mm: make set_mempolicy(MPOL_INTERLEAV) N_HIGH_MEMORY aware

At first, init_task's mems_allowed is initialized as this.
init_task->mems_allowed == node_state[N_POSSIBLE]

And cpuset's top_cpuset mask is initialized as this
top_cpuset->mems_allowed = node_state[N_HIGH_MEMORY]

Before 2.6.29:
policy's mems_allowed is initialized as this.

1. update tasks->mems_allowed by its cpuset->mems_allowed.
2. policy->mems_allowed = nodes_and(tasks->mems_allowed, user's mask)

Updating task's mems_allowed in reference to top_cpuset's one.
cpuset's mems_allowed is aware of N_HIGH_MEMORY, always.

In 2.6.30: After commit 58568d2a8215cb6f55caf2332017d7bdff954e1c
("cpuset,mm: update tasks' mems_allowed in time"), policy's mems_allowed
is initialized as this.

1. policy->mems_allowd = nodes_and(task->mems_allowed, user's mask)

Here, if task is in top_cpuset, task->mems_allowed is not updated from
init's one. Assume user excutes command as #numactrl --interleave=all
,....

policy->mems_allowd = nodes_and(N_POSSIBLE, ALL_SET_MASK)

Then, policy's mems_allowd can includes a possible node, which has no pgdat.

MPOL's INTERLEAVE just scans nodemask of task->mems_allowd and access this
directly.

NODE_DATA(nid)->zonelist even if NODE_DATA(nid)==NULL

Then, what's we need is making policy->mems_allowed be aware of
N_HIGH_MEMORY. This patch does that. But to do so, extra nodemask will
be on statck. Because I know cpumask has a new interface of
CPUMASK_ALLOC(), I added it to node.

This patch stands on old behavior. But I feel this fix itself is just a
Band-Aid. But to do fundametal fix, we have to take care of memory
hotplug and it takes time. (task->mems_allowd should be N_HIGH_MEMORY, I
think.)

mpol_set_nodemask() should be aware of N_HIGH_MEMORY and policy's nodemask
should be includes only online nodes.

In old behavior, this is guaranteed by frequent reference to cpuset's
code. Now, most of them are removed and mempolicy has to check it by
itself.

To do check, a few nodemask_t will be used for calculating nodemask. But,
size of nodemask_t can be big and it's not good to allocate them on stack.

Now, cpumask_t has CPUMASK_ALLOC/FREE an easy code for get scratch area.
NODEMASK_ALLOC/FREE shoudl be there.

[[e-mail]: cleanups & tweaks]
Tested-by: KOSAKI Motohiro <[e-mail]>
Signed-off-by: KAMEZAWA Hiroyuki <[e-mail]>
Cc: Miao Xie <[e-mail]>
Cc: Ingo Molnar <[e-mail]>
Cc: Peter Zijlstra <[e-mail]>
Cc: Christoph Lameter <[e-mail]>
Cc: Paul Menage <[e-mail]>
Cc: Nick Piggin <[e-mail]>
Cc: Yasunori Goto <[e-mail]>
Cc: Pekka Enberg <[e-mail]>
Cc: David Rientjes <[e-mail]>
Cc: Lee Schermerhorn <[e-mail]>
Signed-off-by: Andrew Morton <[e-mail]>
Signed-off-by: Linus Torvalds <[e-mail]>

include/linux/nodemask.h

···
  *    to generate slightly worse code.  So use a simple one-line #define
  *    for node_isset(), instead of wrapping an inline inside a macro, the
  *    way we do the other calls.
+ *
+ * NODEMASK_SCRATCH
+ * When doing above logical AND, OR, XOR, Remap operations the callers tend to
+ * need temporary nodemask_t's on the stack. But if NODES_SHIFT is large,
+ * nodemask_t's consume too much stack space.  NODEMASK_SCRATCH is a helper
+ * for such situations. See below and CPUMASK_ALLOC also.
  */
 
 #include <linux/kernel.h>
···
 
 #define for_each_node(node)	   for_each_node_state(node, N_POSSIBLE)
 #define for_each_online_node(node) for_each_node_state(node, N_ONLINE)
+
+/*
+ * For nodemask scrach area.(See CPUMASK_ALLOC() in cpumask.h)
+ */
+
+#if NODES_SHIFT > 8 /* nodemask_t > 64 bytes */
+#define NODEMASK_ALLOC(x, m) struct x *m = kmalloc(sizeof(*m), GFP_KERNEL)
+#define NODEMASK_FREE(m) kfree(m)
+#else
+#define NODEMASK_ALLOC(x, m) struct x _m, *m = &_m
+#define NODEMASK_FREE(m)
+#endif
+
+/* A example struture for using NODEMASK_ALLOC, used in mempolicy. */
+struct nodemask_scratch {
+	nodemask_t	mask1;
+	nodemask_t	mask2;
+};
+
+#define NODEMASK_SCRATCH(x) NODEMASK_ALLOC(nodemask_scratch, x)
+#define NODEMASK_SCRATCH_FREE(x)  NODEMASK_FREE(x)
+
 
 #endif /* __LINUX_NODEMASK_H */

mm/mempolicy.c

···
  * Must be called holding task's alloc_lock to protect task's mems_allowed
  * and mempolicy.  May also be called holding the mmap_semaphore for write.
  */
-static int mpol_set_nodemask(struct mempolicy *pol, const nodemask_t *nodes)
+static int mpol_set_nodemask(struct mempolicy *pol,
+		     const nodemask_t *nodes, struct nodemask_scratch *nsc)
 {
-	nodemask_t cpuset_context_nmask;
 	int ret;
 
