On Fri, Jun 17, 2016 at 10:40:14AM -0400, Oleg Drokin wrote: > >> The problem is: > >> > >> spin_lock(somelock); > >> structure = some_internal_list_lookup(list); > >> if (structure) > >> goto out; > >> > >> init_new_structure(new_structure); > >> mutex_init(&new_structure->s_mutex); > >> mutex_lock(&new_structure->s_mutex); // XXX CANNOT DO THIS UNDER SPINLOCK! > > > > mutex_trylock(&new_structure->s_mutex); > > > > should work, since you know it cannot be acquired yet by anybody else, > > since you've not published it yet. > > This does work, but suddenly does not look so obvious anymore, does it?
Well, the whole thing is somewhat tricky and would require a comment anyway. > I got some feedback that doing this is not really preferred. Much preferred to adding additional API I would think. Because if we add it here, we'll also have to add it to rt_mutex. Also, I'm not sure I want to promote this construct, it seems like a very special case. > Also once __must_check is added to mutex_try_lock() (surprised it's not yet), > we'll need to also have the useless "but what if it did fail to lock" path? The BUG_ON() suggested by Jeff seems like a good solution ;-)
