[PATCH] cpuset: hotunplug cpus and mems in all cpusets

The cpuset code handling hot unplug of CPUs or Memory Nodes was incorrect - it could remove a CPU or Node from the top cpuset, while leaving it still in some child cpusets. One basic rule of cpusets is that each cpusets cpus and mems are subsets of its parents. The cpuset hot unplug code violated this rule. So the cpuset hotunplug handler must walk down the tree, removing any removed CPU or Node from all cpusets. However, it is not allowed to make a cpusets cpus or mems become empty. They can only transition from empty to non-empty, not back. So if the last CPU or Node would be removed from a cpuset by the above walk, we scan back up the cpuset hierarchy, finding the nearest ancestor that still has something online, and copy its CPU or Memory placement. Signed-off-by: Paul Jackson <pj@sgi.com> Cc: Nathan Lynch <ntl@pobox.com> Cc: Anton Blanchard <anton@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] cpuset: hotunplug cpus and mems in all cpusets
The cpuset code handling hot unplug of CPUs or Memory Nodes was incorrect - it could remove a CPU or Node from the top cpuset, while leaving it still in some child cpusets. One basic rule of cpusets is that each cpusets cpus and mems are subsets of its parents. The cpuset hot unplug code violated this rule. So the cpuset hotunplug handler must walk down the tree, removing any removed CPU or Node from all cpusets. However, it is not allowed to make a cpusets cpus or mems become empty. They can only transition from empty to non-empty, not back. So if the last CPU or Node would be removed from a cpuset by the above walk, we scan back up the cpuset hierarchy, finding the nearest ancestor that still has something online, and copy its CPU or Memory placement. Signed-off-by: Paul Jackson <pj@sgi.com> Cc: Nathan Lynch <ntl@pobox.com> Cc: Anton Blanchard <anton@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Paul Jackson · Linus Torvalds
1 parent 38837fc75a
Showing 1 changed file with 70 additions and 17 deletions Side-by-side Diff
kernel/cpuset.c
@@ -2040,48 +2040,101 @@
 	return err;
 }
  
+#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_MEMORY_HOTPLUG)
 /*
- * The top_cpuset tracks what CPUs and Memory Nodes are online,
- * period.  This is necessary in order to make cpusets transparent
- * (of no affect) on systems that are actively using CPU hotplug
- * but making no active use of cpusets.
+ * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs
+ * or memory nodes, we need to walk over the cpuset hierarchy,
+ * removing that CPU or node from all cpusets.  If this removes the
+ * last CPU or node from a cpuset, then the guarantee_online_cpus()
+ * or guarantee_online_mems() code will use that emptied cpusets
+ * parent online CPUs or nodes.  Cpusets that were already empty of
+ * CPUs or nodes are left empty.
  *
- * This routine ensures that top_cpuset.cpus_allowed tracks
- * cpu_online_map on each CPU hotplug (cpuhp) event.
+ * This routine is intentionally inefficient in a couple of regards.
+ * It will check all cpusets in a subtree even if the top cpuset of
+ * the subtree has no offline CPUs or nodes.  It checks both CPUs and
+ * nodes, even though the caller could have been coded to know that
+ * only one of CPUs or nodes needed to be checked on a given call.
+ * This was done to minimize text size rather than cpu cycles.
+ *
+ * Call with both manage_mutex and callback_mutex held.
+ *
+ * Recursive, on depth of cpuset subtree.
  */
  
-#ifdef CONFIG_HOTPLUG_CPU
-static int cpuset_handle_cpuhp(struct notifier_block *nb,
-				unsigned long phase, void *cpu)
+static void guarantee_online_cpus_mems_in_subtree(const struct cpuset *cur)
 {
+	struct cpuset *c;
+
+	/* Each of our child cpusets mems must be online */
+	list_for_each_entry(c, &cur->children, sibling) {
+		guarantee_online_cpus_mems_in_subtree(c);
+		if (!cpus_empty(c->cpus_allowed))
+			guarantee_online_cpus(c, &c->cpus_allowed);
+		if (!nodes_empty(c->mems_allowed))
+			guarantee_online_mems(c, &c->mems_allowed);
+	}
+}
+
+/*
+ * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track
+ * cpu_online_map and node_online_map.  Force the top cpuset to track
+ * whats online after any CPU or memory node hotplug or unplug event.
+ *
+ * To ensure that we don't remove a CPU or node from the top cpuset
+ * that is currently in use by a child cpuset (which would violate
+ * the rule that cpusets must be subsets of their parent), we first
+ * call the recursive routine guarantee_online_cpus_mems_in_subtree().
+ *
+ * Since there are two callers of this routine, one for CPU hotplug
+ * events and one for memory node hotplug events, we could have coded
+ * two separate routines here.  We code it as a single common routine
+ * in order to minimize text size.
+ */
+
+static void common_cpu_mem_hotplug_unplug(void)
+{
 	mutex_lock(&manage_mutex);
 	mutex_lock(&callback_mutex);
  
+	guarantee_online_cpus_mems_in_subtree(&top_cpuset);
 	top_cpuset.cpus_allowed = cpu_online_map;
+	top_cpuset.mems_allowed = node_online_map;
  
 	mutex_unlock(&callback_mutex);
 	mutex_unlock(&manage_mutex);
+}
+#endif
  
+#ifdef CONFIG_HOTPLUG_CPU
+/*
+ * The top_cpuset tracks what CPUs and Memory Nodes are online,
+ * period.  This is necessary in order to make cpusets transparent
+ * (of no affect) on systems that are actively using CPU hotplug
+ * but making no active use of cpusets.
+ *
+ * This routine ensures that top_cpuset.cpus_allowed tracks
+ * cpu_online_map on each CPU hotplug (cpuhp) event.
+ */
+
+static int cpuset_handle_cpuhp(struct notifier_block *nb,
+				unsigned long phase, void *cpu)
+{
+	common_cpu_mem_hotplug_unplug();
 	return 0;
 }
 #endif
  
+#ifdef CONFIG_MEMORY_HOTPLUG
 /*
  * Keep top_cpuset.mems_allowed tracking node_online_map.
  * Call this routine anytime after you change node_online_map.
  * See also the previous routine cpuset_handle_cpuhp().
  */
  
