On Sat, Mar 21, 2020 at 04:48:56PM -0700, Andrew Morton wrote:
On Tue, 3 Mar 2020 15:35:50 -0800 Roman Gushchin guro@fb.com wrote:
Depending on CONFIG_VMAP_STACK and the THREAD_SIZE / PAGE_SIZE ratio the space for task stacks can be allocated using __vmalloc_node_range(), alloc_pages_node() and kmem_cache_alloc_node(). In the first and the second cases page->mem_cgroup pointer is set, but in the third it's not: memcg membership of a slab page should be determined using the memcg_from_slab_page() function, which looks at page->slab_cache->memcg_params.memcg . In this case, using mod_memcg_page_state() (as in account_kernel_stack()) is incorrect: page->mem_cgroup pointer is NULL even for pages charged to a non-root memory cgroup.
It can lead to kernel_stack per-memcg counters permanently showing 0 on some architectures (depending on the configuration).
In order to fix it, let's introduce a mod_memcg_obj_state() helper, which takes a pointer to a kernel object as a first argument, uses mem_cgroup_from_obj() to get a RCU-protected memcg pointer and calls mod_memcg_state(). It allows to handle all possible configurations (CONFIG_VMAP_STACK and various THREAD_SIZE/PAGE_SIZE values) without spilling any memcg/kmem specifics into fork.c .
Note: this patch has been first posted as a part of the new slab controller patchset. This is a slightly updated version: the fixes tag has been added and the commit log was extended by the advice of Johannes Weiner. Because it's a fix that makes sense by itself, I'm re-posting it as a standalone patch.
Actually, it isn't a standalone patch.
It's true. I only meant it doesn't have to be a part of the slab accounting rework patchset.
--- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -776,6 +776,17 @@ void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) rcu_read_unlock(); } +void mod_memcg_obj_state(void *p, int idx, int val) +{
- struct mem_cgroup *memcg;
- rcu_read_lock();
- memcg = mem_cgroup_from_obj(p);
- if (memcg)
mod_memcg_state(memcg, idx, val);- rcu_read_unlock();
+}
mem_cgroup_from_obj() is later added by http://lkml.kernel.org/r/20200117203609.3146239-1-guro@fb.com
We could merge both mm-memcg-slab-introduce-mem_cgroup_from_obj.patch and this patch, but that's a whole lot of stuff to backport into -stable.
Are you able to come up with a simpler suitable-for-stable fix?
I'll try.
Thank you!