From: Benjamin LaHaise Improve the performance of slab_put_obj(). Without the cast, gcc considers ptrdiff_t a 64 bit signed integer and ends up emitting code to use a full signed 128 bit divide on EM64T, which is substantially slower than a 32 bit unsigned divide. I noticed this when looking at the profile of a case where the slab balance is just on edge and thrashes back and forth freeing a block. Signed-off-by: Benjamin LaHaise Signed-off-by: Andrew Morton --- mm/slab.c | 6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-) diff -puN mm/slab.c~use-32-bit-division-in-slab_put_obj mm/slab.c --- devel/mm/slab.c~use-32-bit-division-in-slab_put_obj 2006-01-22 22:34:48.000000000 -0800 +++ devel-akpm/mm/slab.c 2006-01-22 22:38:19.000000000 -0800 @@ -1398,7 +1398,7 @@ static void check_poison_obj(kmem_cache_ struct slab *slabp = page_get_slab(virt_to_page(objp)); int objnr; - objnr = (objp - slabp->s_mem) / cachep->objsize; + objnr = (unsigned)(objp - slabp->s_mem) / cachep->objsize; if (objnr) { objp = slabp->s_mem + (objnr - 1) * cachep->objsize; realobj = (char *)objp + obj_dbghead(cachep); @@ -2341,7 +2341,7 @@ static void *cache_free_debugcheck(kmem_ if (cachep->flags & SLAB_STORE_USER) *dbg_userword(cachep, objp) = caller; - objnr = (objp - slabp->s_mem) / cachep->objsize; + objnr = (unsigned)(objp - slabp->s_mem) / cachep->objsize; BUG_ON(objnr >= cachep->num); BUG_ON(objp != slabp->s_mem + objnr * cachep->objsize); @@ -2699,7 +2699,7 @@ static void free_block(kmem_cache_t *cac slabp = page_get_slab(virt_to_page(objp)); l3 = cachep->nodelists[node]; list_del(&slabp->list); - objnr = (objp - slabp->s_mem) / cachep->objsize; + objnr = (unsigned)(objp - slabp->s_mem) / cachep->objsize; check_spinlock_acquired_node(cachep, node); check_slabp(cachep, slabp); _