sched, cgroup: replace signal_struct->group_rwsem with a global percpu_rwsem

Note: This commit was originally committed as d59cfc09c32a but got
      reverted by 0c986253b939 due to the performance regression from
      the percpu_rwsem write down/up operations added to cgroup task
      migration path.  percpu_rwsem changes which alleviate the
      performance issue are pending for v4.4-rc1 merge window.
      Re-apply.

The cgroup side of threadgroup locking uses signal_struct->group_rwsem
to synchronize against threadgroup changes.  This per-process rwsem
adds small overhead to thread creation, exit and exec paths, forces
cgroup code paths to do lock-verify-unlock-retry dance in a couple
places and makes it impossible to atomically perform operations across
multiple processes.

This patch replaces signal_struct->group_rwsem with a global
percpu_rwsem cgroup_threadgroup_rwsem which is cheaper on the reader
side and contained in cgroups proper.  This patch converts one-to-one.

This does make writer side heavier and lower the granularity; however,
cgroup process migration is a fairly cold path, we do want to optimize
thread operations over it and cgroup migration operations don't take
enough time for the lower granularity to matter.

Signed-off-by: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/g/55F8097A.7000206@de.ibm.com
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>

2 files changed, 20 insertions, 61 deletions

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 2c9eae6ad970..115091efa889 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -46,6 +46,7 @@
 #include <linux/slab.h>
 #include <linux/spinlock.h>
 #include <linux/rwsem.h>
+#include <linux/percpu-rwsem.h>
 #include <linux/string.h>
 #include <linux/sort.h>
 #include <linux/kmod.h>
@@ -103,6 +104,8 @@ static DEFINE_SPINLOCK(cgroup_idr_lock);
  */
 static DEFINE_SPINLOCK(release_agent_path_lock);
 
+struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
+
 #define cgroup_assert_mutex_or_rcu_locked()				\
 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&			\
 			   !lockdep_is_held(&cgroup_mutex),		\
@@ -871,48 +874,6 @@ static struct css_set *find_css_set(struct css_set *old_cset,
 	return cset;
 }
 
-void cgroup_threadgroup_change_begin(struct task_struct *tsk)
-{
-	down_read(&tsk->signal->group_rwsem);
-}
-
-void cgroup_threadgroup_change_end(struct task_struct *tsk)
-{
-	up_read(&tsk->signal->group_rwsem);
-}
-
-/**
- * threadgroup_lock - lock threadgroup
- * @tsk: member task of the threadgroup to lock
- *
- * Lock the threadgroup @tsk belongs to.  No new task is allowed to enter
- * and member tasks aren't allowed to exit (as indicated by PF_EXITING) or
- * change ->group_leader/pid.  This is useful for cases where the threadgroup
- * needs to stay stable across blockable operations.
- *
- * fork and exit explicitly call threadgroup_change_{begin|end}() for
- * synchronization.  While held, no new task will be added to threadgroup
- * and no existing live task will have its PF_EXITING set.
- *
- * de_thread() does threadgroup_change_{begin|end}() when a non-leader
- * sub-thread becomes a new leader.
- */
-static void threadgroup_lock(struct task_struct *tsk)
-{
-	down_write(&tsk->signal->group_rwsem);
-}
-
-/**
- * threadgroup_unlock - unlock threadgroup
- * @tsk: member task of the threadgroup to unlock
- *
- * Reverse threadgroup_lock().
- */
-static inline void threadgroup_unlock(struct task_struct *tsk)
-{
-	up_write(&tsk->signal->group_rwsem);
-}
-
 static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
 {
 	struct cgroup *root_cgrp = kf_root->kn->priv;
@@ -2113,9 +2074,9 @@ static void cgroup_task_migrate(struct cgroup *old_cgrp,
 	lockdep_assert_held(&css_set_rwsem);
 
