1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright 2020 Google LLC
4 */
5 #define _GNU_SOURCE
6
7 #include <errno.h>
8 #include <stdlib.h>
9 #include <stdio.h>
10 #include <string.h>
11 #include <sys/mman.h>
12 #include <time.h>
13 #include <stdbool.h>
14
15 #include "../kselftest.h"
16
17 #define EXPECT_SUCCESS 0
18 #define EXPECT_FAILURE 1
19 #define NON_OVERLAPPING 0
20 #define OVERLAPPING 1
21 #define NS_PER_SEC 1000000000ULL
22 #define VALIDATION_DEFAULT_THRESHOLD 4 /* 4MB */
23 #define VALIDATION_NO_THRESHOLD 0 /* Verify the entire region */
24
25 #define MIN(X, Y) ((X) < (Y) ? (X) : (Y))
26
27 struct config {
28 unsigned long long src_alignment;
29 unsigned long long dest_alignment;
30 unsigned long long region_size;
31 int overlapping;
32 };
33
34 struct test {
35 const char *name;
36 struct config config;
37 int expect_failure;
38 };
39
40 enum {
41 _1KB = 1ULL << 10, /* 1KB -> not page aligned */
42 _4KB = 4ULL << 10,
43 _8KB = 8ULL << 10,
44 _1MB = 1ULL << 20,
45 _2MB = 2ULL << 20,
46 _4MB = 4ULL << 20,
47 _1GB = 1ULL << 30,
48 _2GB = 2ULL << 30,
49 PMD = _2MB,
50 PUD = _1GB,
51 };
52
53 #define PTE page_size
54
55 #define MAKE_TEST(source_align, destination_align, size, \
56 overlaps, should_fail, test_name) \
57 (struct test){ \
58 .name = test_name, \
59 .config = { \
60 .src_alignment = source_align, \
61 .dest_alignment = destination_align, \
62 .region_size = size, \
63 .overlapping = overlaps, \
64 }, \
65 .expect_failure = should_fail \
66 }
67
68 /*
69 * Returns false if the requested remap region overlaps with an
70 * existing mapping (e.g text, stack) else returns true.
71 */
is_remap_region_valid(void * addr,unsigned long long size)72 static bool is_remap_region_valid(void *addr, unsigned long long size)
73 {
74 void *remap_addr = NULL;
75 bool ret = true;
76
77 /* Use MAP_FIXED_NOREPLACE flag to ensure region is not mapped */
78 remap_addr = mmap(addr, size, PROT_READ | PROT_WRITE,
79 MAP_FIXED_NOREPLACE | MAP_ANONYMOUS | MAP_SHARED,
80 -1, 0);
81
82 if (remap_addr == MAP_FAILED) {
83 if (errno == EEXIST)
84 ret = false;
85 } else {
86 munmap(remap_addr, size);
87 }
88
89 return ret;
90 }
91
92 /* Returns mmap_min_addr sysctl tunable from procfs */
get_mmap_min_addr(void)93 static unsigned long long get_mmap_min_addr(void)
94 {
95 FILE *fp;
96 int n_matched;
97 static unsigned long long addr;
98
99 if (addr)
100 return addr;
101
102 fp = fopen("/proc/sys/vm/mmap_min_addr", "r");
103 if (fp == NULL) {
104 ksft_print_msg("Failed to open /proc/sys/vm/mmap_min_addr: %s\n",
105 strerror(errno));
106 exit(KSFT_SKIP);
107 }
108
109 n_matched = fscanf(fp, "%llu", &addr);
110 if (n_matched != 1) {
111 ksft_print_msg("Failed to read /proc/sys/vm/mmap_min_addr: %s\n",
112 strerror(errno));
113 fclose(fp);
114 exit(KSFT_SKIP);
115 }
116
117 fclose(fp);
118 return addr;
119 }
120
121 /*
122 * Returns the start address of the mapping on success, else returns
123 * NULL on failure.
124 */
get_source_mapping(struct config c)125 static void *get_source_mapping(struct config c)
126 {
127 unsigned long long addr = 0ULL;
128 void *src_addr = NULL;
129 unsigned long long mmap_min_addr;
130
131 mmap_min_addr = get_mmap_min_addr();
132
133 retry:
134 addr += c.src_alignment;
135 if (addr < mmap_min_addr)
136 goto retry;
137
138 src_addr = mmap((void *) addr, c.region_size, PROT_READ | PROT_WRITE,
139 MAP_FIXED_NOREPLACE | MAP_ANONYMOUS | MAP_SHARED,
140 -1, 0);
141 if (src_addr == MAP_FAILED) {
142 if (errno == EPERM || errno == EEXIST)
143 goto retry;
144 goto error;
145 }
146 /*
147 * Check that the address is aligned to the specified alignment.
