A patch that would probably have fixed this issue on x86 systems seems to have been deferred because the current behaviour seemed to be necessary for sparc64. At least if I interpret these mails correctly: http://lkml.org/lkml/2008/4/4/86 http://lkml.org/lkml/2008/4/4/292
Attachment:
signature.asc
Description: This is a digitally signed message part.