On Fri, 6 Jul 2012, Jiang Liu wrote: > On 07/05/2012 10:47 PM, Christoph Lameter wrote: > > On Tue, 3 Jul 2012, Jiang Liu wrote: > > > >> PG_slabobject: mark whether a (compound) page hosts SLUB/SLOB objects. > > > > Any subsystem may allocate a compound page to store metadata. > > > > The compound pages used by SLOB and SLUB are not managed in any way but > > the calls to kfree and kmalloc are converted to calls to the page > > allocator. There is no "management" by the slab allocators for these > > cases and its inaccurate to say that these are SLUB/SLOB objects since the > > allocators never deal with these objects. > > > Hi Chris, > I think there's a little difference with SLUB and SLOB for compound page. > For SLOB, it relies on the page allocator to allocate compound page to fulfill > request bigger than one page. For SLUB, it relies on the page allocator if the > request is bigger than two pages. So SLUB may allocate a 2-pages compound page > to host SLUB managed objects. > My proposal may be summarized as below: > 1) PG_slab flag marks a memory object is allocated from slab allocator. > 2) PG_slabobject marks a (compound) page hosts SLUB/SLOB managed objects. > 3) Only set PG_slab/PG_slabobject on the head page of compound pages. > 4) For SLAB, PG_slabobject is redundant and so not used. > > A summary of proposed usage of PG_slab(S) and PG_slabobject(O) with > SLAB/SLUB/SLOB allocators as below: > pagesize SLAB SLUB SLOB > 1page S S,O S,O > 2page S S,O S > >=4page S S S There is no point of recognizing such objects because those will be kmalloc objects and they can only be freed in a subsystem specific way. There is no standard way to even figure out which subsystem allocated them. So for all practical purposes those are unrecoverable. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>