148 * Addresses which have alignments that are multiples of that
149 * specified are not considered valid. For instance, 1GB address is
150 * 2MB-aligned, however it will not be considered valid for a
151 * requested alignment of 2MB. This is done to reduce coincidental
152 * alignment in the tests.
153 */
154 if (((unsigned long long) src_addr & (c.src_alignment - 1)) ||
155 !((unsigned long long) src_addr & c.src_alignment)) {
156 munmap(src_addr, c.region_size);
157 goto retry;
158 }
159
160 if (!src_addr)
161 goto error;
162
163 return src_addr;
164 error:
165 ksft_print_msg("Failed to map source region: %s\n",
166 strerror(errno));
167 return NULL;
168 }
169
170 /* Returns the time taken for the remap on success else returns -1. */
remap_region(struct config c,unsigned int threshold_mb,char pattern_seed)171 static long long remap_region(struct config c, unsigned int threshold_mb,
172 char pattern_seed)
173 {
174 void *addr, *src_addr, *dest_addr;
175 unsigned long long i;
176 struct timespec t_start = {0, 0}, t_end = {0, 0};
177 long long start_ns, end_ns, align_mask, ret, offset;
178 unsigned long long threshold;
179
180 if (threshold_mb == VALIDATION_NO_THRESHOLD)
181 threshold = c.region_size;
182 else
183 threshold = MIN(threshold_mb * _1MB, c.region_size);
184
185 src_addr = get_source_mapping(c);
186 if (!src_addr) {
187 ret = -1;
188 goto out;
189 }
190
191 /* Set byte pattern */
192 srand(pattern_seed);
193 for (i = 0; i < threshold; i++)
194 memset((char *) src_addr + i, (char) rand(), 1);
195
196 /* Mask to zero out lower bits of address for alignment */
197 align_mask = ~(c.dest_alignment - 1);
198 /* Offset of destination address from the end of the source region */
199 offset = (c.overlapping) ? -c.dest_alignment : c.dest_alignment;
200 addr = (void *) (((unsigned long long) src_addr + c.region_size
201 + offset) & align_mask);
202
203 /* See comment in get_source_mapping() */
204 if (!((unsigned long long) addr & c.dest_alignment))
205 addr = (void *) ((unsigned long long) addr | c.dest_alignment);
206
207 /* Don't destroy existing mappings unless expected to overlap */
208 while (!is_remap_region_valid(addr, c.region_size) && !c.overlapping) {
209 /* Check for unsigned overflow */
210 if (addr + c.dest_alignment < addr) {
211 ksft_print_msg("Couldn't find a valid region to remap to\n");
212 ret = -1;
213 goto out;
214 }
215 addr += c.dest_alignment;
216 }
217
218 clock_gettime(CLOCK_MONOTONIC, &t_start);
219 dest_addr = mremap(src_addr, c.region_size, c.region_size,
220 MREMAP_MAYMOVE|MREMAP_FIXED, (char *) addr);
221 clock_gettime(CLOCK_MONOTONIC, &t_end);
222
223 if (dest_addr == MAP_FAILED) {
224 ksft_print_msg("mremap failed: %s\n", strerror(errno));
225 ret = -1;
226 goto clean_up_src;
227 }
228
229 /* Verify byte pattern after remapping */
230 srand(pattern_seed);
231 for (i = 0; i < threshold; i++) {
232 char c = (char) rand();
233
234 if (((char *) dest_addr)[i] != c) {
235 ksft_print_msg("Data after remap doesn't match at offset %d\n",
236 i);
237 ksft_print_msg("Expected: %#x\t Got: %#x\n", c & 0xff,
238 ((char *) dest_addr)[i] & 0xff);
239 ret = -1;
240 goto clean_up_dest;
241 }
242 }
243
244 start_ns = t_start.tv_sec * NS_PER_SEC + t_start.tv_nsec;
245 end_ns = t_end.tv_sec * NS_PER_SEC + t_end.tv_nsec;
246 ret = end_ns - start_ns;
247
248 /*
249 * Since the destination address is specified using MREMAP_FIXED, subsequent
250 * mremap will unmap any previous mapping at the address range specified by
251 * dest_addr and region_size. This significantly affects the remap time of
252 * subsequent tests. So we clean up mappings after each test.
