• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2016 Fujitsu Ltd.
3  *  Author: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
4  *  Ported: Guangwen Feng <fenggw-fnst@cn.fujitsu.com>
5  *
6  * This program is free software: you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation, either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program, if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 /*
21  * This is a regression test for the race condition between move_pages()
22  * and freeing hugepages, where move_pages() calls follow_page(FOLL_GET)
23  * for hugepages internally and tries to get its refcount without
24  * preventing concurrent freeing.
25  *
26  * This test can crash the buggy kernel, and the bug was fixed in:
27  *
28  *  commit e66f17ff71772b209eed39de35aaa99ba819c93d
29  *  Author: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
30  *  Date:   Wed Feb 11 15:25:22 2015 -0800
31  *
32  *  mm/hugetlb: take page table lock in follow_huge_pmd()
33  */
34 
35 #include <errno.h>
36 #include <unistd.h>
37 #include <string.h>
38 #include <stdio.h>
39 #include <sys/types.h>
40 #include <sys/wait.h>
41 
42 #include "tst_test.h"
43 #include "move_pages_support.h"
44 #include "lapi/mmap.h"
45 
46 #ifdef HAVE_NUMA_V2
47 
48 #define LOOPS	1000
49 #define PATH_MEMINFO	"/proc/meminfo"
50 #define PATH_NR_HUGEPAGES	"/proc/sys/vm/nr_hugepages"
51 #define PATH_HUGEPAGES	"/sys/kernel/mm/hugepages/"
52 #define TEST_PAGES	2
53 #define TEST_NODES	2
54 
55 static int pgsz, hpsz;
56 static long orig_hugepages = -1;
57 static char path_hugepages_node1[PATH_MAX];
58 static char path_hugepages_node2[PATH_MAX];
59 static long orig_hugepages_node1 = -1;
60 static long orig_hugepages_node2 = -1;
61 static unsigned int node1, node2;
62 static void *addr;
63 
do_child(void)64 static void do_child(void)
65 {
66 	int test_pages = TEST_PAGES * hpsz / pgsz;
67 	int i, j;
68 	int *nodes, *status;
69 	void **pages;
70 	pid_t ppid = getppid();
71 
72 	pages = SAFE_MALLOC(sizeof(char *) * test_pages);
73 	nodes = SAFE_MALLOC(sizeof(int) * test_pages);
74 	status = SAFE_MALLOC(sizeof(int) * test_pages);
75 
76 	for (i = 0; i < test_pages; i++)
77 		pages[i] = addr + i * pgsz;
78 
79 	for (i = 0; ; i++) {
80 		for (j = 0; j < test_pages; j++) {
81 			if (i % 2 == 0)
82 				nodes[j] = node1;
83 			else
84 				nodes[j] = node2;
85 			status[j] = 0;
86 		}
87 
88 		TEST(numa_move_pages(ppid, test_pages,
89 			pages, nodes, status, MPOL_MF_MOVE_ALL));
90 		if (TST_RET < 0) {
91 			tst_res(TFAIL | TTERRNO, "move_pages failed");
92 			break;
93 		}
94 	}
95 
96 	exit(0);
97 }
98 
do_test(void)99 static void do_test(void)
100 {
101 	int i;
102 	pid_t cpid = -1;
103 	int status;
104 	unsigned int twenty_percent = (tst_timeout_remaining() / 5);
105 
106 	addr = SAFE_MMAP(NULL, TEST_PAGES * hpsz, PROT_READ | PROT_WRITE,
107 		MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0);
108 
109 	SAFE_MUNMAP(addr, TEST_PAGES * hpsz);
110 
111 	cpid = SAFE_FORK();
112 	if (cpid == 0)
113 		do_child();
114 
115 	for (i = 0; i < LOOPS; i++) {
116 		void *ptr;
117 
118 		ptr = SAFE_MMAP(NULL, TEST_PAGES * hpsz,
119 			PROT_READ | PROT_WRITE,
120 			MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0);
121 		if (ptr != addr)
122 			tst_brk(TBROK, "Failed to mmap at desired addr");
123 
124 		memset(addr, 0, TEST_PAGES * hpsz);
125 
126 		SAFE_MUNMAP(addr, TEST_PAGES * hpsz);
127 
128 		if (tst_timeout_remaining() < twenty_percent)
129 			break;
130 	}
131 
132 	SAFE_KILL(cpid, SIGKILL);
133 	SAFE_WAITPID(cpid, &status, 0);
134 	if (!WIFEXITED(status))
135 		tst_res(TPASS, "Bug not reproduced");
136 }
137 
alloc_free_huge_on_node(unsigned int node,size_t size)138 static void alloc_free_huge_on_node(unsigned int node, size_t size)
139 {
140 	char *mem;
141 	long ret;
142 	struct bitmask *bm;
143 
144 	tst_res(TINFO, "Allocating and freeing %zu hugepages on node %u",