 	/*
-	 * We are synchronized through threadgroup_lock() against PF_EXITING
-	 * setting such that we can't race against cgroup_exit() changing the
-	 * css_set to init_css_set and dropping the old one.
+	 * We are synchronized through cgroup_threadgroup_rwsem against
+	 * PF_EXITING setting such that we can't race against cgroup_exit()
+	 * changing the css_set to init_css_set and dropping the old one.
 	 */
 	WARN_ON_ONCE(tsk->flags & PF_EXITING);
 	old_cset = task_css_set(tsk);
@@ -2172,10 +2133,11 @@ static void cgroup_migrate_finish(struct list_head *preloaded_csets)
  * @src_cset and add it to @preloaded_csets, which should later be cleaned
  * up by cgroup_migrate_finish().
  *
- * This function may be called without holding threadgroup_lock even if the
- * target is a process.  Threads may be created and destroyed but as long
- * as cgroup_mutex is not dropped, no new css_set can be put into play and
- * the preloaded css_sets are guaranteed to cover all migrations.
+ * This function may be called without holding cgroup_threadgroup_rwsem
+ * even if the target is a process.  Threads may be created and destroyed
+ * but as long as cgroup_mutex is not dropped, no new css_set can be put
+ * into play and the preloaded css_sets are guaranteed to cover all
+ * migrations.
  */
 static void cgroup_migrate_add_src(struct css_set *src_cset,
 				   struct cgroup *dst_cgrp,
@@ -2278,7 +2240,7 @@ err:
  * @threadgroup: whether @leader points to the whole process or a single task
  *
  * Migrate a process or task denoted by @leader to @cgrp.  If migrating a
- * process, the caller must be holding threadgroup_lock of @leader.  The
+ * process, the caller must be holding cgroup_threadgroup_rwsem.  The
  * caller is also responsible for invoking cgroup_migrate_add_src() and
  * cgroup_migrate_prepare_dst() on the targets before invoking this
  * function and following up with cgroup_migrate_finish().
@@ -2406,7 +2368,7 @@ out_release_tset:
  * @leader: the task or the leader of the threadgroup to be attached
  * @threadgroup: attach the whole threadgroup?
  *
- * Call holding cgroup_mutex and threadgroup_lock of @leader.
+ * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
  */
 static int cgroup_attach_task(struct cgroup *dst_cgrp,
 			      struct task_struct *leader, bool threadgroup)
@@ -2528,7 +2490,7 @@ retry_find_task:
 	get_task_struct(tsk);
 	rcu_read_unlock();
 
-	threadgroup_lock(tsk);
+	percpu_down_write(&cgroup_threadgroup_rwsem);
 	if (threadgroup) {
 		if (!thread_group_leader(tsk)) {
 			/*
@@ -2538,7 +2500,7 @@ retry_find_task:
 			 * try again; this is
 			 * "double-double-toil-and-trouble-check locking".
 			 */
-			threadgroup_unlock(tsk);
+			percpu_up_write(&cgroup_threadgroup_rwsem);
 			put_task_struct(tsk);
 			goto retry_find_task;
 		}
@@ -2548,7 +2510,7 @@ retry_find_task:
 	if (!ret)
 		ret = cgroup_attach_task(cgrp, tsk, threadgroup);
 
-	threadgroup_unlock(tsk);
+	percpu_up_write(&cgroup_threadgroup_rwsem);
 
 	put_task_struct(tsk);
 out_unlock_cgroup:
@@ -2751,17 +2713,17 @@ static int cgroup_update_dfl_csses(struct cgroup *cgrp)
 				goto out_finish;
 			last_task = task;
 
-			threadgroup_lock(task);
+			percpu_down_write(&cgroup_threadgroup_rwsem);
 			/* raced against de_thread() from another thread? */
 			if (!thread_group_leader(task)) {
-				threadgroup_unlock(task);
+				percpu_up_write(&cgroup_threadgroup_rwsem);
 				put_task_struct(task);
 				continue;
 			}
 
 			ret = cgroup_migrate(src_cset->dfl_cgrp, task, true);
 
-			threadgroup_unlock(task);
+			percpu_up_write(&cgroup_threadgroup_rwsem);
 			put_task_struct(task);
 
 			if (WARN(ret, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret))
@@ -5083,6 +5045,7 @@ int __init cgroup_init(void)
 	unsigned long key;
 	int ssid, err;
 
+	BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
 	BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
 	BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
 
diff --git a/kernel/fork.c b/kernel/fork.c
index 2845623fb582..7d5f0f118a63 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1149,10 +1149,6 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 	tty_audit_fork(sig);
 	sched_autogroup_fork(sig);
 
-#ifdef CONFIG_CGROUPS
-	init_rwsem(&sig->group_rwsem);
-#endif
-
 	sig->oom_score_adj = current->signal->oom_score_adj;
 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;