mm/mm_init.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * mm_init.c - Memory initialisation verification and debugging
   4  *
   5  * Copyright 2008 IBM Corporation, 2008
   6  * Author Mel Gorman <mel@csn.ul.ie>
   7  *
   8  */
   9 #include <linux/kernel.h>
  10 #include <linux/init.h>
  11 #include <linux/kobject.h>
  12 #include <linux/export.h>
  13 #include <linux/memory.h>
  14 #include <linux/notifier.h>
  15 #include <linux/sched.h>
  16 #include <linux/mman.h>
  17 #include "internal.h"
  18
  19 #ifdef CONFIG_DEBUG_MEMORY_INIT
  20 int __meminitdata mminit_loglevel;
  21
  22 #ifndef SECTIONS_SHIFT
  23 #define SECTIONS_SHIFT  0
  24 #endif
  25
  26 /* The zonelists are simply reported, validation is manual. */
  27 void __init mminit_verify_zonelist(void)
  28 {
  29         int nid;
  30
  31         if (mminit_loglevel < MMINIT_VERIFY)
  32                 return;
  33
  34         for_each_online_node(nid) {
  35                 pg_data_t *pgdat = NODE_DATA(nid);
  36                 struct zone *zone;
  37                 struct zoneref *z;
  38                 struct zonelist *zonelist;
  39                 int i, listid, zoneid;
  40
  41                 BUILD_BUG_ON(MAX_ZONELISTS > 2);
  42                 for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
  43
  44                         /* Identify the zone and nodelist */
  45                         zoneid = i % MAX_NR_ZONES;
  46                         listid = i / MAX_NR_ZONES;
  47                         zonelist = &pgdat->node_zonelists[listid];
  48                         zone = &pgdat->node_zones[zoneid];
  49                         if (!populated_zone(zone))
  50                                 continue;
  51
  52                         /* Print information about the zonelist */
  53                         printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
  54                                 listid > 0 ? "thisnode" : "general", nid,
  55                                 zone->name);
  56
  57                         /* Iterate the zonelist */
  58                         for_each_zone_zonelist(zone, z, zonelist, zoneid)
  59                                 pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
  60                         pr_cont("\n");
  61                 }
  62         }
  63 }
  64
  65 void __init mminit_verify_pageflags_layout(void)
  66 {
  67         int shift, width;
  68         unsigned long or_mask, add_mask;
  69
  70         shift = 8 * sizeof(unsigned long);
  71         width = shift - SECTIONS_WIDTH - NODES_WIDTH - ZONES_WIDTH
  72                 - LAST_CPUPID_SHIFT - KASAN_TAG_WIDTH;
  73         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
  74                 "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Flags %d\n",
  75                 SECTIONS_WIDTH,
  76                 NODES_WIDTH,
  77                 ZONES_WIDTH,
  78                 LAST_CPUPID_WIDTH,
  79                 KASAN_TAG_WIDTH,
  80                 NR_PAGEFLAGS);
  81         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
  82                 "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
  83                 SECTIONS_SHIFT,
  84                 NODES_SHIFT,
  85                 ZONES_SHIFT,
  86                 LAST_CPUPID_SHIFT,
  87                 KASAN_TAG_WIDTH);
  88         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
  89                 "Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
  90                 (unsigned long)SECTIONS_PGSHIFT,
  91                 (unsigned long)NODES_PGSHIFT,
  92                 (unsigned long)ZONES_PGSHIFT,
  93                 (unsigned long)LAST_CPUPID_PGSHIFT,
  94                 (unsigned long)KASAN_TAG_PGSHIFT);
  95         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
  96                 "Node/Zone ID: %lu -> %lu\n",
  97                 (unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
  98                 (unsigned long)ZONEID_PGOFF);
  99         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
 100                 "location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
 101                 shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
 102 #ifdef NODE_NOT_IN_PAGE_FLAGS
 103         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
 104                 "Node not in page flags");
 105 #endif
 106 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
 107         mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
 108                 "Last cpupid not in page flags");
 109 #endif
 110
 111         if (SECTIONS_WIDTH) {
 112                 shift -= SECTIONS_WIDTH;
 113                 BUG_ON(shift != SECTIONS_PGSHIFT);
 114         }
 115         if (NODES_WIDTH) {
 116                 shift -= NODES_WIDTH;
 117                 BUG_ON(shift != NODES_PGSHIFT);
 118         }
 119         if (ZONES_WIDTH) {
 120                 shift -= ZONES_WIDTH;
 121                 BUG_ON(shift != ZONES_PGSHIFT);
 122         }
 123
 124         /* Check for bitmask overlaps */
 125         or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
 126                         (NODES_MASK << NODES_PGSHIFT) |
 127                         (SECTIONS_MASK << SECTIONS_PGSHIFT);
 128         add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
 129                         (NODES_MASK << NODES_PGSHIFT) +
 130                         (SECTIONS_MASK << SECTIONS_PGSHIFT);
 131         BUG_ON(or_mask != add_mask);
 132 }
 133
 134 static __init int set_mminit_loglevel(char *str)
 135 {
 136         get_option(&str, &mminit_loglevel);
 137         return 0;
 138 }
 139 early_param("mminit_loglevel", set_mminit_loglevel);
 140 #endif /* CONFIG_DEBUG_MEMORY_INIT */
 141
 142 struct kobject *mm_kobj;
 143 EXPORT_SYMBOL_GPL(mm_kobj);
 144
 145 #ifdef CONFIG_SMP
 146 s32 vm_committed_as_batch = 32;
 147
 148 void mm_compute_batch(int overcommit_policy)
 149 {
 150         u64 memsized_batch;
 151         s32 nr = num_present_cpus();
 152         s32 batch = max_t(s32, nr*2, 32);
 153         unsigned long ram_pages = totalram_pages();
 154
 155         /*
 156          * For policy OVERCOMMIT_NEVER, set batch size to 0.4% of
 157          * (total memory/#cpus), and lift it to 25% for other policies
 158          * to easy the possible lock contention for percpu_counter
 159          * vm_committed_as, while the max limit is INT_MAX
 160          */
 161         if (overcommit_policy == OVERCOMMIT_NEVER)
 162                 memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
 163         else
 164                 memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
 165
 166         vm_committed_as_batch = max_t(s32, memsized_batch, batch);
 167 }
 168
 169 static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
 170                                         unsigned long action, void *arg)
 171 {
 172         switch (action) {
 173         case MEM_ONLINE:
 174         case MEM_OFFLINE:
 175                 mm_compute_batch(sysctl_overcommit_memory);
 176                 break;
 177         default:
 178                 break;
 179         }
 180         return NOTIFY_OK;
 181 }
 182
 183 static struct notifier_block compute_batch_nb __meminitdata = {
 184         .notifier_call = mm_compute_batch_notifier,
 185         .priority = IPC_CALLBACK_PRI, /* use lowest priority */
 186 };
 187
 188 static int __init mm_compute_batch_init(void)
 189 {
 190         mm_compute_batch(sysctl_overcommit_memory);
 191         register_hotmemory_notifier(&compute_batch_nb);
 192
 193         return 0;
 194 }
 195
 196 __initcall(mm_compute_batch_init);
 197
 198 #endif
 199
 200 static int __init mm_sysfs_init(void)
 201 {
 202         mm_kobj = kobject_create_and_add("mm", kernel_kobj);
 203         if (!mm_kobj)
 204                 return -ENOMEM;
 205
 206         return 0;
 207 }
 208 postcore_initcall(mm_sysfs_init);