linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Baolin Wang <baolin.wang@linux.alibaba.com>
To: akpm@linux-foundation.org, ying.huang@intel.com,
	dave.hansen@linux.intel.com
Cc: ziy@nvidia.com, osalvador@suse.de, shy828301@gmail.com,
	baolin.wang@linux.alibaba.com, zhongjiang-ali@linux.alibaba.com,
	xlpang@linux.alibaba.com, linux-mm@kvack.org,
	linux-kernel@vger.kernel.org
Subject: [PATCH v2 2/2] mm: migrate: Allocate the node_demotion structure dynamically
Date: Thu, 11 Nov 2021 15:48:35 +0800	[thread overview]
Message-ID: <e39502af91e12ba1a4bef3be4d05b11b2c7a7a9f.1636616548.git.baolin.wang@linux.alibaba.com> (raw)
In-Reply-To: <cover.1636616548.git.baolin.wang@linux.alibaba.com>
In-Reply-To: <cover.1636616548.git.baolin.wang@linux.alibaba.com>

For the worst case (MAX_NUMNODES=1024), the node_demotion structure can
consume 32k bytes, which appears too large, so we can change to allocate
node_demotion dynamically at initialization time. Meanwhile allocating
the target demotion nodes array dynamically to select a suitable size
according to the MAX_NUMNODES.

Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 mm/migrate.c | 38 +++++++++++++++++++++++++++++---------
 1 file changed, 29 insertions(+), 9 deletions(-)

diff --git a/mm/migrate.c b/mm/migrate.c
index 126e9e6..0145b38 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -1152,10 +1152,11 @@ static int __unmap_and_move(struct page *page, struct page *newpage,
 #define DEFAULT_DEMOTION_TARGET_NODES 15
 struct demotion_nodes {
 	unsigned short nr;
-	short nodes[DEFAULT_DEMOTION_TARGET_NODES];
+	short nodes[];
 };
 
-static struct demotion_nodes node_demotion[MAX_NUMNODES] __read_mostly;
+static struct demotion_nodes *node_demotion[MAX_NUMNODES] __read_mostly;
+static unsigned short target_nodes_max;
 
 /**
  * next_demotion_node() - Get the next node in the demotion path
@@ -1168,10 +1169,13 @@ struct demotion_nodes {
  */
 int next_demotion_node(int node)
 {
-	struct demotion_nodes *nd = &node_demotion[node];
+	struct demotion_nodes *nd = node_demotion[node];
 	unsigned short target_nr, index;
 	int target;
 
+	if (!nd)
+		return NUMA_NO_NODE;
+
 	/*
 	 * node_demotion[] is updated without excluding this
 	 * function from running.  RCU doesn't provide any
@@ -3014,9 +3018,9 @@ static void __disable_all_migrate_targets(void)
 	int node, i;
 
 	for_each_online_node(node) {
-		node_demotion[node].nr = 0;
-		for (i = 0; i < DEFAULT_DEMOTION_TARGET_NODES; i++)
-			node_demotion[node].nodes[i] = NUMA_NO_NODE;
+		node_demotion[node]->nr = 0;
+		for (i = 0; i < target_nodes_max; i++)
+			node_demotion[node]->nodes[i] = NUMA_NO_NODE;
 	}
 }
 
@@ -3048,7 +3052,10 @@ static int establish_migrate_target(int node, nodemask_t *used,
 				    int best_distance)
 {
 	int migration_target, index, val;
-	struct demotion_nodes *nd = &node_demotion[node];
+	struct demotion_nodes *nd = node_demotion[node];
+
+	if (WARN_ONCE(!nd, "Can not set up migration path for node:%d\n", node))
+		return NUMA_NO_NODE;
 
 	migration_target = find_next_best_node(node, used);
 	if (migration_target == NUMA_NO_NODE)
@@ -3067,7 +3074,7 @@ static int establish_migrate_target(int node, nodemask_t *used,
 	}
 
 	index = nd->nr;
-	if (WARN_ONCE(index >= DEFAULT_DEMOTION_TARGET_NODES,
+	if (WARN_ONCE(index >= target_nodes_max,
 		      "Exceeds maximum demotion target nodes\n"))
 		return NUMA_NO_NODE;
 
@@ -3256,7 +3263,20 @@ static int migration_offline_cpu(unsigned int cpu)
 
 static int __init migrate_on_reclaim_init(void)
 {
-	int ret;
+	struct demotion_nodes *nd;
+	int ret, node;
+
+	/* Keep the maximum target demotion nodes are less than MAX_NUMNODES. */
+	target_nodes_max = min_t(unsigned short, DEFAULT_DEMOTION_TARGET_NODES,
+				 MAX_NUMNODES - 1);
+	for_each_node(node) {
+		nd = kmalloc(struct_size(nd, nodes, target_nodes_max),
+			     GFP_KERNEL);
+		if (!nd)
+			continue;
+
+		node_demotion[node] = nd;
+	}
 
 	ret = cpuhp_setup_state_nocalls(CPUHP_MM_DEMOTION_DEAD, "mm/demotion:offline",
 					NULL, migration_offline_cpu);
-- 
1.8.3.1



  parent reply	other threads:[~2021-11-11  7:48 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-11-11  7:48 [PATCH v2 0/2] Support multiple target nodes demotion Baolin Wang
2021-11-11  7:48 ` [PATCH v2 1/2] mm: migrate: " Baolin Wang
2021-11-11  8:20   ` Huang, Ying
2021-11-11 10:52     ` Baolin Wang
2021-11-11  7:48 ` Baolin Wang [this message]
2021-11-11  8:51   ` [PATCH v2 2/2] mm: migrate: Allocate the node_demotion structure dynamically Huang, Ying
2021-11-11 11:20     ` Baolin Wang
2021-11-11 23:39       ` Huang, Ying
2021-11-11 13:42   ` kernel test robot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=e39502af91e12ba1a4bef3be4d05b11b2c7a7a9f.1636616548.git.baolin.wang@linux.alibaba.com \
    --to=baolin.wang@linux.alibaba.com \
    --cc=akpm@linux-foundation.org \
    --cc=dave.hansen@linux.intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=osalvador@suse.de \
    --cc=shy828301@gmail.com \
    --cc=xlpang@linux.alibaba.com \
    --cc=ying.huang@intel.com \
    --cc=zhongjiang-ali@linux.alibaba.com \
    --cc=ziy@nvidia.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox