Andi Kleen | fcc1f2d | 2012-12-11 16:01:36 -0800 | [diff] [blame] | 1 | /* Test selecting other page sizes for mmap/shmget. |
| 2 | |
| 3 | Before running this huge pages for each huge page size must have been |
| 4 | reserved. |
| 5 | For large pages beyond MAX_ORDER (like 1GB on x86) boot options must be used. |
| 6 | Also shmmax must be increased. |
| 7 | And you need to run as root to work around some weird permissions in shm. |
| 8 | And nothing using huge pages should run in parallel. |
| 9 | When the program aborts you may need to clean up the shm segments with |
| 10 | ipcrm -m by hand, like this |
| 11 | sudo ipcs | awk '$1 == "0x00000000" {print $2}' | xargs -n1 sudo ipcrm -m |
| 12 | (warning this will remove all if someone else uses them) */ |
| 13 | |
| 14 | #define _GNU_SOURCE 1 |
| 15 | #include <sys/mman.h> |
| 16 | #include <stdlib.h> |
| 17 | #include <stdio.h> |
| 18 | #include <sys/ipc.h> |
| 19 | #include <sys/shm.h> |
| 20 | #include <sys/stat.h> |
| 21 | #include <glob.h> |
| 22 | #include <assert.h> |
| 23 | #include <unistd.h> |
| 24 | #include <stdarg.h> |
| 25 | #include <string.h> |
| 26 | |
| 27 | #define err(x) perror(x), exit(1) |
| 28 | |
| 29 | #define MAP_HUGE_2MB (21 << MAP_HUGE_SHIFT) |
| 30 | #define MAP_HUGE_1GB (30 << MAP_HUGE_SHIFT) |
| 31 | #define MAP_HUGE_SHIFT 26 |
| 32 | #define MAP_HUGE_MASK 0x3f |
| 33 | #define MAP_HUGETLB 0x40000 |
| 34 | |
| 35 | #define SHM_HUGETLB 04000 /* segment will use huge TLB pages */ |
| 36 | #define SHM_HUGE_SHIFT 26 |
| 37 | #define SHM_HUGE_MASK 0x3f |
| 38 | #define SHM_HUGE_2MB (21 << SHM_HUGE_SHIFT) |
| 39 | #define SHM_HUGE_1GB (30 << SHM_HUGE_SHIFT) |
| 40 | |
| 41 | #define NUM_PAGESIZES 5 |
| 42 | |
| 43 | #define NUM_PAGES 4 |
| 44 | |
| 45 | #define Dprintf(fmt...) // printf(fmt) |
| 46 | |
| 47 | unsigned long page_sizes[NUM_PAGESIZES]; |
| 48 | int num_page_sizes; |
| 49 | |
| 50 | int ilog2(unsigned long v) |
| 51 | { |
| 52 | int l = 0; |
| 53 | while ((1UL << l) < v) |
| 54 | l++; |
| 55 | return l; |
| 56 | } |
| 57 | |
| 58 | void find_pagesizes(void) |
| 59 | { |
| 60 | glob_t g; |
| 61 | int i; |
| 62 | glob("/sys/kernel/mm/hugepages/hugepages-*kB", 0, NULL, &g); |
| 63 | assert(g.gl_pathc <= NUM_PAGESIZES); |
| 64 | for (i = 0; i < g.gl_pathc; i++) { |
| 65 | sscanf(g.gl_pathv[i], "/sys/kernel/mm/hugepages/hugepages-%lukB", |
| 66 | &page_sizes[i]); |
| 67 | page_sizes[i] <<= 10; |
| 68 | printf("Found %luMB\n", page_sizes[i] >> 20); |
| 69 | } |
| 70 | num_page_sizes = g.gl_pathc; |
| 71 | globfree(&g); |
| 72 | } |
| 73 | |
| 74 | unsigned long default_huge_page_size(void) |
| 75 | { |
| 76 | unsigned long hps = 0; |
| 77 | char *line = NULL; |
| 78 | size_t linelen = 0; |
