[Click] Userlevel Click task schedule

[Beyers Cronje]

Hi Eddie,

Your suggested method of using a pipe works 100%, thanks!

Beyers

On 8/27/07, Eddie Kohler <kohler at cs.ucla.edu> wrote:
>
> Hi Beyers, again apologies for the delay!
>
> Without looking at the code, it's hard to tell for sure, but you should
> know that Click's core is not thread safe.  It is not surprising that a
> reschedule() in thread 2 (with Click running on thread 1) could get lost
> or even corrupt memory.
>
> I would implement this a little differently.  For example with a pipe.
> In the element's initialize() method, create a pipe with pipe() and make
> it nonblocking with fcntl(F_SETFL, O_NONBLOCK).  In Thread 2, when
> FromMyDevice receives a packet, it writes a byte to the pipe.  In Thread
> 1 (the main Click thread), add_select(_pipe[0], SELECT_READ).  The
> element's selected() method will read as many bytes from the pipe as
> possible and reschedule the task.  (Will explain further if necessary.)
>
> This is a lot of crap to go through to just reschedule a task, but
> absent a thread-safe core, it is unfortunately necessary.
>
> Eddie
>
>
> Beyers Cronje wrote:
> > Hi all,
> >
> > I have a question regarding userlevel Click task scheduling.
> >
> > I have the following click config:
> >
> > FromMyDevice -> Discard;
> >
> > FromMyDevice implements a separate dedicated thread to communicate with
> the
> > physical interface. This second thread receives packets from the
> physical
> > device and stores then in an internal queue and then calls reschedule()
> on
> > the task. FromMyDevice::run_task() then pulls the packets from the queue
> and
> > pushes them through the normal Click configuration. The problem is that
> > calling reschedule() from the second thread doesnt seem to work as the
> task
> > is never rescheduled. If I do away with reschedule and only make use of
> > fast_reschedule inside run_task() everything works 100%, but there's too
> > much wasted cpu cycles with this method.
> >
> > Does anyone have any idea how I can go about it?
> >
> > Thanks
> >
> > Beyers Cronje
> > _______________________________________________
> > click mailing list
> > click at amsterdam.lcs.mit.edu
> > https://amsterdam.lcs.mit.edu/mailman/listinfo/click
>