On Thu, Jul 28, 2011 at 12:07:53PM +0200, Frank van Maarseveen wrote:
/proc/slabinfo shows an anomaly on a 2.6.39.3 x86_64 system. After running F14 + kde4 for a while it shows high kmalloc-192 numbers:
# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>... kmalloc-192 735694 735723 264 31 2 : tunables 0 0 0 : slabdata 23733 23733 0
Digging in /sys/kernel/slab/kmalloc-192/alloc_calls one line stands out:
701960 nouveau_gart_manager_new+0x5d/0xf0 age=24/7405655/87972315 pid=1-4920 cpus=0-3
There are 709907 allocations total and according to /sys/kernel/slab/kmalloc-192/free_calls there are 52245 free calls total.
This seems to fix it:
diff -Nurp a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c --- a/drivers/gpu/drm/nouveau/nouveau_mem.c 2011-05-19 06:06:34.000000000 +0200 +++ b/drivers/gpu/drm/nouveau/nouveau_mem.c 2011-08-03 10:58:22.919205352 +0200 @@ -848,6 +848,7 @@ nouveau_gart_manager_del(struct ttm_mem_ nouveau_vm_unmap(&node->tmp_vma); nouveau_vm_put(&node->tmp_vma); } + kfree(node); mem->mm_node = NULL; }