| 79 | FILE *f = fopen("/proc/meminfo", "r"); |
| 80 | if (!f) |
| 81 | return 0; |
| 82 | while (getline(&line, &linelen, f) > 0) { |
| 83 | if (sscanf(line, "Hugepagesize: %lu kB", &hps) == 1) { |
| 84 | hps <<= 10; |
| 85 | break; |
| 86 | } |
| 87 | } |
| 88 | free(line); |
| 89 | return hps; |
| 90 | } |
| 91 | |
| 92 | void show(unsigned long ps) |
| 93 | { |
| 94 | char buf[100]; |
| 95 | if (ps == getpagesize()) |
| 96 | return; |
| 97 | printf("%luMB: ", ps >> 20); |
| 98 | fflush(stdout); |
| 99 | snprintf(buf, sizeof buf, |
| 100 | "cat /sys/kernel/mm/hugepages/hugepages-%lukB/free_hugepages", |
| 101 | ps >> 10); |
| 102 | system(buf); |
| 103 | } |
| 104 | |
| 105 | unsigned long read_sysfs(int warn, char *fmt, ...) |
| 106 | { |
| 107 | char *line = NULL; |
| 108 | size_t linelen = 0; |
| 109 | char buf[100]; |
| 110 | FILE *f; |
| 111 | va_list ap; |
| 112 | unsigned long val = 0; |
| 113 | |
| 114 | va_start(ap, fmt); |
| 115 | vsnprintf(buf, sizeof buf, fmt, ap); |
| 116 | va_end(ap); |
| 117 | |
| 118 | f = fopen(buf, "r"); |
| 119 | if (!f) { |
| 120 | if (warn) |
| 121 | printf("missing %s\n", buf); |
| 122 | return 0; |
| 123 | } |
| 124 | if (getline(&line, &linelen, f) > 0) { |
| 125 | sscanf(line, "%lu", &val); |
| 126 | } |
| 127 | fclose(f); |
| 128 | free(line); |
| 129 | return val; |
| 130 | } |
| 131 | |
| 132 | unsigned long read_free(unsigned long ps) |
| 133 | { |
| 134 | return read_sysfs(ps != getpagesize(), |
| 135 | "/sys/kernel/mm/hugepages/hugepages-%lukB/free_hugepages", |
| 136 | ps >> 10); |
| 137 | } |
| 138 | |
| 139 | void test_mmap(unsigned long size, unsigned flags) |
| 140 | { |
| 141 | char *map; |
| 142 | unsigned long before, after; |
| 143 | int err; |
| 144 | |
| 145 | before = read_free(size); |
| 146 | map = mmap(NULL, size*NUM_PAGES, PROT_READ|PROT_WRITE, |
| 147 | MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB|flags, 0, 0); |
| 148 | |
| 149 | if (map == (char *)-1) err("mmap"); |
| 150 | memset(map, 0xff, size*NUM_PAGES); |
| 151 | after = read_free(size); |
| 152 | Dprintf("before %lu after %lu diff %ld size %lu\n", |
| 153 | before, after, before - after, size); |
| 154 | assert(size == getpagesize() || (before - after) == NUM_PAGES); |
| 155 | show(size); |
| 156 | err = munmap(map, size); |
| 157 | assert(!err); |
| 158 | } |
| 159 | |
| 160 | void test_shmget(unsigned long size, unsigned flags) |
| 161 | { |
| 162 | int id; |
| 163 | unsigned long before, after; |
| 164 | int err; |
| 165 | |
| 166 | before = read_free(size); |
| 167 | id = shmget(IPC_PRIVATE, size * NUM_PAGES, IPC_CREAT|0600|flags); |
| 168 | if (id < 0) err("shmget"); |
| 169 | |
| 170 | struct shm_info i; |
