PERFORCE change 125520 for review

[Marko Zec]

On Monday 17 September 2007 20:03:59 Julian Elischer wrote:
> Marko Zec wrote:
> > http://perforce.freebsd.org/chv.cgi?CH=125520
> >
> > Change 125520 by zec at zec_tpx32 on 2007/08/21 23:51:39
> >
> > 	Given that ng_pipe nodes can be connected into arbitrary
> > 	topologies, and therefore it is possible for ngp_rcvdata()
> > 	to be recursively called from a single thread, it is
> > 	necessary to explicitly allow for the ng_pipe_giant mutex
> > 	to be recursively acquireable.
>
> OR use a different locking scheme.

That's right, but I'm just wondering is there anything fundamentally 
wrong with lock recursing (both in general as well as in this 
particular case)?

Marko

> i.e. reference counts or something.
>
> > Affected files ...
> >
> > .. //depot/projects/vimage/src/sys/netgraph/ng_pipe.c#2 edit
> >
> > Differences ...
> >
> > ==== //depot/projects/vimage/src/sys/netgraph/ng_pipe.c#2 (text+ko)
> > ====
> >
> > @@ -1028,7 +1028,7 @@
> >  			error = EEXIST;
> >  		else {
> >  			mtx_init(&ng_pipe_giant, "ng_pipe_giant", NULL,
> > -			    MTX_DEF);
> > +			    MTX_DEF | MTX_RECURSE);
> >  			LIST_INIT(&node_head);
> >  			LIST_INIT(&hook_head);
> >  			ds_handle = timeout((timeout_t *) &pipe_scheduler,