cpuset,mm: update tasks' mems_allowed in time

Fix allocating page cache/slab object on the unallowed node when memory
spread is set by updating tasks' mems_allowed after its cpuset's mems is
changed.

In order to update tasks' mems_allowed in time, we must modify the code of
memory policy.  Because the memory policy is applied in the process's
context originally.  After applying this patch, one task directly
manipulates anothers mems_allowed, and we use alloc_lock in the
task_struct to protect mems_allowed and memory policy of the task.

But in the fast path, we didn't use lock to protect them, because adding a
lock may lead to performance regression.  But if we don't add a lock,the
task might see no nodes when changing cpuset's mems_allowed to some
non-overlapping set.  In order to avoid it, we set all new allowed nodes,
then clear newly disallowed ones.

[lee.schermerhorn@hp.com:
  The rework of mpol_new() to extract the adjusting of the node mask to
  apply cpuset and mpol flags "context" breaks set_mempolicy() and mbind()
  with MPOL_PREFERRED and a NULL nodemask--i.e., explicit local
  allocation.  Fix this by adding the check for MPOL_PREFERRED and empty
  node mask to mpol_new_mpolicy().

  Remove the now unneeded 'nodes = NULL' from mpol_new().

  Note that mpol_new_mempolicy() is always called with a non-NULL
  'nodes' parameter now that it has been removed from mpol_new().
  Therefore, we don't need to test nodes for NULL before testing it for
  'empty'.  However, just to be extra paranoid, add a VM_BUG_ON() to
  verify this assumption.]
[lee.schermerhorn@hp.com:

  I don't think the function name 'mpol_new_mempolicy' is descriptive
  enough to differentiate it from mpol_new().

  This function applies cpuset set context, usually constraining nodes
  to those allowed by the cpuset.  However, when the 'RELATIVE_NODES flag
  is set, it also translates the nodes.  So I settled on
  'mpol_set_nodemask()', because the comment block for mpol_new() mentions
  that we need to call this function to "set nodes".

  Some additional minor line length, whitespace and typo cleanup.]
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Paul Menage <menage@google.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 46 insertions, 138 deletions

diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index af5a83d..7e75a41 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -97,12 +97,6 @@ struct cpuset {
 
 	struct cpuset *parent;		/* my parent */
 
-	/*
-	 * Copy of global cpuset_mems_generation as of the most
-	 * recent time this cpuset changed its mems_allowed.
-	 */
-	int mems_generation;
-
 	struct fmeter fmeter;		/* memory_pressure filter */
 
 	/* partition number for rebuild_sched_domains() */
@@ -176,27 +170,6 @@ static inline int is_spread_slab(const struct cpuset *cs)
 	return test_bit(CS_SPREAD_SLAB, &cs->flags);
 }
 
-/*
- * Increment this integer everytime any cpuset changes its
- * mems_allowed value.  Users of cpusets can track this generation
- * number, and avoid having to lock and reload mems_allowed unless
- * the cpuset they're using changes generation.
- *
- * A single, global generation is needed because cpuset_attach_task() could
- * reattach a task to a different cpuset, which must not have its
- * generation numbers aliased with those of that tasks previous cpuset.
- *
- * Generations are needed for mems_allowed because one task cannot
- * modify another's memory placement.  So we must enable every task,
- * on every visit to __alloc_pages(), to efficiently check whether
- * its current->cpuset->mems_allowed has changed, requiring an update
- * of its current->mems_allowed.
- *
- * Since writes to cpuset_mems_generation are guarded by the cgroup lock
- * there is no need to mark it atomic.
- */
-static int cpuset_mems_generation;
-
 static struct cpuset top_cpuset = {
 	.flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
 };
@@ -228,8 +201,9 @@ static struct cpuset top_cpuset = {
  * If a task is only holding callback_mutex, then it has read-only
  * access to cpusets.
  *
- * The task_struct fields mems_allowed and mems_generation may only
- * be accessed in the context of that task, so require no locks.
+ * Now, the task_struct fields mems_allowed and mempolicy may be changed
+ * by other task, we use alloc_lock in the task_struct fields to protect
+ * them.
  *
  * The cpuset_common_file_read() handlers only hold callback_mutex across
  * small pieces of code, such as when reading out possibly multi-word
@@ -349,69 +323,6 @@ static void cpuset_update_task_spread_flag(struct cpuset *cs,
 		tsk->flags &= ~PF_SPREAD_SLAB;
 }
 
