git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From 5cf1cacb49aee39c3e02ae87068fc3c6430659b0 Mon Sep 17 00:00:00 2001
   2 From: Tejun Heo <tj@kernel.org>
   3 Date: Thu, 21 Apr 2016 19:06:48 -0400
   4 Subject: cgroup, cpuset: replace cpuset_post_attach_flush() with cgroup_subsys->post_attach callback
   5
   6 From: Tejun Heo <tj@kernel.org>
   7
   8 commit 5cf1cacb49aee39c3e02ae87068fc3c6430659b0 upstream.
   9
  10 Since e93ad19d0564 ("cpuset: make mm migration asynchronous"), cpuset
  11 kicks off asynchronous NUMA node migration if necessary during task
  12 migration and flushes it from cpuset_post_attach_flush() which is
  13 called at the end of __cgroup_procs_write().  This is to avoid
  14 performing migration with cgroup_threadgroup_rwsem write-locked which
  15 can lead to deadlock through dependency on kworker creation.
  16
  17 memcg has a similar issue with charge moving, so let's convert it to
  18 an official callback rather than the current one-off cpuset specific
  19 function.  This patch adds cgroup_subsys->post_attach callback and
  20 makes cpuset register cpuset_post_attach_flush() as its ->post_attach.
  21
  22 The conversion is mostly one-to-one except that the new callback is
  23 called under cgroup_mutex.  This is to guarantee that no other
  24 migration operations are started before ->post_attach callbacks are
  25 finished.  cgroup_mutex is one of the outermost mutex in the system
  26 and has never been and shouldn't be a problem.  We can add specialized
  27 synchronization around __cgroup_procs_write() but I don't think
  28 there's any noticeable benefit.
  29
  30 Signed-off-by: Tejun Heo <tj@kernel.org>
  31 Cc: Li Zefan <lizefan@huawei.com>
  32 Cc: Johannes Weiner <hannes@cmpxchg.org>
  33 Cc: Michal Hocko <mhocko@kernel.org>
  34 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  35
  36 ---
  37  include/linux/cgroup-defs.h |    1 +
  38  include/linux/cpuset.h      |    6 ------
  39  kernel/cgroup.c             |    7 +++++--
  40  kernel/cpuset.c             |    4 ++--
  41  4 files changed, 8 insertions(+), 10 deletions(-)
  42
  43 --- a/include/linux/cgroup-defs.h
  44 +++ b/include/linux/cgroup-defs.h
  45 @@ -434,6 +434,7 @@ struct cgroup_subsys {
  46         int (*can_attach)(struct cgroup_taskset *tset);
  47         void (*cancel_attach)(struct cgroup_taskset *tset);
  48         void (*attach)(struct cgroup_taskset *tset);
  49 +       void (*post_attach)(void);
  50         int (*can_fork)(struct task_struct *task, void **priv_p);
  51         void (*cancel_fork)(struct task_struct *task, void *priv);
  52         void (*fork)(struct task_struct *task, void *priv);
  53 --- a/include/linux/cpuset.h
  54 +++ b/include/linux/cpuset.h
  55 @@ -137,8 +137,6 @@ static inline void set_mems_allowed(node
  56         task_unlock(current);
  57  }
  58
  59 -extern void cpuset_post_attach_flush(void);
  60 -
  61  #else /* !CONFIG_CPUSETS */
  62
  63  static inline bool cpusets_enabled(void) { return false; }
  64 @@ -245,10 +243,6 @@ static inline bool read_mems_allowed_ret
  65         return false;
  66  }
  67
  68 -static inline void cpuset_post_attach_flush(void)
  69 -{
  70 -}
  71 -
  72  #endif /* !CONFIG_CPUSETS */
  73
  74  #endif /* _LINUX_CPUSET_H */
  75 --- a/kernel/cgroup.c
  76 +++ b/kernel/cgroup.c
  77 @@ -2721,9 +2721,10 @@ static ssize_t __cgroup_procs_write(stru
  78                                     size_t nbytes, loff_t off, bool threadgroup)
  79  {
  80         struct task_struct *tsk;
  81 +       struct cgroup_subsys *ss;
  82         struct cgroup *cgrp;
  83         pid_t pid;
  84 -       int ret;
  85 +       int ssid, ret;
  86
  87         if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
  88                 return -EINVAL;
  89 @@ -2771,8 +2772,10 @@ out_unlock_rcu:
  90         rcu_read_unlock();
  91  out_unlock_threadgroup:
  92         percpu_up_write(&cgroup_threadgroup_rwsem);
  93 +       for_each_subsys(ss, ssid)
  94 +               if (ss->post_attach)
  95 +                       ss->post_attach();
  96         cgroup_kn_unlock(of->kn);
  97 -       cpuset_post_attach_flush();
  98         return ret ?: nbytes;
  99  }
 100
 101 --- a/kernel/cpuset.c
 102 +++ b/kernel/cpuset.c
 103 @@ -57,7 +57,6 @@
 104  #include <asm/uaccess.h>
 105  #include <linux/atomic.h>
 106  #include <linux/mutex.h>
 107 -#include <linux/workqueue.h>
 108  #include <linux/cgroup.h>
 109  #include <linux/wait.h>
 110
 111 @@ -1015,7 +1014,7 @@ static void cpuset_migrate_mm(struct mm_
 112         }
 113  }
 114
 115 -void cpuset_post_attach_flush(void)
 116 +static void cpuset_post_attach(void)
 117  {
 118         flush_workqueue(cpuset_migrate_mm_wq);
 119  }
 120 @@ -2083,6 +2082,7 @@ struct cgroup_subsys cpuset_cgrp_subsys
 121         .can_attach     = cpuset_can_attach,
 122         .cancel_attach  = cpuset_cancel_attach,
 123         .attach         = cpuset_attach,
 124 +       .post_attach    = cpuset_post_attach,
 125         .bind           = cpuset_bind,
 126         .legacy_cftypes = files,
 127         .early_init     = 1,