Skip to contentSkip to conversation

mm/hotplug: prevent memory leak when reuse pgdat

PATCH In review Wei Yang 3 participants 3 messages Aug 13, 2019

Subsystems (2)
lore
Share & tools
lore

Wei Yang

author
#0 lore raw

PATCH 1/1 mm/hotplug: prevent memory leak when reuse pgdat

When offline a node in try_offline_node, pgdat is not released. So that pgdat could be reused in hotadd_new_pgdat. While we re-allocate pgdat->per_cpu_nodestats if this pgdat is reused.

This patch prevents the memory leak by just allocate per_cpu_nodestats when it is a new pgdat.

NOTE: This is not tested since I didn't manage to create a case to offline a whole node. If my analysis is not correct, please let me know.

---
 mm/memory_hotplug.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
mm/memory_hotplug.c +9 -1
index c73f09913165..efaf9e6f580a 100644
@@ -933,8 +933,11 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 		if (!pgdat)
 			return NULL;
 
+		pgdat->per_cpu_nodestats =
+			alloc_percpu(struct per_cpu_nodestat);
 		arch_refresh_nodedata(nid, pgdat);
 	} else {
+		int cpu;
 		/*
 		 * Reset the nr_zones, order and classzone_idx before reuse.
 		 * Note that kswapd will init kswapd_classzone_idx properly
@@ -943,6 +946,12 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 		pgdat->nr_zones = 0;
 		pgdat->kswapd_order = 0;
 		pgdat->kswapd_classzone_idx = 0;
+		for_each_online_cpu(cpu) {
+			struct per_cpu_nodestat *p;
+
+			p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
+			memset(p, 0, sizeof(*p));
+		}
 	}
 
 	/* we can use NODE_DATA(nid) from here */
@@ -952,7 +961,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 
 	/* init node's zones as empty zones, we don't have any present pages.*/
 	free_area_init_core_hotplug(nid);
-	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
 
 	/*
 	 * The node we allocated has no zone fallback lists. For avoiding
Signature · 1 line
2.17.1
Acked-by on v1 1/1
#1 lore raw
Quoted 6 lines · Wei Yang

When offline a node in try_offline_node, pgdat is not released. So that pgdat could be reused in hotadd_new_pgdat. While we re-allocate pgdat->per_cpu_nodestats if this pgdat is reused.

This patch prevents the memory leak by just allocate per_cpu_nodestats when it is a new pgdat.

Yes this makes sense! I was slightly confused why we haven't initialized the allocated pcp area because __alloc_percpu does GFP_KERNEL without __GFP_ZERO but then I've just found out that the zeroying is done regardless. A bit unexpected...

Quoted 4 lines

NOTE: This is not tested since I didn't manage to create a case to offline a whole node. If my analysis is not correct, please let me know.

Thanks!

Quoted 44 lines
---
 mm/memory_hotplug.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
mm/memory_hotplug.c +9 -1
index c73f09913165..efaf9e6f580a 100644
@@ -933,8 +933,11 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 		if (!pgdat)
 			return NULL;
 
+		pgdat->per_cpu_nodestats =
+			alloc_percpu(struct per_cpu_nodestat);
 		arch_refresh_nodedata(nid, pgdat);
 	} else {
+		int cpu;
 		/*
 		 * Reset the nr_zones, order and classzone_idx before reuse.
 		 * Note that kswapd will init kswapd_classzone_idx properly
@@ -943,6 +946,12 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 		pgdat->nr_zones = 0;
 		pgdat->kswapd_order = 0;
 		pgdat->kswapd_classzone_idx = 0;
+		for_each_online_cpu(cpu) {
+			struct per_cpu_nodestat *p;
+
+			p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
+			memset(p, 0, sizeof(*p));
+		}
 	}
 
 	/* we can use NODE_DATA(nid) from here */
@@ -952,7 +961,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 
 	/* init node's zones as empty zones, we don't have any present pages.*/
 	free_area_init_core_hotplug(nid);
-	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
 
 	/*
 	 * The node we allocated has no zone fallback lists. For avoiding
Signature · 1 line
2.17.1
Signature · 2 lines
Michal Hocko
SUSE Labs
#2 lore raw
Quoted 13 lines · Michal Hocko
Quoted 6 lines · Wei Yang

When offline a node in try_offline_node, pgdat is not released. So that pgdat could be reused in hotadd_new_pgdat. While we re-allocate pgdat->per_cpu_nodestats if this pgdat is reused.

This patch prevents the memory leak by just allocate per_cpu_nodestats when it is a new pgdat.

Yes this makes sense! I was slightly confused why we haven't initialized the allocated pcp area because __alloc_percpu does GFP_KERNEL without __GFP_ZERO but then I've just found out that the zeroying is done regardless. A bit unexpected...

Quoted 9 lines
Quoted 4 lines

NOTE: This is not tested since I didn't manage to create a case to offline a whole node. If my analysis is not correct, please let me know.

Thanks!

Thanks :-)

Quoted 48 lines
Quoted 44 lines
---
 mm/memory_hotplug.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
mm/memory_hotplug.c +9 -1
index c73f09913165..efaf9e6f580a 100644
@@ -933,8 +933,11 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 		if (!pgdat)
 			return NULL;
 
+		pgdat->per_cpu_nodestats =
+			alloc_percpu(struct per_cpu_nodestat);
 		arch_refresh_nodedata(nid, pgdat);
 	} else {
+		int cpu;
 		/*
 		 * Reset the nr_zones, order and classzone_idx before reuse.
 		 * Note that kswapd will init kswapd_classzone_idx properly
@@ -943,6 +946,12 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 		pgdat->nr_zones = 0;
 		pgdat->kswapd_order = 0;
 		pgdat->kswapd_classzone_idx = 0;
+		for_each_online_cpu(cpu) {
+			struct per_cpu_nodestat *p;
+
+			p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
+			memset(p, 0, sizeof(*p));
+		}
 	}
 
 	/* we can use NODE_DATA(nid) from here */
@@ -952,7 +961,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 
 	/* init node's zones as empty zones, we don't have any present pages.*/
 	free_area_init_core_hotplug(nid);
-	pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
 
 	/*
 	 * The node we allocated has no zone fallback lists. For avoiding
Signature · 1 line
2.17.1
Signature · 2 lines
Michal Hocko
SUSE Labs
Signature · 2 lines
Wei Yang
Help you, Help me