-/**
- * cpuset_update_task_memory_state - update task memory placement
- *
- * If the current tasks cpusets mems_allowed changed behind our
- * backs, update current->mems_allowed, mems_generation and task NUMA
- * mempolicy to the new value.
- *
- * Task mempolicy is updated by rebinding it relative to the
- * current->cpuset if a task has its memory placement changed.
- * Do not call this routine if in_interrupt().
- *
- * Call without callback_mutex or task_lock() held.  May be
- * called with or without cgroup_mutex held.  Thanks in part to
- * 'the_top_cpuset_hack', the task's cpuset pointer will never
- * be NULL.  This routine also might acquire callback_mutex during
- * call.
- *
- * Reading current->cpuset->mems_generation doesn't need task_lock
- * to guard the current->cpuset derefence, because it is guarded
- * from concurrent freeing of current->cpuset using RCU.
- *
- * The rcu_dereference() is technically probably not needed,
- * as I don't actually mind if I see a new cpuset pointer but
- * an old value of mems_generation.  However this really only
- * matters on alpha systems using cpusets heavily.  If I dropped
- * that rcu_dereference(), it would save them a memory barrier.
- * For all other arch's, rcu_dereference is a no-op anyway, and for
- * alpha systems not using cpusets, another planned optimization,
- * avoiding the rcu critical section for tasks in the root cpuset
- * which is statically allocated, so can't vanish, will make this
- * irrelevant.  Better to use RCU as intended, than to engage in
- * some cute trick to save a memory barrier that is impossible to
- * test, for alpha systems using cpusets heavily, which might not
- * even exist.
- *
- * This routine is needed to update the per-task mems_allowed data,
- * within the tasks context, when it is trying to allocate memory
- * (in various mm/mempolicy.c routines) and notices that some other
- * task has been modifying its cpuset.
- */
-
-void cpuset_update_task_memory_state(void)
-{
-	int my_cpusets_mem_gen;
-	struct task_struct *tsk = current;
-	struct cpuset *cs;
-
-	rcu_read_lock();
-	my_cpusets_mem_gen = task_cs(tsk)->mems_generation;
-	rcu_read_unlock();
-
-	if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
-		mutex_lock(&callback_mutex);
-		task_lock(tsk);
-		cs = task_cs(tsk); /* Maybe changed when task not locked */
-		guarantee_online_mems(cs, &tsk->mems_allowed);
-		tsk->cpuset_mems_generation = cs->mems_generation;
-		task_unlock(tsk);
-		mutex_unlock(&callback_mutex);
-		mpol_rebind_task(tsk, &tsk->mems_allowed);
-	}
-}
-
 /*
  * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
  *
@@ -1017,14 +928,6 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
  *    other task, the task_struct mems_allowed that we are hacking
  *    is for our current task, which must allocate new pages for that
  *    migrating memory region.
- *
- *    We call cpuset_update_task_memory_state() before hacking
- *    our tasks mems_allowed, so that we are assured of being in
- *    sync with our tasks cpuset, and in particular, callbacks to
- *    cpuset_update_task_memory_state() from nested page allocations
- *    won't see any mismatch of our cpuset and task mems_generation
- *    values, so won't overwrite our hacked tasks mems_allowed
- *    nodemask.
  */
 
 static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
@@ -1032,22 +935,37 @@ static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
 {
 	struct task_struct *tsk = current;
 
-	cpuset_update_task_memory_state();
-
-	mutex_lock(&callback_mutex);
 	tsk->mems_allowed = *to;
-	mutex_unlock(&callback_mutex);
 
 	do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
 
-	mutex_lock(&callback_mutex);
 	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
-	mutex_unlock(&callback_mutex);
 }
 
 /*
- * Rebind task's vmas to cpuset's new mems_allowed, and migrate pages to new
- * nodes if memory_migrate flag is set. Called with cgroup_mutex held.
+ * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy
+ * @tsk: the task to change
+ * @newmems: new nodes that the task will be set
+ *
+ * In order to avoid seeing no nodes if the old and new nodes are disjoint,
+ * we structure updates as setting all new allowed nodes, then clearing newly
+ * disallowed ones.
+ *
+ * Called with task's alloc_lock held
+ */
+static void cpuset_change_task_nodemask(struct task_struct *tsk,
+					nodemask_t *newmems)
+{
+	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
+	mpol_rebind_task(tsk, &tsk->mems_allowed);
+	mpol_rebind_task(tsk, newmems);
+	tsk->mems_allowed = *newmems;
+}
+
+/*
+ * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy
+ * of it to cpuset's new mems_allowed, and migrate pages to new nodes if
+ * memory_migrate flag is set. Called with cgroup_mutex held.
  */
 static void cpuset_change_nodemask(struct task_struct *p,
 				   struct cgroup_scanner *scan)
@@ -1056,12 +974,19 @@ static void cpuset_change_nodemask(struct task_struct *p,
 	struct cpuset *cs;
 	int migrate;
 	const nodemask_t *oldmem = scan->data;
+	nodemask_t newmems;
+
+	cs = cgroup_cs(scan->cg);
+	guarantee_online_mems(cs, &newmems);
+
+	task_lock(p);
+	cpuset_change_task_nodemask(p, &newmems);
+	task_unlock(p);
 
 	mm = get_task_mm(p);
 	if (!mm)
 		return;
 
