VDSO clock function may be miss-optimised leading to hangs on SMP systems

commit 1ddf0b1b11aa8a90cef6706e935fc31c75c406ba
Author: Andy Lutomirski <email address hidden>
Date: Sun Dec 21 08:57:46 2014 -0800

    x86, vdso: Use asm volatile in __getcpu

    In Linux 3.18 and below, GCC hoists the lsl instructions in the
    pvclock code all the way to the beginning of __vdso_clock_gettime,
    slowing the non-paravirt case significantly. For unknown reasons,
    presumably related to the removal of a branch, the performance issue
    is gone as of

    e76b027e6408 x86,vdso: Use LSL unconditionally for vgetcpu

    but I don't trust GCC enough to expect the problem to stay fixed.

    There should be no correctness issue, because the __getcpu calls in
    __vdso_vlock_gettime were never necessary in the first place.

    Note to stable maintainers: In 3.18 and below, depending on
    configuration, gcc 4.9.2 generates code like this:

         9c3: 44 0f 03 e8 lsl %ax,%r13d
         9c7: 45 89 eb mov %r13d,%r11d
         9ca: 0f 03 d8 lsl %ax,%ebx

    This patch won't apply as is to any released kernel, but I'll send a
    trivial backported version if needed.

    Fixes: 51c19b4f5927 x86: vdso: pvclock gettime support
    Cc: <email address hidden> # 3.8+
    Cc: Marcelo Tosatti <email address hidden>
    Acked-by: Paolo Bonzini <email address hidden>
    Signed-off-by: Andy Lutomirski <email address hidden>

===
break-fix: 51c19b4f5927f5a646e93d69f73c7e89ea14e737 1ddf0b1b11aa8a90cef6706e935fc31c75c406ba