Why spinlock in linux kernel is in the ".subsection 1" (or ".text.lock.smth")?

In linux kernel in the implementation of spinlocks, e.g. http://lxr.linux.no/#linux+v2.6.18/include/asm-i386/semaphore.h#L97

  97static inline void down(struct semaphore * sem)
  98{
  99        might_sleep();
 100        __asm__ __volatile__(
 101                "# atomic down operation\n\t"
 102                LOCK_PREFIX "decl %0\n\t"     /* --sem->count */
 103                "js 2f\n"
 104                "1:\n"
 105                LOCK_SECTION_START("")
 106                "2:\tlea %0,%%eax\n\t"
 107                "call __down_failed\n\t"
 108                "jmp 1b\n"
 109                LOCK_SECTION_END
 110                :"+m" (sem->count)
 111                :
 112                :"memory","ax");
 113}

LOCK_SECTION_START and LOCK_SECTION_END are used. They are defined in http://lxr.linux.no/#linux+v2.6.18/include/linux/spinlock.h#L63 as

  61#define LOCK_SECTION_NAME ".text.lock."KBUILD_BASENAME
  62
  63#define LOCK_SECTION_START(extra)               \
  64        ".subsection 1\n\t"                     \
  65        extra                                   \
  66        ".ifndef " LOCK_SECTION_NAME "\n\t"     \
  67       开发者_如何学JAVA LOCK_SECTION_NAME ":\n\t"               \
  68        ".endif\n"
  69
  70#define LOCK_SECTION_END                        \
  71        ".previous\n\t"

So all locked operations are partly putted in subsection 1 or section .text.lock.SMTH_STRING.

What reason is for it?

I'm not 100% sure, but I think this is related to SMP alternatives for i386, where kernel can switch between SMP and UP at boot or even at runtime.