xref: /openbmc/linux/mm/shuffle.c (revision 132db935)
1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright(c) 2018 Intel Corporation. All rights reserved.
3 
4 #include <linux/mm.h>
5 #include <linux/init.h>
6 #include <linux/mmzone.h>
7 #include <linux/random.h>
8 #include <linux/moduleparam.h>
9 #include "internal.h"
10 #include "shuffle.h"
11 
12 DEFINE_STATIC_KEY_FALSE(page_alloc_shuffle_key);
13 static unsigned long shuffle_state __ro_after_init;
14 
15 /*
16  * Depending on the architecture, module parameter parsing may run
17  * before, or after the cache detection. SHUFFLE_FORCE_DISABLE prevents,
18  * or reverts the enabling of the shuffle implementation. SHUFFLE_ENABLE
19  * attempts to turn on the implementation, but aborts if it finds
20  * SHUFFLE_FORCE_DISABLE already set.
21  */
22 __meminit void page_alloc_shuffle(enum mm_shuffle_ctl ctl)
23 {
24 	if (ctl == SHUFFLE_FORCE_DISABLE)
25 		set_bit(SHUFFLE_FORCE_DISABLE, &shuffle_state);
26 
27 	if (test_bit(SHUFFLE_FORCE_DISABLE, &shuffle_state)) {
28 		if (test_and_clear_bit(SHUFFLE_ENABLE, &shuffle_state))
29 			static_branch_disable(&page_alloc_shuffle_key);
30 	} else if (ctl == SHUFFLE_ENABLE
31 			&& !test_and_set_bit(SHUFFLE_ENABLE, &shuffle_state))
32 		static_branch_enable(&page_alloc_shuffle_key);
33 }
34 
35 static bool shuffle_param;
36 static int shuffle_show(char *buffer, const struct kernel_param *kp)
37 {
38 	return sprintf(buffer, "%c\n", test_bit(SHUFFLE_ENABLE, &shuffle_state)
39 			? 'Y' : 'N');
40 }
41 
42 static __meminit int shuffle_store(const char *val,
43 		const struct kernel_param *kp)
44 {
45 	int rc = param_set_bool(val, kp);
46 
47 	if (rc < 0)
48 		return rc;
49 	if (shuffle_param)
50 		page_alloc_shuffle(SHUFFLE_ENABLE);
51 	else
52 		page_alloc_shuffle(SHUFFLE_FORCE_DISABLE);
53 	return 0;
54 }
55 module_param_call(shuffle, shuffle_store, shuffle_show, &shuffle_param, 0400);
56 
57 /*
58  * For two pages to be swapped in the shuffle, they must be free (on a
59  * 'free_area' lru), have the same order, and have the same migratetype.
60  */
61 static struct page * __meminit shuffle_valid_page(unsigned long pfn, int order)
62 {
63 	struct page *page;
64 
65 	/*
66 	 * Given we're dealing with randomly selected pfns in a zone we
67 	 * need to ask questions like...
68 	 */
69 
70 	/* ...is the pfn even in the memmap? */
71 	if (!pfn_valid_within(pfn))
72 		return NULL;
73 
74 	/* ...is the pfn in a present section or a hole? */
75 	if (!pfn_in_present_section(pfn))
76 		return NULL;
77 
78 	/* ...is the page free and currently on a free_area list? */
79 	page = pfn_to_page(pfn);
80 	if (!PageBuddy(page))
81 		return NULL;
82 
83 	/*
84 	 * ...is the page on the same list as the page we will
85 	 * shuffle it with?
86 	 */
87 	if (page_order(page) != order)
88 		return NULL;
89 
90 	return page;
91 }
92 
93 /*
94  * Fisher-Yates shuffle the freelist which prescribes iterating through an
95  * array, pfns in this case, and randomly swapping each entry with another in
96  * the span, end_pfn - start_pfn.
