@@ -755,6 +755,26 @@ static inline void set_next_region_in_freelist(struct free_list *free_list)
}
}
+static inline int is_mem_pwr_work_in_progress(struct mem_power_ctrl *mpc)
+{
+ if (mpc->work_status == MEM_PWR_WORK_IN_PROGRESS)
+ return 1;
+ return 0;
+}
+
+static inline void set_mem_pwr_work_in_progress(struct mem_power_ctrl *mpc)
+{
+ mpc->work_status = MEM_PWR_WORK_IN_PROGRESS;
+ smp_mb();
+}
+
+static inline void set_mem_pwr_work_complete(struct mem_power_ctrl *mpc)
+{
+ mpc->work_status = MEM_PWR_WORK_COMPLETE;
+ mpc->region = NULL;
+ smp_mb();
+}
+
#ifdef SECTION_IN_PAGE_FLAGS
static inline void set_page_section(struct page *page, unsigned long section)
{
@@ -17,6 +17,7 @@
#include <linux/pageblock-flags.h>
#include <linux/page-flags-layout.h>
#include <linux/atomic.h>
+#include <linux/workqueue.h>
#include <asm/page.h>
/* Free memory management - zoned buddy allocator. */
@@ -337,6 +338,24 @@ enum zone_type {
#ifndef __GENERATING_BOUNDS_H
+/*
+ * In order to evacuate a memory region, if the no. of pages to be migrated
+ * via compaction is more than this number, the effort is considered too
+ * costly and should be aborted.
+ */
+#define MAX_NR_MEM_PWR_MIGRATE_PAGES 32
+
+enum {
+ MEM_PWR_WORK_COMPLETE = 0,
+ MEM_PWR_WORK_IN_PROGRESS
+};
+
+struct mem_power_ctrl {
+ struct work_struct work;
+ struct zone_mem_region *region;
+ int work_status;
+};
+
struct zone_mem_region {
unsigned long start_pfn;
unsigned long end_pfn;
@@ -405,6 +424,8 @@ struct zone {
struct zone_mem_region zone_regions[MAX_NR_ZONE_REGIONS];
int nr_zone_regions;
+ struct mem_power_ctrl mem_power_ctrl;
+
#ifndef CONFIG_SPARSEMEM
/*
* Flags for a pageblock_nr_pages block. See pageblock-flags.h.
@@ -5002,6 +5002,35 @@ static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
#endif /* CONFIG_FLAT_NODE_MEM_MAP */
}
+static void mem_power_mgmt_fn(struct work_struct *work)
+{
+ struct mem_power_ctrl *mpc;
+ struct zone_mem_region *region;
+ unsigned long pages_in_use;
+ struct zone *zone;
+
+ mpc = container_of(work, struct mem_power_ctrl, work);
+
+ if (!mpc->region)
+ return; /* No work to do */
+
+ zone = container_of(mpc, struct zone, mem_power_ctrl);
+ region = mpc->region;
+
+ if (region == zone->zone_regions)
+ return; /* No point compacting region 0. */
+
+ pages_in_use = region->present_pages - region->nr_free;
+
+ if (pages_in_use > 0 &&
+ (pages_in_use <= MAX_NR_MEM_PWR_MIGRATE_PAGES)) {
+
+ evacuate_mem_region(zone, region);
+ }
+
+ set_mem_pwr_work_complete(mpc);
+}
+
static void __meminit init_node_memory_regions(struct pglist_data *pgdat)
{
int nid = pgdat->node_id;
@@ -5094,6 +5123,10 @@ static void __meminit init_zone_memory_regions(struct pglist_data *pgdat)
zone_init_free_lists_late(z);
+ INIT_WORK(&z->mem_power_ctrl.work, mem_power_mgmt_fn);
+ z->mem_power_ctrl.region = NULL;
+ set_mem_pwr_work_complete(&z->mem_power_ctrl);
+
/*
* Revisit the last visited node memory region, in case it
* spans multiple zones.
We are going to invoke the memory compaction algorithms for region-evacuation from worker threads, instead of dedicating a separate kthread to it. So add the worker infrastructure to perform this. In the worker, we calculate the cost of migration/compaction for a given region - if we need to migrate less than 32 pages, then we go ahead, else we deem the effort to be too costly and abort the compaction. Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com> --- include/linux/mm.h | 20 ++++++++++++++++++++ include/linux/mmzone.h | 21 +++++++++++++++++++++ mm/page_alloc.c | 33 +++++++++++++++++++++++++++++++++ 3 files changed, 74 insertions(+) -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html