the-tree/mm/mm_init.c

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * mm_init.c - Memory initialisation verification and debugging
0004  *
0005  * Copyright 2008 IBM Corporation, 2008
0006  * Author Mel Gorman <mel@csn.ul.ie>
0007  *
0008  */
0009 #include <linux/kernel.h>
0010 #include <linux/init.h>
0011 #include <linux/kobject.h>
0012 #include <linux/export.h>
0013 #include <linux/memory.h>
0014 #include <linux/notifier.h>
0015 #include <linux/sched.h>
0016 #include <linux/mman.h>
0017 #include "internal.h"
0018
0019 #ifdef CONFIG_DEBUG_MEMORY_INIT
0020 int __meminitdata mminit_loglevel;
0021
0022 /* The zonelists are simply reported, validation is manual. */
0023 void __init mminit_verify_zonelist(void)
0024 {
0025     int nid;
0026
0027     if (mminit_loglevel < MMINIT_VERIFY)
0028         return;
0029
0030     for_each_online_node(nid) {
0031         pg_data_t *pgdat = NODE_DATA(nid);
0032         struct zone *zone;
0033         struct zoneref *z;
0034         struct zonelist *zonelist;
0035         int i, listid, zoneid;
0036
0037         BUILD_BUG_ON(MAX_ZONELISTS > 2);
0038         for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
0039
0040             /* Identify the zone and nodelist */
0041             zoneid = i % MAX_NR_ZONES;
0042             listid = i / MAX_NR_ZONES;
0043             zonelist = &pgdat->node_zonelists[listid];
0044             zone = &pgdat->node_zones[zoneid];
0045             if (!populated_zone(zone))
0046                 continue;
0047
0048             /* Print information about the zonelist */
0049             printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
0050                 listid > 0 ? "thisnode" : "general", nid,
0051                 zone->name);
0052
0053             /* Iterate the zonelist */
0054             for_each_zone_zonelist(zone, z, zonelist, zoneid)
0055                 pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
0056             pr_cont("\n");
0057         }
0058     }
0059 }
0060
0061 void __init mminit_verify_pageflags_layout(void)
0062 {
0063     int shift, width;
0064     unsigned long or_mask, add_mask;
0065
0066     shift = 8 * sizeof(unsigned long);
0067     width = shift - SECTIONS_WIDTH - NODES_WIDTH - ZONES_WIDTH
0068         - LAST_CPUPID_SHIFT - KASAN_TAG_WIDTH;
0069     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
0070         "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Flags %d\n",
0071         SECTIONS_WIDTH,
0072         NODES_WIDTH,
0073         ZONES_WIDTH,
0074         LAST_CPUPID_WIDTH,
0075         KASAN_TAG_WIDTH,
0076         NR_PAGEFLAGS);
0077     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
0078         "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
0079         SECTIONS_SHIFT,
0080         NODES_SHIFT,
0081         ZONES_SHIFT,
0082         LAST_CPUPID_SHIFT,
0083         KASAN_TAG_WIDTH);
0084     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
0085         "Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
0086         (unsigned long)SECTIONS_PGSHIFT,
0087         (unsigned long)NODES_PGSHIFT,
0088         (unsigned long)ZONES_PGSHIFT,
0089         (unsigned long)LAST_CPUPID_PGSHIFT,
0090         (unsigned long)KASAN_TAG_PGSHIFT);
0091     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
0092         "Node/Zone ID: %lu -> %lu\n",
0093         (unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
0094         (unsigned long)ZONEID_PGOFF);
0095     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
0096         "location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
0097         shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
0098 #ifdef NODE_NOT_IN_PAGE_FLAGS
0099     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
0100         "Node not in page flags");
0101 #endif
0102 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
0103     mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
0104         "Last cpupid not in page flags");
0105 #endif
0106
0107     if (SECTIONS_WIDTH) {
0108         shift -= SECTIONS_WIDTH;
0109         BUG_ON(shift != SECTIONS_PGSHIFT);
0110     }
0111     if (NODES_WIDTH) {
0112         shift -= NODES_WIDTH;
0113         BUG_ON(shift != NODES_PGSHIFT);
0114     }
0115     if (ZONES_WIDTH) {
0116         shift -= ZONES_WIDTH;
0117         BUG_ON(shift != ZONES_PGSHIFT);
0118     }
0119
0120     /* Check for bitmask overlaps */
0121     or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
0122             (NODES_MASK << NODES_PGSHIFT) |
0123             (SECTIONS_MASK << SECTIONS_PGSHIFT);
0124     add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
0125             (NODES_MASK << NODES_PGSHIFT) +
0126             (SECTIONS_MASK << SECTIONS_PGSHIFT);
0127     BUG_ON(or_mask != add_mask);
0128 }
0129
0130 static __init int set_mminit_loglevel(char *str)
0131 {
0132     get_option(&str, &mminit_loglevel);
0133     return 0;
0134 }
0135 early_param("mminit_loglevel", set_mminit_loglevel);
0136 #endif /* CONFIG_DEBUG_MEMORY_INIT */
0137
0138 struct kobject *mm_kobj;
0139 EXPORT_SYMBOL_GPL(mm_kobj);
0140
0141 #ifdef CONFIG_SMP
0142 s32 vm_committed_as_batch = 32;
0143
0144 void mm_compute_batch(int overcommit_policy)
0145 {
0146     u64 memsized_batch;
0147     s32 nr = num_present_cpus();
0148     s32 batch = max_t(s32, nr*2, 32);
0149     unsigned long ram_pages = totalram_pages();
0150
0151     /*
0152      * For policy OVERCOMMIT_NEVER, set batch size to 0.4% of
0153      * (total memory/#cpus), and lift it to 25% for other policies
0154      * to easy the possible lock contention for percpu_counter
0155      * vm_committed_as, while the max limit is INT_MAX
0156      */
0157     if (overcommit_policy == OVERCOMMIT_NEVER)
0158         memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
0159     else
0160         memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
0161
0162     vm_committed_as_batch = max_t(s32, memsized_batch, batch);
0163 }
0164
0165 static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
0166                     unsigned long action, void *arg)
0167 {
0168     switch (action) {
0169     case MEM_ONLINE:
0170     case MEM_OFFLINE:
0171         mm_compute_batch(sysctl_overcommit_memory);
0172         break;
0173     default:
0174         break;
0175     }
0176     return NOTIFY_OK;
0177 }
0178
0179 static struct notifier_block compute_batch_nb __meminitdata = {
0180     .notifier_call = mm_compute_batch_notifier,
0181     .priority = IPC_CALLBACK_PRI, /* use lowest priority */
0182 };
0183
0184 static int __init mm_compute_batch_init(void)
0185 {
0186     mm_compute_batch(sysctl_overcommit_memory);
0187     register_hotmemory_notifier(&compute_batch_nb);
0188
0189     return 0;
0190 }
0191
0192 __initcall(mm_compute_batch_init);
0193
0194 #endif
0195
0196 static int __init mm_sysfs_init(void)
0197 {
0198     mm_kobj = kobject_create_and_add("mm", kernel_kobj);
0199     if (!mm_kobj)
0200         return -ENOMEM;
0201
0202     return 0;
0203 }
0204 postcore_initcall(mm_sysfs_init);