-#ifdef CONFIG_MEMORY_HOTPLUG
 void cpuset_track_online_nodes()
 {
-	mutex_lock(&manage_mutex);
-	mutex_lock(&callback_mutex);
-
-	top_cpuset.mems_allowed = node_online_map;
-
-	mutex_unlock(&callback_mutex);
-	mutex_unlock(&manage_mutex);
+	common_cpu_mem_hotplug_unplug();
 }
 #endif
...	...	@@ -2040,48 +2040,101 @@
2040	2040	return err;
2041	2041	}
2042	2042
	2043	+#if defined(CONFIG_HOTPLUG_CPU) \|\| defined(CONFIG_MEMORY_HOTPLUG)
2043	2044	/*
2044		- * The top_cpuset tracks what CPUs and Memory Nodes are online,
2045		- * period. This is necessary in order to make cpusets transparent
2046		- * (of no affect) on systems that are actively using CPU hotplug
2047		- * but making no active use of cpusets.
	2045	+ * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs
	2046	+ * or memory nodes, we need to walk over the cpuset hierarchy,
	2047	+ * removing that CPU or node from all cpusets. If this removes the
	2048	+ * last CPU or node from a cpuset, then the guarantee_online_cpus()
	2049	+ * or guarantee_online_mems() code will use that emptied cpusets
	2050	+ * parent online CPUs or nodes. Cpusets that were already empty of
	2051	+ * CPUs or nodes are left empty.
2048	2052	*
2049		- * This routine ensures that top_cpuset.cpus_allowed tracks
2050		- * cpu_online_map on each CPU hotplug (cpuhp) event.
	2053	+ * This routine is intentionally inefficient in a couple of regards.
	2054	+ * It will check all cpusets in a subtree even if the top cpuset of
	2055	+ * the subtree has no offline CPUs or nodes. It checks both CPUs and
	2056	+ * nodes, even though the caller could have been coded to know that
	2057	+ * only one of CPUs or nodes needed to be checked on a given call.
	2058	+ * This was done to minimize text size rather than cpu cycles.
	2059	+ *
	2060	+ * Call with both manage_mutex and callback_mutex held.
	2061	+ *
	2062	+ * Recursive, on depth of cpuset subtree.
2051	2063	*/
2052	2064
2053		-#ifdef CONFIG_HOTPLUG_CPU
2054		-static int cpuset_handle_cpuhp(struct notifier_block *nb,
2055		- unsigned long phase, void *cpu)
	2065	+static void guarantee_online_cpus_mems_in_subtree(const struct cpuset *cur)
2056	2066	{
	2067	+ struct cpuset *c;
	2068	+
	2069	+ /* Each of our child cpusets mems must be online */
	2070	+ list_for_each_entry(c, &cur->children, sibling) {
	2071	+ guarantee_online_cpus_mems_in_subtree(c);
	2072	+ if (!cpus_empty(c->cpus_allowed))
	2073	+ guarantee_online_cpus(c, &c->cpus_allowed);
	2074	+ if (!nodes_empty(c->mems_allowed))
	2075	+ guarantee_online_mems(c, &c->mems_allowed);
	2076	+ }
	2077	+}
	2078	+
	2079	+/*
	2080	+ * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track
	2081	+ * cpu_online_map and node_online_map. Force the top cpuset to track
	2082	+ * whats online after any CPU or memory node hotplug or unplug event.
	2083	+ *
	2084	+ * To ensure that we don't remove a CPU or node from the top cpuset
	2085	+ * that is currently in use by a child cpuset (which would violate
	2086	+ * the rule that cpusets must be subsets of their parent), we first
	2087	+ * call the recursive routine guarantee_online_cpus_mems_in_subtree().
	2088	+ *
	2089	+ * Since there are two callers of this routine, one for CPU hotplug
	2090	+ * events and one for memory node hotplug events, we could have coded
	2091	+ * two separate routines here. We code it as a single common routine
	2092	+ * in order to minimize text size.
	2093	+ */
	2094	+
	2095	+static void common_cpu_mem_hotplug_unplug(void)
	2096	+{
2057	2097	mutex_lock(&manage_mutex);
2058	2098	mutex_lock(&callback_mutex);
2059	2099
	2100	+ guarantee_online_cpus_mems_in_subtree(&top_cpuset);
2060	2101	top_cpuset.cpus_allowed = cpu_online_map;
	2102	+ top_cpuset.mems_allowed = node_online_map;
2061	2103
2062	2104	mutex_unlock(&callback_mutex);
2063	2105	mutex_unlock(&manage_mutex);
	2106	+}
	2107	+#endif
2064	2108
	2109	+#ifdef CONFIG_HOTPLUG_CPU
	2110	+/*
	2111	+ * The top_cpuset tracks what CPUs and Memory Nodes are online,
	2112	+ * period. This is necessary in order to make cpusets transparent
	2113	+ * (of no affect) on systems that are actively using CPU hotplug
	2114	+ * but making no active use of cpusets.
	2115	+ *
	2116	+ * This routine ensures that top_cpuset.cpus_allowed tracks
	2117	+ * cpu_online_map on each CPU hotplug (cpuhp) event.
	2118	+ */
	2119	+
	2120	+static int cpuset_handle_cpuhp(struct notifier_block *nb,
	2121	+ unsigned long phase, void *cpu)
	2122	+{
	2123	+ common_cpu_mem_hotplug_unplug();
2065	2124	return 0;
2066	2125	}
2067	2126	#endif
2068	2127
	2128	+#ifdef CONFIG_MEMORY_HOTPLUG
2069	2129	/*
2070	2130	* Keep top_cpuset.mems_allowed tracking node_online_map.
2071	2131	* Call this routine anytime after you change node_online_map.
2072	2132	* See also the previous routine cpuset_handle_cpuhp().
2073	2133	*/
2074	2134
2075		-#ifdef CONFIG_MEMORY_HOTPLUG
2076	2135	void cpuset_track_online_nodes()
2077	2136	{
2078		- mutex_lock(&manage_mutex);
2079		- mutex_lock(&callback_mutex);
2080		-
2081		- top_cpuset.mems_allowed = node_online_map;
2082		-
2083		- mutex_unlock(&callback_mutex);
2084		- mutex_unlock(&manage_mutex);
	2137	+ common_cpu_mem_hotplug_unplug();
2085	2138	}
2086	2139	#endif
2087	2140