Linus Torvalds wrote:

> I'm convinced. However, I still want som elargely cosmetic changes to the 
> patch, Ingo:
> [...]

I'm not Ingo but changing the patch seemed easy enough to do, even for
me.  I append the patch below.  It is not meant for inclusion since
there is onw more problem to solve.  Ingo patch had another bug.  The
user_tid field in the child didn't get set if CLONE_CHILD_CLEARTID was
set.  This obviously has bad results.

I've changed the test to read

+       if (clone_flags & (CLONE_CHILD_SETTID|CLONE_CHILD_CLEARTID))

which works for me.  But since in schedule_tail() the code reads

+       if (current->user_tid)
+               put_user(current->pid, current->user_tid);

this enables writing the TID even if CLONE_CHILD_SETTID isn't set.  The
question is: how to access the clone flag information in the child?

Anyway, since this distinction isn't important for my tests I've applied
the attached patch and all runs fine.

-- 
--------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------