nvidia and rmap (again)
Quoted 3 lines · John
I don't quite know how to divine the right info out of slabinfo; I've attached it below.
If you search for a leak then you'd have to check the first number in each row: the number of active objects. Just look at the first row: it documents the fields in the following rows. If a number is huge and constantly increasing, then there is would be a leak. Typically the inode, dentry and buffer_head caches are large, the rest are small.
But: slab manages only small objects. I assume that nvidia allocates pages with alloc_pages() and then plays with the page flags. This is the layer below slab, you must look at /proc/meminfo to detect leaks.
--
Manfred