xref: /openbmc/linux/mm/mm_init.c (revision 6e63153d)
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 - LRU_GEN_WIDTH - LRU_REFS_WIDTH;
69 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
70 		"Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Gen %d Tier %d Flags %d\n",
71 		SECTIONS_WIDTH,
72 		NODES_WIDTH,
73 		ZONES_WIDTH,
74 		LAST_CPUPID_WIDTH,
75 		KASAN_TAG_WIDTH,
76 		LRU_GEN_WIDTH,
77 		LRU_REFS_WIDTH,
78 		NR_PAGEFLAGS);
79 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
80 		"Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
81 		SECTIONS_SHIFT,
82 		NODES_SHIFT,
83 		ZONES_SHIFT,
84 		LAST_CPUPID_SHIFT,
85 		KASAN_TAG_WIDTH);
86 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
87 		"Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
88 		(unsigned long)SECTIONS_PGSHIFT,
89 		(unsigned long)NODES_PGSHIFT,
90 		(unsigned long)ZONES_PGSHIFT,
91 		(unsigned long)LAST_CPUPID_PGSHIFT,
92 		(unsigned long)KASAN_TAG_PGSHIFT);
93 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
94 		"Node/Zone ID: %lu -> %lu\n",
95 		(unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
96 		(unsigned long)ZONEID_PGOFF);
97 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
98 		"location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
99 		shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
100 #ifdef NODE_NOT_IN_PAGE_FLAGS
101 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
102 		"Node not in page flags");
103 #endif
104 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
105 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
106 		"Last cpupid not in page flags");
107 #endif
108 
109 	if (SECTIONS_WIDTH) {
110 		shift -= SECTIONS_WIDTH;
111 		BUG_ON(shift != SECTIONS_PGSHIFT);
112 	}
113 	if (NODES_WIDTH) {
114 		shift -= NODES_WIDTH;
115 		BUG_ON(shift != NODES_PGSHIFT);
116 	}
117 	if (ZONES_WIDTH) {
118 		shift -= ZONES_WIDTH;
119 		BUG_ON(shift != ZONES_PGSHIFT);
120 	}
121 
122 	/* Check for bitmask overlaps */
123 	or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
124 			(NODES_MASK << NODES_PGSHIFT) |
125 			(SECTIONS_MASK << SECTIONS_PGSHIFT);
126 	add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
127 			(NODES_MASK << NODES_PGSHIFT) +
128 			(SECTIONS_MASK << SECTIONS_PGSHIFT);
129 	BUG_ON(or_mask != add_mask);
130 }
131 
132 static __init int set_mminit_loglevel(char *str)
133 {
134 	get_option(&str, &mminit_loglevel);
135 	return 0;
136 }
137 early_param("mminit_loglevel", set_mminit_loglevel);
138 #endif /* CONFIG_DEBUG_MEMORY_INIT */
139 
140 struct kobject *mm_kobj;
141 EXPORT_SYMBOL_GPL(mm_kobj);
142 
143 #ifdef CONFIG_SMP
144 s32 vm_committed_as_batch = 32;
145 
146 void mm_compute_batch(int overcommit_policy)
147 {
148 	u64 memsized_batch;
149 	s32 nr = num_present_cpus();
150 	s32 batch = max_t(s32, nr*2, 32);
151 	unsigned long ram_pages = totalram_pages();
152 
153 	/*
154 	 * For policy OVERCOMMIT_NEVER, set batch size to 0.4% of
155 	 * (total memory/#cpus), and lift it to 25% for other policies
156 	 * to easy the possible lock contention for percpu_counter
157 	 * vm_committed_as, while the max limit is INT_MAX
158 	 */
159 	if (overcommit_policy == OVERCOMMIT_NEVER)
160 		memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
161 	else
162 		memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
163 
164 	vm_committed_as_batch = max_t(s32, memsized_batch, batch);
165 }
166 
167 static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
168 					unsigned long action, void *arg)
169 {
170 	switch (action) {
171 	case MEM_ONLINE:
172 	case MEM_OFFLINE:
173 		mm_compute_batch(sysctl_overcommit_memory);
174 		break;
175 	default:
176 		break;
177 	}
178 	return NOTIFY_OK;
179 }
180 
181 static struct notifier_block compute_batch_nb __meminitdata = {
182 	.notifier_call = mm_compute_batch_notifier,
183 	.priority = IPC_CALLBACK_PRI, /* use lowest priority */
184 };
185 
186 static int __init mm_compute_batch_init(void)
187 {
188 	mm_compute_batch(sysctl_overcommit_memory);
189 	register_hotmemory_notifier(&compute_batch_nb);
190 
191 	return 0;
192 }
193 
194 __initcall(mm_compute_batch_init);
195 
196 #endif
197 
198 static int __init mm_sysfs_init(void)
199 {
200 	mm_kobj = kobject_create_and_add("mm", kernel_kobj);
201 	if (!mm_kobj)
202 		return -ENOMEM;
203 
204 	return 0;
205 }
206 postcore_initcall(mm_sysfs_init);
207