aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--drivers/char/sysrq.c2
-rw-r--r--include/linux/oom.h4
-rw-r--r--mm/oom_kill.c46
-rw-r--r--mm/page_alloc.c22
4 files changed, 53 insertions, 21 deletions
diff --git a/drivers/char/sysrq.c b/drivers/char/sysrq.c
index 44203ff599da..1ae2de7d8b4f 100644
--- a/drivers/char/sysrq.c
+++ b/drivers/char/sysrq.c
@@ -339,7 +339,7 @@ static struct sysrq_key_op sysrq_term_op = {
339 339
340static void moom_callback(struct work_struct *ignored) 340static void moom_callback(struct work_struct *ignored)
341{ 341{
342 out_of_memory(node_zonelist(0, GFP_KERNEL), GFP_KERNEL, 0); 342 out_of_memory(node_zonelist(0, GFP_KERNEL), GFP_KERNEL, 0, NULL);
343} 343}
344 344
345static DECLARE_WORK(moom_work, moom_callback); 345static DECLARE_WORK(moom_work, moom_callback);
diff --git a/include/linux/oom.h b/include/linux/oom.h
index 6aac5fe4f6f1..537662315627 100644
--- a/include/linux/oom.h
+++ b/include/linux/oom.h
@@ -10,6 +10,7 @@
10#ifdef __KERNEL__ 10#ifdef __KERNEL__
11 11
12#include <linux/types.h> 12#include <linux/types.h>
13#include <linux/nodemask.h>
13 14
14struct zonelist; 15struct zonelist;
15struct notifier_block; 16struct notifier_block;
@@ -26,7 +27,8 @@ enum oom_constraint {
26extern int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_flags); 27extern int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_flags);
27extern void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_flags); 28extern void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_flags);
28 29
29extern void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order); 30extern void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
31 int order, nodemask_t *mask);
30extern int register_oom_notifier(struct notifier_block *nb); 32extern int register_oom_notifier(struct notifier_block *nb);
31extern int unregister_oom_notifier(struct notifier_block *nb); 33extern int unregister_oom_notifier(struct notifier_block *nb);
32 34
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 6bb8a7a7ec9a..25c679e0288a 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -196,27 +196,46 @@ unsigned long badness(struct task_struct *p, unsigned long uptime)
196/* 196/*
197 * Determine the type of allocation constraint. 197 * Determine the type of allocation constraint.
198 */ 198 */
199static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
200 gfp_t gfp_mask)
201{
202#ifdef CONFIG_NUMA 199#ifdef CONFIG_NUMA
200static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
201 gfp_t gfp_mask, nodemask_t *nodemask)
202{
203 struct zone *zone; 203 struct zone *zone;
204 struct zoneref *z; 204 struct zoneref *z;
205 enum zone_type high_zoneidx = gfp_zone(gfp_mask); 205 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
206 nodemask_t nodes = node_states[N_HIGH_MEMORY];
207 206
208 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) 207 /*
209 if (cpuset_zone_allowed_softwall(zone, gfp_mask)) 208 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
210 node_clear(zone_to_nid(zone), nodes); 209 * to kill current.We have to random task kill in this case.
211 else 210 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
212 return CONSTRAINT_CPUSET; 211 */
212 if (gfp_mask & __GFP_THISNODE)
213 return CONSTRAINT_NONE;
213 214
214 if (!nodes_empty(nodes)) 215 /*
216 * The nodemask here is a nodemask passed to alloc_pages(). Now,
217 * cpuset doesn't use this nodemask for its hardwall/softwall/hierarchy
218 * feature. mempolicy is an only user of nodemask here.
219 * check mempolicy's nodemask contains all N_HIGH_MEMORY
220 */
221 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask))
215 return CONSTRAINT_MEMORY_POLICY; 222 return CONSTRAINT_MEMORY_POLICY;
216#endif 223
224 /* Check this allocation failure is caused by cpuset's wall function */
225 for_each_zone_zonelist_nodemask(zone, z, zonelist,
226 high_zoneidx, nodemask)
227 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
228 return CONSTRAINT_CPUSET;
217 229
218 return CONSTRAINT_NONE; 230 return CONSTRAINT_NONE;
219} 231}
232#else
233static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
234 gfp_t gfp_mask, nodemask_t *nodemask)
235{
236 return CONSTRAINT_NONE;
237}
238#endif
220 239
221/* 240/*
222 * Simple selection loop. We chose the process with the highest 241 * Simple selection loop. We chose the process with the highest
@@ -613,7 +632,8 @@ rest_and_return:
613 * OR try to be smart about which process to kill. Note that we 632 * OR try to be smart about which process to kill. Note that we
614 * don't have to be perfect here, we just have to be good. 633 * don't have to be perfect here, we just have to be good.
615 */ 634 */
616void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order) 635void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
636 int order, nodemask_t *nodemask)
617{ 637{
618 unsigned long freed = 0; 638 unsigned long freed = 0;
619 enum oom_constraint constraint; 639 enum oom_constraint constraint;
@@ -632,7 +652,7 @@ void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
632 * Check if there were limitations on the allocation (only relevant for 652 * Check if there were limitations on the allocation (only relevant for
633 * NUMA) that may require different handling. 653 * NUMA) that may require different handling.
634 */ 654 */
635 constraint = constrained_alloc(zonelist, gfp_mask); 655 constraint = constrained_alloc(zonelist, gfp_mask, nodemask);
636 read_lock(&tasklist_lock); 656 read_lock(&tasklist_lock);
637 657
638 switch (constraint) { 658 switch (constraint) {
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 59d2e88fb47c..850c4a7e2fe5 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1654,12 +1654,22 @@ __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
1654 if (page) 1654 if (page)
1655 goto out; 1655 goto out;
1656 1656
1657 /* The OOM killer will not help higher order allocs */ 1657 if (!(gfp_mask & __GFP_NOFAIL)) {
1658 if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_NOFAIL)) 1658 /* The OOM killer will not help higher order allocs */
1659 goto out; 1659 if (order > PAGE_ALLOC_COSTLY_ORDER)
1660 1660 goto out;
1661 /*
1662 * GFP_THISNODE contains __GFP_NORETRY and we never hit this.
1663 * Sanity check for bare calls of __GFP_THISNODE, not real OOM.
1664 * The caller should handle page allocation failure by itself if
1665 * it specifies __GFP_THISNODE.
1666 * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER.
1667 */
1668 if (gfp_mask & __GFP_THISNODE)
1669 goto out;
1670 }
1661 /* Exhausted what can be done so it's blamo time */ 1671 /* Exhausted what can be done so it's blamo time */
1662 out_of_memory(zonelist, gfp_mask, order); 1672 out_of_memory(zonelist, gfp_mask, order, nodemask);
1663 1673
1664out: 1674out:
1665 clear_zonelist_oom(zonelist, gfp_mask); 1675 clear_zonelist_oom(zonelist, gfp_mask);
@@ -3123,7 +3133,7 @@ static int __cpuinit process_zones(int cpu)
3123 3133
3124 if (percpu_pagelist_fraction) 3134 if (percpu_pagelist_fraction)
3125 setup_pagelist_highmark(zone_pcp(zone, cpu), 3135 setup_pagelist_highmark(zone_pcp(zone, cpu),
3126 (zone->present_pages / percpu_pagelist_fraction)); 3136 (zone->present_pages / percpu_pagelist_fraction));
3127 } 3137 }
3128 3138
3129 return 0; 3139 return 0;