On Fri, 22 Mar 2013 20:29:18 +0100, Benjamin Herrenschmidt <b...@kernel.crashing.org> wrote: > On Fri, 2013-03-22 at 13:03 -0500, Nathan Fontenot wrote: > > We don't ever free old property values, mainly I assume since we don't keep > > reference counts and can't know when it is safe to do so. The problem I > > am starting to see on pseries is that we are getting very large properties. > > One of the biggest culprits is the property on pseries systems to describe > > the memory on the system in the device tree. These are big and getting > > bigger as memory increases, additionally this property is update every > > time memory is DLPAR added or removed from the system which can lead to > > leaving a bunch of memory that should be free'ed. > > > > Given that, is there (or has there been) any discussion on adding reference > > counts to properties in the device tree? With the myriad ways to get at > > the value of a property this may not be feasible but I would like to hear > > any thoughts from the community. > > My assumption was always that the lifetime of property values is tied > the the lifetime of the node they are in. IE, we wouldn't free a > property removed from a node but we could free all properties when > the node goes away... > > Not the best but would do...
As a middle way, we could modify the property in-place if the size of the new prop is <= the size of the old. Anyone foresee a problem with that approach? It could expose race conditions between reading and writing the new property value into the buffer and updating the length field. g. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