253 */
254 clean_up_dest:
255 munmap(dest_addr, c.region_size);
256 clean_up_src:
257 munmap(src_addr, c.region_size);
258 out:
259 return ret;
260 }
261
run_mremap_test_case(struct test test_case,int * failures,unsigned int threshold_mb,unsigned int pattern_seed)262 static void run_mremap_test_case(struct test test_case, int *failures,
263 unsigned int threshold_mb,
264 unsigned int pattern_seed)
265 {
266 long long remap_time = remap_region(test_case.config, threshold_mb,
267 pattern_seed);
268
269 if (remap_time < 0) {
270 if (test_case.expect_failure)
271 ksft_test_result_xfail("%s\n\tExpected mremap failure\n",
272 test_case.name);
273 else {
274 ksft_test_result_fail("%s\n", test_case.name);
275 *failures += 1;
276 }
277 } else {
278 /*
279 * Comparing mremap time is only applicable if entire region
280 * was faulted in.
281 */
282 if (threshold_mb == VALIDATION_NO_THRESHOLD ||
283 test_case.config.region_size <= threshold_mb * _1MB)
284 ksft_test_result_pass("%s\n\tmremap time: %12lldns\n",
285 test_case.name, remap_time);
286 else
287 ksft_test_result_pass("%s\n", test_case.name);
288 }
289 }
290
usage(const char * cmd)291 static void usage(const char *cmd)
292 {
293 fprintf(stderr,
294 "Usage: %s [[-t <threshold_mb>] [-p <pattern_seed>]]\n"
295 "-t\t only validate threshold_mb of the remapped region\n"
296 " \t if 0 is supplied no threshold is used; all tests\n"
297 " \t are run and remapped regions validated fully.\n"
298 " \t The default threshold used is 4MB.\n"
299 "-p\t provide a seed to generate the random pattern for\n"
300 " \t validating the remapped region.\n", cmd);
301 }
302
parse_args(int argc,char ** argv,unsigned int * threshold_mb,unsigned int * pattern_seed)303 static int parse_args(int argc, char **argv, unsigned int *threshold_mb,
304 unsigned int *pattern_seed)
305 {
306 const char *optstr = "t:p:";
307 int opt;
308
309 while ((opt = getopt(argc, argv, optstr)) != -1) {
310 switch (opt) {
311 case 't':
312 *threshold_mb = atoi(optarg);
313 break;
314 case 'p':
315 *pattern_seed = atoi(optarg);
316 break;
317 default:
318 usage(argv[0]);
319 return -1;
320 }
321 }
322
323 if (optind < argc) {
324 usage(argv[0]);
325 return -1;
326 }
327
328 return 0;
329 }
330
331 #define MAX_TEST 13
332 #define MAX_PERF_TEST 3
main(int argc,char ** argv)333 int main(int argc, char **argv)
334 {
335 int failures = 0;
336 int i, run_perf_tests;
337 unsigned int threshold_mb = VALIDATION_DEFAULT_THRESHOLD;
338 unsigned int pattern_seed;
339 struct test test_cases[MAX_TEST];
340 struct test perf_test_cases[MAX_PERF_TEST];
341 int page_size;
342 time_t t;
343
344 pattern_seed = (unsigned int) time(&t);
345
346 if (parse_args(argc, argv, &threshold_mb, &pattern_seed) < 0)
347 exit(EXIT_FAILURE);
348
349 ksft_print_msg("Test configs:\n\tthreshold_mb=%u\n\tpattern_seed=%u\n\n",
350 threshold_mb, pattern_seed);
351
352 page_size = sysconf(_SC_PAGESIZE);
353
354 /* Expected mremap failures */
355 test_cases[0] = MAKE_TEST(page_size, page_size, page_size,
356 OVERLAPPING, EXPECT_FAILURE,
357 "mremap - Source and Destination Regions Overlapping");
358
359 test_cases[1] = MAKE_TEST(page_size, page_size/4, page_size,