 	/* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
 	if (pol == NULL)
 		return 0;
+	/* Check N_HIGH_MEMORY */
+	nodes_and(nsc->mask1,
+		  cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]);
 
 	VM_BUG_ON(!nodes);
 	if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
 		nodes = NULL;	/* explicit local allocation */
 	else {
 		if (pol->flags & MPOL_F_RELATIVE_NODES)
-			mpol_relative_nodemask(&cpuset_context_nmask, nodes,
-					       &cpuset_current_mems_allowed);
+			mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1);
 		else
-			nodes_and(cpuset_context_nmask, *nodes,
-				  cpuset_current_mems_allowed);
+			nodes_and(nsc->mask2, *nodes, nsc->mask1);
+
 		if (mpol_store_user_nodemask(pol))
 			pol->w.user_nodemask = *nodes;
 		else
···
 						cpuset_current_mems_allowed;
 	}
 
-	ret = mpol_ops[pol->mode].create(pol,
-				nodes ? &cpuset_context_nmask : NULL);
+	if (nodes)
+		ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
+	else
+		ret = mpol_ops[pol->mode].create(pol, NULL);
 	return ret;
 }
 
···
 {
 	struct mempolicy *new, *old;
 	struct mm_struct *mm = current->mm;
+	NODEMASK_SCRATCH(scratch);
 	int ret;
 
-	new = mpol_new(mode, flags, nodes);
-	if (IS_ERR(new))
-		return PTR_ERR(new);
+	if (!scratch)
+		return -ENOMEM;
 
+	new = mpol_new(mode, flags, nodes);
+	if (IS_ERR(new)) {
+		ret = PTR_ERR(new);
+		goto out;
+	}
 	/*
 	 * prevent changing our mempolicy while show_numa_maps()
 	 * is using it.
···
 	if (mm)
 		down_write(&mm->mmap_sem);
 	task_lock(current);
-	ret = mpol_set_nodemask(new, nodes);
+	ret = mpol_set_nodemask(new, nodes, scratch);
 	if (ret) {
 		task_unlock(current);
 		if (mm)
 			up_write(&mm->mmap_sem);
 		mpol_put(new);
-		return ret;
+		goto out;
 	}
 	old = current->mempolicy;
 	current->mempolicy = new;
···
 		up_write(&mm->mmap_sem);
 
 	mpol_put(old);
-	return 0;
+	ret = 0;
+out:
+	NODEMASK_SCRATCH_FREE(scratch);
+	return ret;
 }
 
 /*
···
 		if (err)
 			return err;
 	}
-	down_write(&mm->mmap_sem);
-	task_lock(current);
-	err = mpol_set_nodemask(new, nmask);
-	task_unlock(current);
+	{
+		NODEMASK_SCRATCH(scratch);
+		if (scratch) {
+			down_write(&mm->mmap_sem);
+			task_lock(current);
+			err = mpol_set_nodemask(new, nmask, scratch);
+			task_unlock(current);
+			if (err)
+				up_write(&mm->mmap_sem);
+		} else
+			err = -ENOMEM;
+		NODEMASK_SCRATCH_FREE(scratch);
+	}
 	if (err) {
-		up_write(&mm->mmap_sem);
 		mpol_put(new);
 		return err;
 	}
···
  * Install non-NULL @mpol in inode's shared policy rb-tree.
  * On entry, the current task has a reference on a non-NULL @mpol.
  * This must be released on exit.
+ * This is called at get_inode() calls and we can use GFP_KERNEL.
  */
 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
 {
···
 	if (mpol) {
 		struct vm_area_struct pvma;
 		struct mempolicy *new;
+		NODEMASK_SCRATCH(scratch);
 
+		if (!scratch)
+			return;
 		/* contextualize the tmpfs mount point mempolicy */
 		new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
 		if (IS_ERR(new)) {
 			mpol_put(mpol);	/* drop our ref on sb mpol */
+			NODEMASK_SCRATCH_FREE(scratch);
 			return;		/* no valid nodemask intersection */
 		}
 
 		task_lock(current);
-		ret = mpol_set_nodemask(new, &mpol->w.user_nodemask);
+		ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
 		task_unlock(current);
 		mpol_put(mpol);	/* drop our ref on sb mpol */
 		if (ret) {
+			NODEMASK_SCRATCH_FREE(scratch);
 			mpol_put(new);
 			return;
 		}
···
 		pvma.vm_end = TASK_SIZE;	/* policy covers entire file */
 		mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
 		mpol_put(new);			/* drop initial ref */
+		NODEMASK_SCRATCH_FREE(scratch);
 	}
 }
 
···
 		err = 1;
 	else {
 		int ret;
-
-		task_lock(current);
-		ret = mpol_set_nodemask(new, &nodes);
-		task_unlock(current);
-		if (ret)
+		NODEMASK_SCRATCH(scratch);
+		if (scratch) {
+			task_lock(current);
+			ret = mpol_set_nodemask(new, &nodes, scratch);
+			task_unlock(current);
+		} else
+			ret = -ENOMEM;
+		NODEMASK_SCRATCH_FREE(scratch);
+		if (ret) {
 			err = 1;
-		else if (no_context) {
+			mpol_put(new);
+		} else if (no_context) {
 			/* save for contextualization */
 			new->w.user_nodemask = nodes;
 		}