| 171 | if (shmctl(id, SHM_INFO, (void *)&i) < 0) err("shmctl"); |
| 172 | Dprintf("alloc %lu res %lu\n", i.shm_tot, i.shm_rss); |
| 173 | |
| 174 | |
| 175 | Dprintf("id %d\n", id); |
| 176 | char *map = shmat(id, NULL, 0600); |
| 177 | if (map == (char*)-1) err("shmat"); |
| 178 | |
| 179 | shmctl(id, IPC_RMID, NULL); |
| 180 | |
| 181 | memset(map, 0xff, size*NUM_PAGES); |
| 182 | after = read_free(size); |
| 183 | |
| 184 | Dprintf("before %lu after %lu diff %ld size %lu\n", |
| 185 | before, after, before - after, size); |
| 186 | assert(size == getpagesize() || (before - after) == NUM_PAGES); |
| 187 | show(size); |
| 188 | err = shmdt(map); |
| 189 | assert(!err); |
| 190 | } |
| 191 | |
| 192 | void sanity_checks(void) |
| 193 | { |
| 194 | int i; |
| 195 | unsigned long largest = getpagesize(); |
| 196 | |
| 197 | for (i = 0; i < num_page_sizes; i++) { |
| 198 | if (page_sizes[i] > largest) |
| 199 | largest = page_sizes[i]; |
| 200 | |
| 201 | if (read_free(page_sizes[i]) < NUM_PAGES) { |
| 202 | printf("Not enough huge pages for page size %lu MB, need %u\n", |
| 203 | page_sizes[i] >> 20, |
| 204 | NUM_PAGES); |
| 205 | exit(0); |
| 206 | } |
| 207 | } |
| 208 | |
| 209 | if (read_sysfs(0, "/proc/sys/kernel/shmmax") < NUM_PAGES * largest) { |
| 210 | printf("Please do echo %lu > /proc/sys/kernel/shmmax", largest * NUM_PAGES); |
| 211 | exit(0); |
| 212 | } |
| 213 | |
| 214 | #if defined(__x86_64__) |
| 215 | if (largest != 1U<<30) { |
| 216 | printf("No GB pages available on x86-64\n" |
| 217 | "Please boot with hugepagesz=1G hugepages=%d\n", NUM_PAGES); |
| 218 | exit(0); |
| 219 | } |
| 220 | #endif |
| 221 | } |
| 222 | |
| 223 | int main(void) |
| 224 | { |
| 225 | int i; |
| 226 | unsigned default_hps = default_huge_page_size(); |
| 227 | |
| 228 | find_pagesizes(); |
| 229 | |
| 230 | sanity_checks(); |
| 231 | |
| 232 | for (i = 0; i < num_page_sizes; i++) { |
| 233 | unsigned long ps = page_sizes[i]; |
| 234 | int arg = ilog2(ps) << MAP_HUGE_SHIFT; |
| 235 | printf("Testing %luMB mmap with shift %x\n", ps >> 20, arg); |
| 236 | test_mmap(ps, MAP_HUGETLB | arg); |
| 237 | } |
| 238 | printf("Testing default huge mmap\n"); |
| 239 | test_mmap(default_hps, SHM_HUGETLB); |
| 240 | |
| 241 | puts("Testing non-huge shmget"); |
| 242 | test_shmget(getpagesize(), 0); |
| 243 | |
| 244 | for (i = 0; i < num_page_sizes; i++) { |
| 245 | unsigned long ps = page_sizes[i]; |
| 246 | int arg = ilog2(ps) << SHM_HUGE_SHIFT; |
| 247 | printf("Testing %luMB shmget with shift %x\n", ps >> 20, arg); |
| 248 | test_shmget(ps, SHM_HUGETLB | arg); |
| 249 | } |
| 250 | puts("default huge shmget"); |
| 251 | test_shmget(default_hps, SHM_HUGETLB); |
| 252 | |
| 253 | return 0; |
| 254 | } |