xref: /linux/mm/mm_init.c (revision 23c48a124b469cee2eb0c75e6d22d366d1caa118)
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 /* The zonelists are simply reported, validation is manual. */
23 void __init mminit_verify_zonelist(void)
24 {
25 	int nid;
26 
27 	if (mminit_loglevel < MMINIT_VERIFY)
28 		return;
29 
30 	for_each_online_node(nid) {
31 		pg_data_t *pgdat = NODE_DATA(nid);
32 		struct zone *zone;
33 		struct zoneref *z;
34 		struct zonelist *zonelist;
35 		int i, listid, zoneid;
36 
37 		BUILD_BUG_ON(MAX_ZONELISTS > 2);
38 		for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
39 
40 			/* Identify the zone and nodelist */
41 			zoneid = i % MAX_NR_ZONES;
42 			listid = i / MAX_NR_ZONES;
43 			zonelist = &pgdat->node_zonelists[listid];
44 			zone = &pgdat->node_zones[zoneid];
45 			if (!populated_zone(zone))
46 				continue;
47 
48 			/* Print information about the zonelist */
49 			printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
50 				listid > 0 ? "thisnode" : "general", nid,
51 				zone->name);
52 
53 			/* Iterate the zonelist */
54 			for_each_zone_zonelist(zone, z, zonelist, zoneid)
55 				pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
56 			pr_cont("\n");
57 		}
58 	}
59 }
60 
61 void __init mminit_verify_pageflags_layout(void)
62 {
63 	int shift, width;
64 	unsigned long or_mask, add_mask;
65 
66 	shift = 8 * sizeof(unsigned long);
67 	width = shift - SECTIONS_WIDTH - NODES_WIDTH - ZONES_WIDTH
68 		- LAST_CPUPID_SHIFT - KASAN_TAG_WIDTH;
69 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
70 		"Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Flags %d\n",
71 		SECTIONS_WIDTH,
72 		NODES_WIDTH,
73 		ZONES_WIDTH,
74 		LAST_CPUPID_WIDTH,
75 		KASAN_TAG_WIDTH,
76 		NR_PAGEFLAGS);
77 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
78 		"Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
79 		SECTIONS_SHIFT,
80 		NODES_SHIFT,
81 		ZONES_SHIFT,
82 		LAST_CPUPID_SHIFT,
83 		KASAN_TAG_WIDTH);
84 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
85 		"Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
86 		(unsigned long)SECTIONS_PGSHIFT,
87 		(unsigned long)NODES_PGSHIFT,
88 		(unsigned long)ZONES_PGSHIFT,
89 		(unsigned long)LAST_CPUPID_PGSHIFT,
90 		(unsigned long)KASAN_TAG_PGSHIFT);
91 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
92 		"Node/Zone ID: %lu -> %lu\n",
93 		(unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
94 		(unsigned long)ZONEID_PGOFF);
95 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
96 		"location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
97 		shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
98 #ifdef NODE_NOT_IN_PAGE_FLAGS
99 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
100 		"Node not in page flags");
101 #endif
102 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
103 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
104 		"Last cpupid not in page flags");
105 #endif
106 
107 	if (SECTIONS_WIDTH) {
108 		shift -= SECTIONS_WIDTH;
109 		BUG_ON(shift != SECTIONS_PGSHIFT);
110 	}
111 	if (NODES_WIDTH) {
112 		shift -= NODES_WIDTH;
113 		BUG_ON(shift != NODES_PGSHIFT);
114 	}
115 	if (ZONES_WIDTH) {
116 		shift -= ZONES_WIDTH;
117 		BUG_ON(shift != ZONES_PGSHIFT);
118 	}
119 
120 	/* Check for bitmask overlaps */
121 	or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
122 			(NODES_MASK << NODES_PGSHIFT) |
123 			(SECTIONS_MASK << SECTIONS_PGSHIFT);
124 	add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
125 			(NODES_MASK << NODES_PGSHIFT) +
126 			(SECTIONS_MASK << SECTIONS_PGSHIFT);
127 	BUG_ON(or_mask != add_mask);
128 }
129 
130 static __init int set_mminit_loglevel(char *str)
131 {
132 	get_option(&str, &mminit_loglevel);
133 	return 0;
134 }
135 early_param("mminit_loglevel", set_mminit_loglevel);
136 #endif /* CONFIG_DEBUG_MEMORY_INIT */
137 
138 struct kobject *mm_kobj;
139 EXPORT_SYMBOL_GPL(mm_kobj);
140 
141 #ifdef CONFIG_SMP
142 s32 vm_committed_as_batch = 32;
143 
144 void mm_compute_batch(int overcommit_policy)
145 {
146 	u64 memsized_batch;
147 	s32 nr = num_present_cpus();
148 	s32 batch = max_t(s32, nr*2, 32);
149 	unsigned long ram_pages = totalram_pages();
150 
151 	/*
152 	 * For policy OVERCOMMIT_NEVER, set batch size to 0.4% of
153 	 * (total memory/#cpus), and lift it to 25% for other policies
154 	 * to easy the possible lock contention for percpu_counter
155 	 * vm_committed_as, while the max limit is INT_MAX
156 	 */
157 	if (overcommit_policy == OVERCOMMIT_NEVER)
158 		memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
159 	else
160 		memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
161 
162 	vm_committed_as_batch = max_t(s32, memsized_batch, batch);
163 }
164 
165 static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
166 					unsigned long action, void *arg)
167 {
168 	switch (action) {
169 	case MEM_ONLINE:
170 	case MEM_OFFLINE:
171 		mm_compute_batch(sysctl_overcommit_memory);
172 		break;
173 	default:
174 		break;
175 	}
176 	return NOTIFY_OK;
177 }
178 
179 static struct notifier_block compute_batch_nb __meminitdata = {
180 	.notifier_call = mm_compute_batch_notifier,
181 	.priority = IPC_CALLBACK_PRI, /* use lowest priority */
182 };
183 
184 static int __init mm_compute_batch_init(void)
185 {
186 	mm_compute_batch(sysctl_overcommit_memory);
187 	register_hotmemory_notifier(&compute_batch_nb);
188 
189 	return 0;
190 }
191 
192 __initcall(mm_compute_batch_init);
193 
194 #endif
195 
196 static int __init mm_sysfs_init(void)
197 {
198 	mm_kobj = kobject_create_and_add("mm", kernel_kobj);
199 	if (!mm_kobj)
200 		return -ENOMEM;
201 
202 	return 0;
203 }
204 postcore_initcall(mm_sysfs_init);
205