Very small fix for llvm/clang

[Alex Hornung]

This would fix compilation issues with llvm/clang due to the ugly gcc
extension that allows to specify a register to be used for a variable.
The clobber list of the inline assembly already specifies fsbase as
input to ecx.

The diff is attached and on http://gist.github.com/91146



Sincerely,
Alex
--- a/sys/cpu/amd64/include/cpufunc.h
+++ b/sys/cpu/amd64/include/cpufunc.h
@@ -633,7 +633,7 @@ load_es(u_int sel)
 static __inline void
 load_fs(u_int sel)
 {
- register u_int32_t fsbase __asm("ecx");
+ u_int32_t fsbase;
 
        /* Preserve the fsbase value across the selector load */
        fsbase = MSR_FSBASE;
@@ -647,8 +647,7 @@ load_fs(u_int sel)
 static __inline void
 load_gs(u_int sel)
 {
- register u_int32_t gsbase __asm("ecx");
+ u_int32_t gsbase;
 
        /*
         * Preserve the gsbase value across the selector load.
         * Note that we have to disable interrupts because the gsbase