spinlock_t typedef visibility and uninitialized spinlock
I'm porting CIPE 1.6.0 kernel module to Kernel 2.6.8 and had problems with "spin_is_locked on uninitialized spinlock".
After tracing the problem I found that the spinlock_t structure is not visible to the module code. A 'gcc -E' yields:
typedef struct { } spinlock_t;In spinlock.h, this declaration is inside a #ifdef CONFIG_DEBUG_SPINLOCK block, so it becomes visible only CONFIG_DEBUG_SPINLOCK is 'y'.
If I turn CONFIG_DEBUG_SPINLOCK on, the module loads nicely. Otherwise I get a nasty error in syslog and sometimes a system crash, as if in CIPE the struct was not allocated (what is the case if the compiler uses the typedef as it is above).
The question is: is this a bug or a feature? ;-)
Should the declaration be out of the #ifdef CONFIG_DEBUG_SPINLOCK? Or should I use a special compiler flag?
Carlos
Quoted 16 lines · Carlos Vidal
I'm porting CIPE 1.6.0 kernel module to Kernel 2.6.8 and had problems with "
spin_is_lockedon uninitialized spinlock".After tracing the problem I found that the
spinlock_tstructure is not visible to the module code. A 'gcc -E' yields:typedef struct { } spinlock_t;In
spinlock.h, this declaration is inside a #ifdefCONFIG_DEBUG_SPINLOCKblock, so it becomes visible onlyCONFIG_DEBUG_SPINLOCKis 'y'.If I turn
CONFIG_DEBUG_SPINLOCKon, the module loads nicely. Otherwise I get a nasty error in syslog and sometimes a system crash, as if in CIPE the struct was not allocated (what is the case if the compiler uses the typedef as it is above).
Quoted 4 lines
The question is: is this a bug or a feature? ;-)
Should the declaration be out of the #ifdef
CONFIG_DEBUG_SPINLOCK? Or should I use a special compiler flag?
Try including <linux/config.h> at the top of the compilation unit and run through the preprocessor again. By the way, spinlock code got shuffled around in 2.6.9...
Zwane