kernel module event handler question

[YONETANI Tomokazu]

On Sat, Jun 17, 2006 at 03:18:11PM +0200, Attilio Rao wrote:
> 2006/6/17, YONETANI Tomokazu <qhwt+dfly at xxxxxxxxxx>:
> >Hello.
> >If the event handler in a kernel module returns an error upon registration,
> >isn't the module supposed to be removed from the kldstat's list?
> >
> >I wrote a simple kernel module(attached) which always returns EOPNOTSUPP
> >for MOD_LOAD event.  It adds two sysctl nodes debug.test and 
> >debug.test_int.
> >Reading from or writing to debug.test calls test_sysctl_proc() function.
> >When loaded, although module_register_init prints an error message, the
> >modules sits there, and I can even set different values to debug.test.
> >
> >Is this an expected behavior?  Do I need to do something special to
> >unregister my kernel module when something went wrong?
> 
> kldload(8) is a simply binary wrapper to kldload(2), which is supposed
> to invoke kernel linker in order to load (and locate in memory) kld
> symbols. It doesn't provide a real error-prone infrastructure, so it
> is expected you binary is loaded in memory but its consistency is not
> assured.

No, I'm not just talking about when the module is loaded by kldload(8),
but it can be loaded from the boot loader too.  While I understand your
point, I'm still believe that there's a room for improvement.

OK, here's another question: when a kernel file(*.ko) contains one or more
modules and none of them have positive reference count after the call to
linker_file_sysinit(), should that kernel file still remain loaded?
I also thought about making SYSINIT callback return a (recoverable) error
code, but that breaks driver API and it's something I want to avoid.