> Darren J Moffat wrote: > > Richard L. Hamilton wrote: > >>> Unfortunately, I'm not sure how well it works: an > >>> DEBUG kernel can try > >>> to unload the module at any time. > >>> > >> > >> But a loadable module can refuse to be unloaded > based on the return > >> from _fini(), > >> right? > > > > Yes but that is considered pretty rude, especially > if this module > > would otherwise be able to be unloaded and the only > reason for doing > > that is to avoid the modload vs 'modunload -i 0' > race condition. > > For drivers (you can't do this for misc modules) you > can use a special > property: ddi-no-autodetach. This property, if > non-zero, indicates that > modunload -i 0 will not try to detach instances of > your device in > response to modunload -i 0. I know of particular > devices where this is > important -- the audio drivers do it so that they > won't lose settings > just because nobody is using the device at the > precise time that > modunload -i 0 is called. > > Note that this properly still allows the device to be > properly unloaded > by a targetted modunload.
Why can that not be done for misc modules - can they not have properties, or what? Seems like if there ever is a reason for a misc module that's not loaded on demand, but only by explicit modload, that least surprise would be if it could request to also only be _unloaded_ explicitly (leaving refusing to unload for narrower purposes, like "busy", whatever that means for a module that seems to have few means of interacting with anything else). -- This message posted from opensolaris.org _______________________________________________ opensolaris-code mailing list opensolaris-code@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
