Messages in this thread | | | Date | Mon, 24 Feb 2014 13:49:18 +0900 | From | Joonsoo Kim <> | Subject | Re: [PATCH 9/9] slab: remove a useless lockdep annotation |
| |
On Tue, Feb 18, 2014 at 10:21:10AM -0600, Christoph Lameter wrote: > On Mon, 17 Feb 2014, Joonsoo Kim wrote: > > > > Why change the BAD_ALIEN_MAGIC? > > > > Hello, Christoph. > > > > BAD_ALIEN_MAGIC is only checked by slab_set_lock_classes(). We remove this > > function in this patch, so returning BAD_ALIEN_MAGIC is useless. > > Its not useless. The point is if there is a pointer deref then we will see > this as a pointer value and know that it is realted to alien cache > processing. > > > And, in fact, BAD_ALIEN_MAGIC is already useless, because alloc_alien_cache() > > can't be called on !CONFIG_NUMA. This function is called if use_alien_caches > > is positive, but on !CONFIG_NUMA, use_alien_caches is always 0. So we don't > > have any chance to meet this BAD_ALIEN_MAGIC in runtime. > > Maybe it no longer serves a point. But note that caches may not be > populated because processors/nodes are not up yet.
Hello,
Let me clarify about alloc_alien_cache().
alloc_alien_cache() has two definitions, one for !CONFIG_NUMA, and the other for CONFIG_NUMA. BAD_ALIEN_MAGIC is only assigned on !CONFIG_NUMA definition. On CONFIG_NUMA, alloc_alien_cache() doesn't use BAD_ALIEN_MAGIC. So it is sufficient to consider just !CONFIG_NUMA case.
As I mentioned before, this function isn't called if use_alien_caches is zero and use_alien_caches is always zero on !CONFIG_NUMA. Therefore we cannot see BAD_ALIEN_MAGIC on any configuration. I don't know why BAD_ALIEN_MAGIC is introduced, however, it no longer serves a point, so it is better to remove it.
There are lots of code to check whether processor/nodes are up or not and these doesn't use BAD_ALIEN_MAGIC. Instead, it checks NULL on alien_cache of specific node. So removing BAD_ALIEN_MAGIC doesn't harm anything here.
Thanks.
| |