Fix MIPS32 frame-pointer forcing for more recent GCC

["Joseph S. Myers" <joseph@codesourcery.com>]

The alloca call in FORCE_FRAME_POINTER in 
sysdeps/unix/sysv/linux/mips/mips32/sysdep.h is optimized away by GCC 4.7, 
resulting in failures of nptl/tst-cancel4, tst-cancel5, tst-cancelx4, 
tst-cancelx5.  I've applied this patch that fixes those failures by 
assigning the result of alloca to a volatile variable, so preventing it 
from being optimized away.

diff --git a/ChangeLog.mips b/ChangeLog.mips
index 853f286..9469b5c 100644
--- a/ChangeLog.mips
+++ b/ChangeLog.mips
@@ -1,3 +1,9 @@
+2012-06-14  Joseph Myers  <joseph@codesourcery.com>
+
+	* sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
+	(FORCE_FRAME_POINTER): Assign result of alloca to a volatile
+	variable.
+
 2012-06-01  Joseph Myers  <joseph@codesourcery.com>
 
 	* sysdeps/mips/mips64/n32/s_fma.c: New file.
diff --git a/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h b/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
index e51f329..8c024b0 100644
--- a/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
+++ b/sysdeps/unix/sysv/linux/mips/mips32/sysdep.h
@@ -194,8 +194,9 @@
 /* We need to use a frame pointer for the functions in which we
    adjust $sp around the syscall, or debug information and unwind
    information will be $sp relative and thus wrong during the syscall.  As
-   of GCC 3.4.3, this is sufficient.  */
-#define FORCE_FRAME_POINTER alloca (4)
+   of GCC 4.7, this is sufficient.  */
+#define FORCE_FRAME_POINTER						\
+  void *volatile __fp_force __attribute__ ((unused)) = alloca (4)
 
 #define internal_syscall5(ncs_init, cs_init, input, err, arg1, arg2, arg3, arg4, arg5)\
 ({									\

-- 
Joseph S. Myers
joseph@codesourcery.com