On Thu, 2004-10-14 at 20:13, Currie Reid wrote:
> Hello Philippe,
>
> The patch works, but raises one or two questions (sorry!). First, I
> understood what you meant about the disappearing handler crashing the box,
> so what I did in the root level domain is on module unloading I reset the
> handler to null -
>
> static void __exit my_cleanup( void ){
>
> adeos_catch_event( ADEOS_EXIT_PROCESS, 0 );
> adeos_unregister_domain( &domain_desc );
> }
>
> which works in the root domain. So I figured that this might be a proper
> clean-up procedure in higher priority domains. So I did the same thing in
> the module cleanup routine when registering a domain for the same event.
> When this module is unloaded, the handler does not default back to the
> root domain handler. Without the cleanup, it works as I would have
> expected, but if I reset the handler in the higher priority domain, it
> seems to wipe out the handler further down the pipe. Again, I don't know
> if this is intended behaviour, but I would have thought that setting the
> event handler in a higher priority domain to null would have no affect on
> lower priority (or root) domains. These lower priority domains would
> retain the handler attributed to them at the time of their entry.
>
First guess: maybe because you are using adeos_catch_event() to nullify
it, which implicitely does
adeos_catch_event_from(adp_current,event,NULL) [i.e. it's a shorthand] ,
and on behalf of a module cleanup routine, adp_current == adp_root, i.e.
Linux?
So, if you want to clean it up at the most prioritary level, you have to
write adeos_catch_event_from(&domain_desc,event,NULL), this way, the
current domain (Linux) won't be implicitely used for the operation.
However, unregistering a domain (now) properly releases all events for
it since the stage it occupies disappears, so you don't need to clear
the event handlers manually for this domain.
> Currie
>
>
> On Thu, 14 Oct 2004, Philippe Gerum wrote:
>
> > On Thu, 2004-10-14 at 15:41, Currie Reid wrote:
> > > Hello all,
> > >
> > > I am just trying to understand the underlying behaviour of adeos, and in
> > > my testing I have come across some behaviour that I want to verify is
> > > correct. If I set a handler for ADEOS_EXIT_PROCESS without registering a
> > > domain, then the handler is in the root domain and behaves as expected
> > > (thank you printk!). Next, I do a handler in a registered domain for
> > > ADEOS_EXIT_PROCESS and perform 2 tests - 1 where the handler doesn't
> > > propagate the event to the root domain, and one where it does. These both
> > > behave as expected - when propagated, I see the handler executing once for
> > > each domain. But when I unload the module and unregister the domain, the
> > > handler in the root domain no longer executes - I would have thought that
> > > the handler would still be in place.
> > >
> > > If I register 2 domains and duplicate the test, when the higher priority
> > > domain is unregistered, the lower priority domain starts handling the
> > > events again, as expected. I haven't delved into to code to see how easy
> > > this is to change - mostly because I'm not sure that this isn't
> > > uninitended behaviour.
> > >
> >
> > Actually, this is. I mean: a bug. Please try applying the following
> > patch to adeos/generic.c and let me know of the outcome. This said, keep
> > in mind that fixing this bug will make your box crash if you intend to
> > keep a handler attached which ends up disappearing from the address
> > space after the module is unloaded.
> > Thanks for the report.
> >
> > --- generic.c 4 Oct 2004 17:34:19 -0000 1.17
> > +++ generic.c 14 Oct 2004 16:24:53 -0000
> > @@ -173,7 +173,7 @@
> >
> > for (event = 0; event < ADEOS_NR_EVENTS; event++)
> > /* Need this to update the monitor count. */
> > - adeos_catch_event(event,NULL);
> > + adeos_catch_event_from(adp,event,NULL);
> >
> > #ifdef CONFIG_SMP
> > {
> > --
> >
> > Philippe.
> >
> >
>
>
> _______________________________________________
> Adeos-main mailing list
> [email protected]
> https://mail.gna.org/listinfo/adeos-main
--
Philippe.
