On 8/11/22 08:53, Hyeonggon Yoo wrote:
>
> I started to wonder...
>
> Do we need slab_lock() after this patch for debugging caches?
>
> if SLUB never takes a slab from partial list (unless moving to another list or freeing it),
> and if you must acquire list_lock before accessing (to alloc/free objects) per-node partial list,
> Why do we need it?
Yeah, looks like we shouldn't need it anymore as n->list_lock will cover
everything now. Great observation!
> Of course, it is still needed for ARCHs that does not support
> cmpxchg_double().
>
> But as code for debugging caches is separated after this patch,
> maybe we can simply drop slab_lock() in:
> - alloc_debug_processing()
> - free_debug_processing()
> - alloc_single_from_{new_slab,partial}()
> - validate_slab()
list_slab_objects() too, with some care for the locking in get_map().
> And also relevant comments.
As it's becoming too large, I will split this to more patches. Thanks!
