Re: [PATCH] Futex Generalization Patch

[Hubertus Franke]

On Wednesday 10 April 2002 03:59 pm, Bill Abt wrote:
> On 04/10/2002 at 02:47:50 PM AST, Hubertus Franke <[e-mail]>
>
> wrote:
> > The current interface is
> >
> > (A)
> > async wait:
> >    sys_futex (uaddr, FUTEX_AWAIT, value, (struct timespec*) sig);
> > upon signal handling
> >    sys_futex(uaddrs[], FUTEX_WAIT, size, NULL);
> >    to retrieve the uaddrs that got woken up...
>
> This is actually the preferred way.  I must've misinterpeted what you said
> earlier.  I believe this is actually a more generic way of handling.  A
> thread package can specify the signal to used much in the way the current
> LinuxThreads pre-allocates and uses certain real-time signals...
>
> > I am mainly concerned that SIGIO can be overloaded in a thread package ?
> > How would you know whether a SIGIO came from the futex or from other file
> >
> > handle.
>
> By keep with the original interface, we don't have to contend with this
> problem.  The thread package can use the signal that most suits its'
> implementation...
>
> Make sense?
>

I wasn't precise... 
There are ++ and -- for the file descriptors approach
++)
(a) it automatically cleans up the notify queue associated with the FD.
     [ notify queue could still be global, if we want that ]
     when the process disappears, no callback required in do_exit()
--) There is more overhead in the wakeup and the wait because I need to 
     move from the fd to the filestruct which is always some lookup and
     I have to verify that the file descriptor is actually a valid /dev/futex
     file (yikes).    

Another problem I see is that we only have one additional 
argument (void *utime) to piggyback the fd and the signal on. 

What could be done is the following.
Make utime a (void *utime) argument
In FUTEX_WAIT   interpret it as a pointer <struct timespec>
In FUTEX_AWAIT  inteprest it as { short fd, short sig };
There should be no limitation by casting it to shorts ?

Comments, anyone....

I suggest, we leave it as is right now until more comments come.

> Regards,
>       Bill Abt
>       Senior Software Engineer
>       Next Generation POSIX Threading for Linux
>       IBM Cambridge, MA, USA 02142
>       Ext: +(00)1 617-693-1591
>       T/L: 693-1591 (M/W/F)
>       T/L: 253-9938 (T/Th/Eves.)
>       Cell: +(00)1 617-803-7514
>       [e-mail] or [e-mail]
>       http://oss.software.ibm.com/developerworks/opensource/pthreads

-- 
-- Hubertus Franke  ([e-mail])