTION_START(extra)               \
         ".subsection 1\n\t"                     \

Reverting this patch fixes the compilation problem.

Thierry Godefroy.

#1 lore raw
Quoted 2 lines · Thierry Godefroy

Paragon_NTFS_3.x.v5.1 fails to compile (with gcc v3.4.3) with the following errors:

It's not a bug when an upgrade causes third party kernel modules not to compile - your code needs to be updated to work with 2.6.16.

Linux makes no effort to guarantee source or binary compatibility with out of tree kernel modules, or userspace code that includes kernel headers.

Lee

#2 lore raw
Quoted 6 lines · Lee Revell
Quoted 2 lines · Thierry Godefroy

Paragon_NTFS_3.x.v5.1 fails to compile (with gcc v3.4.3) with the following errors:

It's not a bug when an upgrade causes third party kernel modules not to compile - your code needs to be updated to work with 2.6.16.

Oh, yeah ?... Really ?... Please, read the errors before drawing hasted conclusions... The errors occur after the mere #inclusion of Linux headers. Here is a simple "code" which will trigger the error:

#ifndef __KERNEL__
#define __KERNEL__
#endif

#include <linux/module.h>

int main() {
	return 0;
}

And I don't see anything wrong in that code...

Quoted 3 lines

Linux makes no effort to guarantee source or binary compatibility with out of tree kernel modules, or userspace code that includes kernel headers.

That's a pity... Non-regression should be guaranteed. The problem with Linux is that each new version brings a load of incompatibilities with third parties drivers (graphic cards, win/smart modems, wifi cards, etc, etc), most of which using proprietary code which sources are unavailable to the end user. The fact that Linux keeps breaking those drivers code is not going to encourage hardware vendors to make Linux drivers for their products. For example, ATI takes a couple of months to catchup with the kernel changes, and each time they release a new driver the next Linux release breaks it. :-(

Thierry Godefroy.

#3 lore raw
Quoted 17 lines · Thierry Godefroy
Quoted 6 lines · Lee Revell
Quoted 2 lines · Thierry Godefroy

Paragon_NTFS_3.x.v5.1 fails to compile (with gcc v3.4.3) with the following errors:

It's not a bug when an upgrade causes third party kernel modules not to compile - your code needs to be updated to work with 2.6.16.

Oh, yeah ?... Really ?... Please, read the errors before drawing hasted conclusions... The errors occur after the mere #inclusion of Linux headers. Here is a simple "code" which will trigger the error:

#ifndef __KERNEL__
#define __KERNEL__
#endif

Quoted 7 lines

#include <linux/module.h>

int main() {
	return 0;
}

And I don't see anything wrong in that code...

then look again. Userspace that defines __KERNEL__ .... man that isn't going ot work. __KERNEL__ is the ONLY safeguard we have to protect userspace from miscompiling, and we highly assume userspace doesn't set it. If this program does that then it really deserves all it gets.