[Devel] [PATCH v5 0/2] fix static_key disabling problem in memcg

[Glauber Costa]

Hi, Tejun, Kame,

This series is composed of the two patches of the last fix, with no changes
(only exception is the removal of x = false assignments that Tejun requested,
that is done now). Note also that patch 1 of this series was reused by me
in the slab accounting patches for memcg.

The first patch, that adds a mutex to memcg is dropped. I didn't posted it
before so I could wait for Kame to get back from his vacations and properly
review it.

Kame: Steven Rostedt pointed out that our analysis of the static branch updates
were wrong, so the mutex is really not needed. 

The key to understand that, is that atomic_inc_not_zero will only return right
away if the value is not yet zero - as the name implies - but the update in the
atomic variable only happens after the code is patched.

Therefore, if two callers enters with a key value of zero, both will be held at
the jump_label_lock() call, effectively guaranteeing the behavior we need.

Glauber Costa (2):
  Always free struct memcg through schedule_work()
  decrement static keys on real destroy time

 include/net/sock.h        |    9 ++++++++
 mm/memcontrol.c           |   50 +++++++++++++++++++++++++++++++++-----------
 net/ipv4/tcp_memcontrol.c |   32 ++++++++++++++++++++++------
 3 files changed, 71 insertions(+), 20 deletions(-)

-- 
1.7.7.6