Hi!

Running 3.1 x86_64 on top of 3.1 x86_64 works, but 2.6.38 x86_64 on
top of  3.1 x86_64 not?
Please confirm, so that I can try to produce the issue.

However I suspect we have to back port f1c2bb8 (um: implement a x86_64 vDSO).

Thanks,
//richard

So, you're using gcc 4.5. You have to apply the patch below to your UML kernel.
Anyway, consider using a recent UML kernel. Or at least one which an
active stable tree.

commit 534e3adbd22efa327e6ff27cf2d8ebaad8382ecd
Author: Richard Weinberger <richard <at> nod.at>
Date:   Wed Apr 27 15:26:54 2011 -0700

   um: adjust current_thread_info() for newer gcc versions

   In some cases gcc >= 4.5.2 will optimize away current_thread_info().  To
   prevent gcc from doing so the stack address has to be obtained via inline
   asm.

   Signed-off-by: Richard Weinberger <richard <at> nod.at>
   Acked-by: Kirill A. Shutemov <kirill <at> shutemov.name>
   Signed-off-by: Andrew Morton <akpm <at> linux-foundation.org>
   Signed-off-by: Linus Torvalds <torvalds <at> linux-foundation.org>

diff --git a/arch/um/include/asm/thread_info.h
b/arch/um/include/asm/thread_info.h
index e2cf786..5bd1bad 100644
--- a/arch/um/include/asm/thread_info.h
+++ b/arch/um/include/asm/thread_info.h
<at> <at> -49,7 +49,10 <at> <at> static inline struct thread_info *current_thread_info(void)
 {
       struct thread_info *ti;
       unsigned long mask = THREAD_SIZE - 1;
-       ti = (struct thread_info *) (((unsigned long) &ti) & ~mask);
+       void *p;
+
+       asm volatile ("" : "=r" (p) : "0" (&ti));
+       ti = (struct thread_info *) (((unsigned long)p) & ~mask);
       return ti;
 }

--
Thanks,
//richard