-	cs = cgroup_cs(scan->cg);
 	migrate = is_memory_migrate(cs);
 
 	mpol_rebind_mm(mm, &cs->mems_allowed);
@@ -1114,10 +1039,10 @@ static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
 /*
  * Handle user request to change the 'mems' memory placement
  * of a cpuset.  Needs to validate the request, update the
- * cpusets mems_allowed and mems_generation, and for each
- * task in the cpuset, rebind any vma mempolicies and if
- * the cpuset is marked 'memory_migrate', migrate the tasks
- * pages to the new memory.
+ * cpusets mems_allowed, and for each task in the cpuset,
+ * update mems_allowed and rebind task's mempolicy and any vma
+ * mempolicies and if the cpuset is marked 'memory_migrate',
+ * migrate the tasks pages to the new memory.
  *
  * Call with cgroup_mutex held.  May take callback_mutex during call.
  * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
@@ -1170,7 +1095,6 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
 
 	mutex_lock(&callback_mutex);
 	cs->mems_allowed = trialcs->mems_allowed;
-	cs->mems_generation = cpuset_mems_generation++;
 	mutex_unlock(&callback_mutex);
 
 	update_tasks_nodemask(cs, &oldmem, &heap);
@@ -1434,15 +1358,18 @@ static void cpuset_attach(struct cgroup_subsys *ss,
 
 	if (cs == &top_cpuset) {
 		cpumask_copy(cpus_attach, cpu_possible_mask);
+		to = node_possible_map;
 	} else {
-		mutex_lock(&callback_mutex);
 		guarantee_online_cpus(cs, cpus_attach);
-		mutex_unlock(&callback_mutex);
+		guarantee_online_mems(cs, &to);
 	}
 	err = set_cpus_allowed_ptr(tsk, cpus_attach);
 	if (err)
 		return;
 
+	task_lock(tsk);
+	cpuset_change_task_nodemask(tsk, &to);
+	task_unlock(tsk);
 	cpuset_update_task_spread_flag(cs, tsk);
 
 	from = oldcs->mems_allowed;
@@ -1848,8 +1775,6 @@ static struct cgroup_subsys_state *cpuset_create(
 	struct cpuset *parent;
 
 	if (!cont->parent) {
-		/* This is early initialization for the top cgroup */
-		top_cpuset.mems_generation = cpuset_mems_generation++;
 		return &top_cpuset.css;
 	}
 	parent = cgroup_cs(cont->parent);
@@ -1861,7 +1786,6 @@ static struct cgroup_subsys_state *cpuset_create(
 		return ERR_PTR(-ENOMEM);
 	}
 
-	cpuset_update_task_memory_state();
 	cs->flags = 0;
 	if (is_spread_page(parent))
 		set_bit(CS_SPREAD_PAGE, &cs->flags);
@@ -1870,7 +1794,6 @@ static struct cgroup_subsys_state *cpuset_create(
 	set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
 	cpumask_clear(cs->cpus_allowed);
 	nodes_clear(cs->mems_allowed);
-	cs->mems_generation = cpuset_mems_generation++;
 	fmeter_init(&cs->fmeter);
 	cs->relax_domain_level = -1;
 
@@ -1889,8 +1812,6 @@ static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
 {
 	struct cpuset *cs = cgroup_cs(cont);
 
-	cpuset_update_task_memory_state();
-
 	if (is_sched_load_balance(cs))
 		update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
 
@@ -1911,21 +1832,6 @@ struct cgroup_subsys cpuset_subsys = {
 	.early_init = 1,
 };
 
-/*
- * cpuset_init_early - just enough so that the calls to
- * cpuset_update_task_memory_state() in early init code
- * are harmless.
- */
-
-int __init cpuset_init_early(void)
-{
-	alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_NOWAIT);
-
-	top_cpuset.mems_generation = cpuset_mems_generation++;
-	return 0;
-}
-
-
 /**
  * cpuset_init - initialize cpusets at system boot
  *
@@ -1936,11 +1842,13 @@ int __init cpuset_init(void)
 {
 	int err = 0;
 
+	if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL))
+		BUG();
+
 	cpumask_setall(top_cpuset.cpus_allowed);
 	nodes_setall(top_cpuset.mems_allowed);
 
 	fmeter_init(&top_cpuset.fmeter);
-	top_cpuset.mems_generation = cpuset_mems_generation++;
 	set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
 	top_cpuset.relax_domain_level = -1;