145 		size / hpsz, node);
146 
147 	mem = mmap(NULL, size, PROT_READ | PROT_WRITE,
148 		   MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -1, 0);
149 	if (mem == MAP_FAILED) {
150 		if (errno == ENOMEM)
151 			tst_brk(TCONF, "Cannot allocate huge pages");
152 
153 		tst_brk(TBROK | TERRNO, "mmap(..., MAP_HUGETLB, ...) failed");
154 	}
155 
156 	bm = numa_bitmask_alloc(numa_max_possible_node() + 1);
157 	if (!bm)
158 		tst_brk(TBROK | TERRNO, "numa_bitmask_alloc() failed");
159 
160 	numa_bitmask_setbit(bm, node);
161 
162 	ret = mbind(mem, size, MPOL_BIND, bm->maskp, bm->size + 1, 0);
163 	if (ret) {
164 		if (errno == ENOMEM)
165 			tst_brk(TCONF, "Cannot mbind huge pages");
166 
167 		tst_brk(TBROK | TERRNO, "mbind() failed");
168 	}
169 
170 	TEST(mlock(mem, size));
171 	if (TST_RET) {
172 		SAFE_MUNMAP(mem, size);
173 		if (TST_ERR == ENOMEM || TST_ERR == EAGAIN)
174 			tst_brk(TCONF, "Cannot lock huge pages");
175 		tst_brk(TBROK | TTERRNO, "mlock failed");
176 	}
177 
178 	numa_bitmask_free(bm);
179 
180 	SAFE_MUNMAP(mem, size);
181 }
182 
setup(void)183 static void setup(void)
184 {
185 	int ret;
186 	long memfree;
187 
188 	check_config(TEST_NODES);
189 
190 	if (access(PATH_HUGEPAGES, F_OK))
191 		tst_brk(TCONF, "Huge page not supported");
192 
193 	ret = get_allowed_nodes(NH_MEMS, TEST_NODES, &node1, &node2);
194 	if (ret < 0)
195 		tst_brk(TBROK | TERRNO, "get_allowed_nodes: %d", ret);
196 
197 	pgsz = (int)get_page_size();
198 	SAFE_FILE_LINES_SCANF(PATH_MEMINFO, "Hugepagesize: %d", &hpsz);
199 
200 	SAFE_FILE_LINES_SCANF(PATH_MEMINFO, "MemFree: %ld", &memfree);
201 	tst_res(TINFO, "Free RAM %ld kB", memfree);
202 
203 	if (4 * hpsz > memfree)
204 		tst_brk(TBROK, "Not enough free RAM");
205 
206 	snprintf(path_hugepages_node1, sizeof(path_hugepages_node1),
207 		 "/sys/devices/system/node/node%u/hugepages/hugepages-%dkB/nr_hugepages",
208 		 node1, hpsz);
209 
210 	snprintf(path_hugepages_node2, sizeof(path_hugepages_node2),
211 		 "/sys/devices/system/node/node%u/hugepages/hugepages-%dkB/nr_hugepages",
212 		 node2, hpsz);
213 
214 	if (!access(path_hugepages_node1, F_OK)) {
215 		SAFE_FILE_SCANF(path_hugepages_node1,
216 				"%ld", &orig_hugepages_node1);
217 		tst_res(TINFO,
218 			"Increasing %dkB hugepages pool on node %u to %ld",
219 			hpsz, node1, orig_hugepages_node1 + 4);
220 		SAFE_FILE_PRINTF(path_hugepages_node1,
221 				 "%ld", orig_hugepages_node1 + 4);
222 	}
223 
224 	if (!access(path_hugepages_node2, F_OK)) {
225 		SAFE_FILE_SCANF(path_hugepages_node2,
226 				"%ld", &orig_hugepages_node2);
227 		tst_res(TINFO,
228 			"Increasing %dkB hugepages pool on node %u to %ld",
229 			hpsz, node2, orig_hugepages_node2 + 4);
230 		SAFE_FILE_PRINTF(path_hugepages_node2,
231 				 "%ld", orig_hugepages_node2 + 4);
232 	}
233 
234 	hpsz *= 1024;
235 
236 	if (orig_hugepages_node1 == -1 || orig_hugepages_node2 == -1) {
237 		SAFE_FILE_SCANF(PATH_NR_HUGEPAGES, "%ld", &orig_hugepages);
238 		tst_res(TINFO, "Increasing global hugepages pool to %ld",
239 			orig_hugepages + 8);
240 		SAFE_FILE_PRINTF(PATH_NR_HUGEPAGES, "%ld", orig_hugepages + 8);
241 	}
242 
243 	alloc_free_huge_on_node(node1, 4L * hpsz);
244 	alloc_free_huge_on_node(node2, 4L * hpsz);
245 }
246 
cleanup(void)247 static void cleanup(void)
248 {
249 	if (orig_hugepages != -1)
250 		SAFE_FILE_PRINTF(PATH_NR_HUGEPAGES, "%ld", orig_hugepages);
251 
252 	if (orig_hugepages_node1 != -1) {
253 		SAFE_FILE_PRINTF(path_hugepages_node1,
254 				 "%ld", orig_hugepages_node1);
255 	}
256 
257 	if (orig_hugepages_node2 != -1) {
258 		SAFE_FILE_PRINTF(path_hugepages_node2,
259 				 "%ld", orig_hugepages_node2);
260 	}
261 }
262 
263 static struct tst_test test = {
264 	.min_kver = "2.6.32",
265 	.needs_root = 1,
266 	.forks_child = 1,
267 	.setup = setup,
268 	.cleanup = cleanup,
269 	.test_all = do_test,
270 };
271 
272 #else
273 	TST_TEST_TCONF(NUMA_ERROR_MSG);
274 #endif
275