OK, I have realised a problem with the previous attempts. Embedding the release() methods into dvb_unregister_frontend() causes problems with detaching. The first is - if we leave the release() calls in dvb_unregister_frontend(), it forces a card driver developer to either use ALL dvb_attach() or ALL static linkages for demod+sec+tuner modules. This might not be wanted. Next: I remembered the initialisation sequence of a frontend: 1) attach demod 2) attach tuner 3) attach SEC (if necessary) 4) register frontend Step 4 can fail. If it does, we have to undo steps 3,2, and 1. Therefore we required a dvb_detach() macro to allow this. Combine this with the first problem, and you realise you have to: a) Remove the releases from dvb_unregister_frontend(). b) dvb_unregister_dvb_attached_frontend() is pointless. c) And you have make the card drivers do the detach themselves so they can be flexible about what they do. This is exactly what js suggested: he saw this issue long before I did. So I've committed a new patch to my repos converting all the previous code to handle this. I wasn't sure whether you would prefer me to merge all the patches together like I have been doing, or if you would prefer a single patch illustrating this change. I've left it as a single patch - if you want me to regenerate a complete patchset instead just say so and I will sort it. I've made no attempt at supporting customising static builds yet; this issue was too important. Also I _think_ I detected a bug in the cx88/saa7134 drivers. If dvb_register_frontend() failed, it did not call the release() operations on the attached frontend objects. I've corrected this in the patch. If I'm mistaken, please let me know. (BTW: remember the patches in the repos will all be redone before/if any submission is done to mainline hg) _______________________________________________ linux-dvb@xxxxxxxxxxx http://www.linuxtv.org/cgi-bin/mailman/listinfo/linux-dvb