97  *
98  * To keep the implementation simple it does not attempt to correct for sources
99  * of bias in the distribution, like modulo bias or pseudo-random number
100  * generator bias. I.e. the expectation is that this shuffling raises the bar
101  * for attacks that exploit the predictability of page allocations, but need not
102  * be a perfect shuffle.
103  */
104 #define SHUFFLE_RETRY 10
105 void __meminit __shuffle_zone(struct zone *z)
106 {
107 	unsigned long i, flags;
108 	unsigned long start_pfn = z->zone_start_pfn;
109 	unsigned long end_pfn = zone_end_pfn(z);
110 	const int order = SHUFFLE_ORDER;
111 	const int order_pages = 1 << order;
112 
113 	spin_lock_irqsave(&z->lock, flags);
114 	start_pfn = ALIGN(start_pfn, order_pages);
115 	for (i = start_pfn; i < end_pfn; i += order_pages) {
116 		unsigned long j;
117 		int migratetype, retry;
118 		struct page *page_i, *page_j;
119 
120 		/*
121 		 * We expect page_i, in the sub-range of a zone being added
122 		 * (@start_pfn to @end_pfn), to more likely be valid compared to
123 		 * page_j randomly selected in the span @zone_start_pfn to
124 		 * @spanned_pages.
125 		 */
126 		page_i = shuffle_valid_page(i, order);
127 		if (!page_i)
128 			continue;
129 
130 		for (retry = 0; retry < SHUFFLE_RETRY; retry++) {
131 			/*
132 			 * Pick a random order aligned page in the zone span as
133 			 * a swap target. If the selected pfn is a hole, retry
134 			 * up to SHUFFLE_RETRY attempts find a random valid pfn
135 			 * in the zone.
136 			 */
137 			j = z->zone_start_pfn +
138 				ALIGN_DOWN(get_random_long() % z->spanned_pages,
139 						order_pages);
140 			page_j = shuffle_valid_page(j, order);
141 			if (page_j && page_j != page_i)
142 				break;
143 		}
144 		if (retry >= SHUFFLE_RETRY) {
145 			pr_debug("%s: failed to swap %#lx\n", __func__, i);
146 			continue;
147 		}
148 
149 		/*
150 		 * Each migratetype corresponds to its own list, make sure the
151 		 * types match otherwise we're moving pages to lists where they
152 		 * do not belong.
153 		 */
154 		migratetype = get_pageblock_migratetype(page_i);
155 		if (get_pageblock_migratetype(page_j) != migratetype) {
156 			pr_debug("%s: migratetype mismatch %#lx\n", __func__, i);
157 			continue;
158 		}
159 
160 		list_swap(&page_i->lru, &page_j->lru);
161 
162 		pr_debug("%s: swap: %#lx -> %#lx\n", __func__, i, j);
163 
164 		/* take it easy on the zone lock */
165 		if ((i % (100 * order_pages)) == 0) {
166 			spin_unlock_irqrestore(&z->lock, flags);
167 			cond_resched();
168 			spin_lock_irqsave(&z->lock, flags);
169 		}
170 	}
171 	spin_unlock_irqrestore(&z->lock, flags);
172 }
173 
174 /**
175  * shuffle_free_memory - reduce the predictability of the page allocator
176  * @pgdat: node page data
177  */
178 void __meminit __shuffle_free_memory(pg_data_t *pgdat)
179 {
180 	struct zone *z;
181 
182 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
183 		shuffle_zone(z);
184 }
185 
186 bool shuffle_pick_tail(void)
187 {
188 	static u64 rand;
189 	static u8 rand_bits;
190 	bool ret;
191 
192 	/*
193 	 * The lack of locking is deliberate. If 2 threads race to
194 	 * update the rand state it just adds to the entropy.
195 	 */
196 	if (rand_bits == 0) {
197 		rand_bits = 64;
198 		rand = get_random_u64();
199 	}
200 
201 	ret = rand & 1;
202 
203 	rand_bits--;
204 	rand >>= 1;
205 
206 	return ret;
207 }
208