On Sunday 14 May 2006 11:38, Johannes Stezenbach wrote: > On Sat, May 13, 2006, Oliver Endriss wrote: > > Trent Piepho wrote: > > > First, let me say that chained frontends *isn't* something new that is > > > mixed in with this attach/detach stuff. It's already there. See the > > > lnbp21 driver for an example. > > > > Is chaining used anywhere else than in the lnbp21 driver? > > > > I introduced it because I thought that it would be a good idea to limit > > the number of private pointers in struct dvb_frontend. ;-) > > > > Obviously it makes life harder, not easier. So just remove chaining from > > lnbp21.[ch], and rename misc_priv to something like sec_priv... > > > > Imho it's not worth the trouble. ;-) > > IMHO this "chaining" of the release function is a bad > concept. The card driver does the attach, so it should also > do the release. > > Consider: > foo = foo_attach(); > bar = bar_attach(foo); > ... > bar->release(); > // do we leak foo now is it magically cleaned up? > > Please keep it simple and stupid, the four drivers which > use lnbp21 can also keep a lnbp21 pointer around and > release it explicitly. Hmm, well we could make a rule that you never chain the release() function, even though you are allowed to do so for the other function pointers... would this be ok? Then we would always have a pointer to each used module, and hence using symbol_put_addr() would be fine. As for symbol_put_addr - well we've had absolutely zero response to the patches - both you and Trent have tried to get someone to look at 'em on LKML, and there's been complete silence... did we have a compat.h patch so that we can experiment using symbol_put_addr() in the meantime? If one of you wants to send it, I'll update the repos to use it and remove the chaining so we can see what that looks like. _______________________________________________ linux-dvb@xxxxxxxxxxx http://www.linuxtv.org/cgi-bin/mailman/listinfo/linux-dvb
