[PATCH] Reduce size of huge boot per_cpu_pageset
authorChristoph Lameter <christoph@graphe.net>
Wed, 22 Jun 2005 00:15:00 +0000 (17:15 -0700)
committerLinus Torvalds <torvalds@ppc970.osdl.org>
Wed, 22 Jun 2005 01:46:18 +0000 (18:46 -0700)
Reduce size of the huge per_cpu_pageset structure in __initdata introduced
into mm1 with the pageset localization patchset.  Use one specially
configured pageset per cpu for all zones and nodes during bootup.

- Avoid duplication of pageset initialization code.
- do the adding to the pageset list before potential free_pages_bulk
  in free_hot_cold_page (otherwise we would have to hold a page
  in a pageset during the period that the boot pagesets are in use).
- remove mistaken __cpuinitdata attribute and revert back to __initdata
  for the boot pageset. A boot pageset is not necessary for cpu hotplug.

Tested for UP SMP NUMA on x86_64 (2.6.12-rc6-mm1): UP SMP NUMA Tested on
IA64 (2.6.12-rc5-mm2): NUMA (2.6.12-rc6-mm1 broken for IA64 because of
sparsemem patches)

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
mm/page_alloc.c

index a95e72d7f9452e28a6ea77b34f8a483a321392e4..418102a02921aa74016a90aaac50f7d933a117a0 100644 (file)
@@ -71,11 +71,6 @@ EXPORT_SYMBOL(nr_swap_pages);
 struct zone *zone_table[1 << (ZONES_SHIFT + NODES_SHIFT)];
 EXPORT_SYMBOL(zone_table);
 
-#ifdef CONFIG_NUMA
-static struct per_cpu_pageset
-       pageset_table[MAX_NR_ZONES*MAX_NUMNODES*NR_CPUS] __initdata;
-#endif
-
 static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
 int min_free_kbytes = 1024;
 
@@ -652,10 +647,10 @@ static void fastcall free_hot_cold_page(struct page *page, int cold)
        free_pages_check(__FUNCTION__, page);
        pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
        local_irq_save(flags);
-       if (pcp->count >= pcp->high)
-               pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
        list_add(&page->lru, &pcp->list);
        pcp->count++;
+       if (pcp->count >= pcp->high)
+               pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
        local_irq_restore(flags);
        put_cpu();
 }
@@ -1714,57 +1709,55 @@ static int __devinit zone_batchsize(struct zone *zone)
        return batch;
 }
 
+inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
+{
+       struct per_cpu_pages *pcp;
+
+       pcp = &p->pcp[0];               /* hot */
+       pcp->count = 0;
+       pcp->low = 2 * batch;
+       pcp->high = 6 * batch;
+       pcp->batch = max(1UL, 1 * batch);
+       INIT_LIST_HEAD(&pcp->list);
+
+       pcp = &p->pcp[1];               /* cold*/
+       pcp->count = 0;
+       pcp->low = 0;
+       pcp->high = 2 * batch;
+       pcp->batch = max(1UL, 1 * batch);
+       INIT_LIST_HEAD(&pcp->list);
+}
+
 #ifdef CONFIG_NUMA
 /*
- * Dynamicaly allocate memory for the
+ * Boot pageset table. One per cpu which is going to be used for all
+ * zones and all nodes. The parameters will be set in such a way
+ * that an item put on a list will immediately be handed over to
+ * the buddy list. This is safe since pageset manipulation is done
+ * with interrupts disabled.
+ *
+ * Some NUMA counter updates may also be caught by the boot pagesets.
+ * These will be discarded when bootup is complete.
+ */
+static struct per_cpu_pageset
+       boot_pageset[NR_CPUS] __initdata;
+
+/*
+ * Dynamically allocate memory for the
  * per cpu pageset array in struct zone.
  */
 static int __devinit process_zones(int cpu)
 {
        struct zone *zone, *dzone;
-       int i;
 
        for_each_zone(zone) {
-               struct per_cpu_pageset *npageset = NULL;
 
-               npageset = kmalloc_node(sizeof(struct per_cpu_pageset),
+               zone->pageset[cpu] = kmalloc_node(sizeof(struct per_cpu_pageset),
                                         GFP_KERNEL, cpu_to_node(cpu));
-               if (!npageset) {
-                       zone->pageset[cpu] = NULL;
+               if (!zone->pageset[cpu])
                        goto bad;
-               }
 
-               if (zone->pageset[cpu]) {
-                       memcpy(npageset, zone->pageset[cpu],
-                                       sizeof(struct per_cpu_pageset));
-
-                       /* Relocate lists */
-                       for (i = 0; i < 2; i++) {
-                               INIT_LIST_HEAD(&npageset->pcp[i].list);
-                               list_splice(&zone->pageset[cpu]->pcp[i].list,
-                                       &npageset->pcp[i].list);
-                       }
-               } else {
-                       struct per_cpu_pages *pcp;
-                       unsigned long batch;
-
-                       batch = zone_batchsize(zone);
-
-                       pcp = &npageset->pcp[0];                /* hot */
-                       pcp->count = 0;
-                       pcp->low = 2 * batch;
-                       pcp->high = 6 * batch;
-                       pcp->batch = 1 * batch;
-                       INIT_LIST_HEAD(&pcp->list);
-
-                       pcp = &npageset->pcp[1];                /* cold*/
-                       pcp->count = 0;
-                       pcp->low = 0;
-                       pcp->high = 2 * batch;
-                       pcp->batch = 1 * batch;
-                       INIT_LIST_HEAD(&pcp->list);
-               }
-               zone->pageset[cpu] = npageset;
+               setup_pageset(zone->pageset[cpu], zone_batchsize(zone));
        }
 
        return 0;
@@ -1878,30 +1871,13 @@ static void __init free_area_init_core(struct pglist_data *pgdat,
                batch = zone_batchsize(zone);
 
                for (cpu = 0; cpu < NR_CPUS; cpu++) {
-                       struct per_cpu_pages *pcp;
 #ifdef CONFIG_NUMA
-                       struct per_cpu_pageset *pgset;
-                       pgset = &pageset_table[nid*MAX_NR_ZONES*NR_CPUS +
-                                       (j * NR_CPUS) + cpu];
-
-                       zone->pageset[cpu] = pgset;
+                       /* Early boot. Slab allocator not functional yet */
+                       zone->pageset[cpu] = &boot_pageset[cpu];
+                       setup_pageset(&boot_pageset[cpu],0);
 #else
-                       struct per_cpu_pageset *pgset = zone_pcp(zone, cpu);
+                       setup_pageset(zone_pcp(zone,cpu), batch);
 #endif
-
-                       pcp = &pgset->pcp[0];                   /* hot */
-                       pcp->count = 0;
-                       pcp->low = 2 * batch;
-                       pcp->high = 6 * batch;
-                       pcp->batch = 1 * batch;
-                       INIT_LIST_HEAD(&pcp->list);
-
-                       pcp = &pgset->pcp[1];                   /* cold */
-                       pcp->count = 0;
-                       pcp->low = 0;
-                       pcp->high = 2 * batch;
-                       pcp->batch = 1 * batch;
-                       INIT_LIST_HEAD(&pcp->list);
                }
                printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%lu\n",
                                zone_names[j], realsize, batch);