Re: Pollable Semaphores

[Ulrich Drepper <drepper@gmail.com>]

On Fri, 21 Jan 2005 23:05:04 -0800, Chris Wright <chrisw@osdl.org> wrote:
> Yeah, here it is.  I refreshed it against a current kernel.  It passes my
> same old test, where I select on /proc/<pid>/status fd in exceptfds.

Looks certainly attractive to me.  Nice small patch.  How quickly
after the death of the process is proc_pid_flush() called?

If this could go in and the futex stuff is handled, there is "only"
async I/O to handle.  After that we could finally create a uniform
event mechanism at userlevel which binds all these events (I/O,
process/thread termination, sync primitives) together.  Maybe support
for legacy sync primitives (SysV semaphores, msg queues) is needed as
well, don't know yet.  Note that I assume that polling of POSIX
mqueues works as it did the last time I tried it.