360 NON_OVERLAPPING, EXPECT_FAILURE,
361 "mremap - Destination Address Misaligned (1KB-aligned)");
362 test_cases[2] = MAKE_TEST(page_size/4, page_size, page_size,
363 NON_OVERLAPPING, EXPECT_FAILURE,
364 "mremap - Source Address Misaligned (1KB-aligned)");
365
366 /* Src addr PTE aligned */
367 test_cases[3] = MAKE_TEST(PTE, PTE, PTE * 2,
368 NON_OVERLAPPING, EXPECT_SUCCESS,
369 "8KB mremap - Source PTE-aligned, Destination PTE-aligned");
370
371 /* Src addr 1MB aligned */
372 test_cases[4] = MAKE_TEST(_1MB, PTE, _2MB, NON_OVERLAPPING, EXPECT_SUCCESS,
373 "2MB mremap - Source 1MB-aligned, Destination PTE-aligned");
374 test_cases[5] = MAKE_TEST(_1MB, _1MB, _2MB, NON_OVERLAPPING, EXPECT_SUCCESS,
375 "2MB mremap - Source 1MB-aligned, Destination 1MB-aligned");
376
377 /* Src addr PMD aligned */
378 test_cases[6] = MAKE_TEST(PMD, PTE, _4MB, NON_OVERLAPPING, EXPECT_SUCCESS,
379 "4MB mremap - Source PMD-aligned, Destination PTE-aligned");
380 test_cases[7] = MAKE_TEST(PMD, _1MB, _4MB, NON_OVERLAPPING, EXPECT_SUCCESS,
381 "4MB mremap - Source PMD-aligned, Destination 1MB-aligned");
382 test_cases[8] = MAKE_TEST(PMD, PMD, _4MB, NON_OVERLAPPING, EXPECT_SUCCESS,
383 "4MB mremap - Source PMD-aligned, Destination PMD-aligned");
384
385 /* Src addr PUD aligned */
386 test_cases[9] = MAKE_TEST(PUD, PTE, _2GB, NON_OVERLAPPING, EXPECT_SUCCESS,
387 "2GB mremap - Source PUD-aligned, Destination PTE-aligned");
388 test_cases[10] = MAKE_TEST(PUD, _1MB, _2GB, NON_OVERLAPPING, EXPECT_SUCCESS,
389 "2GB mremap - Source PUD-aligned, Destination 1MB-aligned");
390 test_cases[11] = MAKE_TEST(PUD, PMD, _2GB, NON_OVERLAPPING, EXPECT_SUCCESS,
391 "2GB mremap - Source PUD-aligned, Destination PMD-aligned");
392 test_cases[12] = MAKE_TEST(PUD, PUD, _2GB, NON_OVERLAPPING, EXPECT_SUCCESS,
393 "2GB mremap - Source PUD-aligned, Destination PUD-aligned");
394
395 perf_test_cases[0] = MAKE_TEST(page_size, page_size, _1GB, NON_OVERLAPPING, EXPECT_SUCCESS,
396 "1GB mremap - Source PTE-aligned, Destination PTE-aligned");
397 /*
398 * mremap 1GB region - Page table level aligned time
399 * comparison.
400 */
401 perf_test_cases[1] = MAKE_TEST(PMD, PMD, _1GB, NON_OVERLAPPING, EXPECT_SUCCESS,
402 "1GB mremap - Source PMD-aligned, Destination PMD-aligned");
403 perf_test_cases[2] = MAKE_TEST(PUD, PUD, _1GB, NON_OVERLAPPING, EXPECT_SUCCESS,
404 "1GB mremap - Source PUD-aligned, Destination PUD-aligned");
405
406 run_perf_tests = (threshold_mb == VALIDATION_NO_THRESHOLD) ||
407 (threshold_mb * _1MB >= _1GB);
408
409 ksft_set_plan(ARRAY_SIZE(test_cases) + (run_perf_tests ?
410 ARRAY_SIZE(perf_test_cases) : 0));
411
412 for (i = 0; i < ARRAY_SIZE(test_cases); i++)
413 run_mremap_test_case(test_cases[i], &failures, threshold_mb,
414 pattern_seed);
415
416 if (run_perf_tests) {
417 ksft_print_msg("\n%s\n",
418 "mremap HAVE_MOVE_PMD/PUD optimization time comparison for 1GB region:");
419 for (i = 0; i < ARRAY_SIZE(perf_test_cases); i++)
420 run_mremap_test_case(perf_test_cases[i], &failures,
421 threshold_mb, pattern_seed);
422 }
423
424 if (failures > 0)
425 ksft_exit_fail();
426 else
427 ksft_exit_pass();
428 }
429