Move CPU features into its own static library.

src/trusted/cpu_features/arch/x86/cpu_xgetbv.S

 /*
  * Copyright (c) 2012 The Native Client Authors. All rights reserved.
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 /*
  * This exists only because Windows 64 doesn't support trivial inline assembly.
  * Visual Studio 2010 SP1 has an intrinsic _xgetbv() for this.
  * But earlier versions do not.
  */
 
 #include "native_client/src/trusted/service_runtime/nacl_config.h"
 
         .text
 
 DEFINE_GLOBAL_HIDDEN_IDENTIFIER(NaClXGETBV):
+#if NACL_ARCH(NACL_BUILD_ARCH) == NACL_x86
 #if NACL_BUILD_SUBARCH == 32
         movl 4(%esp), %ecx
 #elif !NACL_WINDOWS
         movl %edi, %ecx
 #endif
         /*
          * This is "xgetbv" (no operands).  Some assemblers don't know it yet.
          */
         .byte 0x0f, 0x01, 0xd0
 #if NACL_BUILD_SUBARCH == 64
         shlq $32, %rdx
         orq %rdx, %rax
 #endif
         ret
+#endif

[bsy, 2013/01/11 21:11:30]

isn't this a tad dangerous -- if this is not NACL_x86, we get a text label with
no code.  if any arm/mips code actually called this function, we'd get control
passed to whatever code happens to follow.  maybe use NACL_HALT in an else
clause, to be defensive?

[Roland McGrath, 2013/01/11 21:15:00]

In other cases we just put #error in an #else clause.
But this is an x86-only file and should not be compiled for other machines
anyway.  So I really wonder what the introduction of the #if was meant for.  It
seems suspect.

[bsy, 2013/01/11 21:19:53]

the problem is that this is getting built for all platforms.  the idea is that
the goal is that ncval would run on any platform and validate code targeting any
other platform.