Commit | Line | Data |
---|---|---|
3f21a6b7 URS |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
3 | /* | |
4 | * Test module for stress and analyze performance of vmalloc allocator. | |
5 | * (C) 2018 Uladzislau Rezki (Sony) <urezki@gmail.com> | |
6 | */ | |
7 | #include <linux/init.h> | |
8 | #include <linux/kernel.h> | |
9 | #include <linux/module.h> | |
10 | #include <linux/vmalloc.h> | |
11 | #include <linux/random.h> | |
12 | #include <linux/kthread.h> | |
13 | #include <linux/moduleparam.h> | |
14 | #include <linux/completion.h> | |
15 | #include <linux/delay.h> | |
16 | #include <linux/rwsem.h> | |
17 | #include <linux/mm.h> | |
da4fc00a URS |
18 | #include <linux/rcupdate.h> |
19 | #include <linux/slab.h> | |
3f21a6b7 URS |
20 | |
21 | #define __param(type, name, init, msg) \ | |
22 | static type name = init; \ | |
23 | module_param(name, type, 0444); \ | |
24 | MODULE_PARM_DESC(name, msg) \ | |
25 | ||
80f47599 URS |
26 | __param(int, nr_threads, 0, |
27 | "Number of workers to perform tests(min: 1 max: USHRT_MAX)"); | |
3f21a6b7 URS |
28 | |
29 | __param(bool, sequential_test_order, false, | |
30 | "Use sequential stress tests order"); | |
31 | ||
32 | __param(int, test_repeat_count, 1, | |
33 | "Set test repeat counter"); | |
34 | ||
35 | __param(int, test_loop_count, 1000000, | |
36 | "Set test loop counter"); | |
37 | ||
f8bcbecf URS |
38 | __param(int, nr_pages, 0, |
39 | "Set number of pages for fix_size_alloc_test(default: 1)"); | |
40 | ||
3f21a6b7 URS |
41 | __param(int, run_test_mask, INT_MAX, |
42 | "Set tests specified in the mask.\n\n" | |
da4fc00a URS |
43 | "\t\tid: 1, name: fix_size_alloc_test\n" |
44 | "\t\tid: 2, name: full_fit_alloc_test\n" | |
45 | "\t\tid: 4, name: long_busy_list_alloc_test\n" | |
46 | "\t\tid: 8, name: random_size_alloc_test\n" | |
47 | "\t\tid: 16, name: fix_align_alloc_test\n" | |
48 | "\t\tid: 32, name: random_size_align_alloc_test\n" | |
49 | "\t\tid: 64, name: align_shift_alloc_test\n" | |
50 | "\t\tid: 128, name: pcpu_alloc_test\n" | |
51 | "\t\tid: 256, name: kvfree_rcu_1_arg_vmalloc_test\n" | |
52 | "\t\tid: 512, name: kvfree_rcu_2_arg_vmalloc_test\n" | |
3f21a6b7 URS |
53 | /* Add a new test case description here. */ |
54 | ); | |
55 | ||
3f21a6b7 URS |
56 | /* |
57 | * Read write semaphore for synchronization of setup | |
58 | * phase that is done in main thread and workers. | |
59 | */ | |
60 | static DECLARE_RWSEM(prepare_for_test_rwsem); | |
61 | ||
62 | /* | |
63 | * Completion tracking for worker threads. | |
64 | */ | |
65 | static DECLARE_COMPLETION(test_all_done_comp); | |
66 | static atomic_t test_n_undone = ATOMIC_INIT(0); | |
67 | ||
68 | static inline void | |
69 | test_report_one_done(void) | |
70 | { | |
71 | if (atomic_dec_and_test(&test_n_undone)) | |
72 | complete(&test_all_done_comp); | |
73 | } | |
74 | ||
75 | static int random_size_align_alloc_test(void) | |
76 | { | |
77 | unsigned long size, align, rnd; | |
78 | void *ptr; | |
79 | int i; | |
80 | ||
81 | for (i = 0; i < test_loop_count; i++) { | |
82 | get_random_bytes(&rnd, sizeof(rnd)); | |
83 | ||
84 | /* | |
85 | * Maximum 1024 pages, if PAGE_SIZE is 4096. | |
86 | */ | |
87 | align = 1 << (rnd % 23); | |
88 | ||
89 | /* | |
90 | * Maximum 10 pages. | |
91 | */ | |
92 | size = ((rnd % 10) + 1) * PAGE_SIZE; | |
93 | ||
c3f896dc CH |
94 | ptr = __vmalloc_node(size, align, GFP_KERNEL | __GFP_ZERO, 0, |
95 | __builtin_return_address(0)); | |
3f21a6b7 URS |
96 | if (!ptr) |
97 | return -1; | |
98 | ||
99 | vfree(ptr); | |
100 | } | |
101 | ||
102 | return 0; | |
103 | } | |
104 | ||
105 | /* | |
106 | * This test case is supposed to be failed. | |
107 | */ | |
108 | static int align_shift_alloc_test(void) | |
109 | { | |
110 | unsigned long align; | |
111 | void *ptr; | |
112 | int i; | |
113 | ||
114 | for (i = 0; i < BITS_PER_LONG; i++) { | |
115 | align = ((unsigned long) 1) << i; | |
116 | ||
c3f896dc CH |
117 | ptr = __vmalloc_node(PAGE_SIZE, align, GFP_KERNEL|__GFP_ZERO, 0, |
118 | __builtin_return_address(0)); | |
3f21a6b7 URS |
119 | if (!ptr) |
120 | return -1; | |
121 | ||
122 | vfree(ptr); | |
123 | } | |
124 | ||
125 | return 0; | |
126 | } | |
127 | ||
128 | static int fix_align_alloc_test(void) | |
129 | { | |
130 | void *ptr; | |
131 | int i; | |
132 | ||
133 | for (i = 0; i < test_loop_count; i++) { | |
c3f896dc CH |
134 | ptr = __vmalloc_node(5 * PAGE_SIZE, THREAD_ALIGN << 1, |
135 | GFP_KERNEL | __GFP_ZERO, 0, | |
136 | __builtin_return_address(0)); | |
3f21a6b7 URS |
137 | if (!ptr) |
138 | return -1; | |
139 | ||
140 | vfree(ptr); | |
141 | } | |
142 | ||
143 | return 0; | |
144 | } | |
145 | ||
146 | static int random_size_alloc_test(void) | |
147 | { | |
148 | unsigned int n; | |
149 | void *p; | |
150 | int i; | |
151 | ||
152 | for (i = 0; i < test_loop_count; i++) { | |
153 | get_random_bytes(&n, sizeof(i)); | |
154 | n = (n % 100) + 1; | |
155 | ||
156 | p = vmalloc(n * PAGE_SIZE); | |
157 | ||
158 | if (!p) | |
159 | return -1; | |
160 | ||
161 | *((__u8 *)p) = 1; | |
162 | vfree(p); | |
163 | } | |
164 | ||
165 | return 0; | |
166 | } | |
167 | ||
168 | static int long_busy_list_alloc_test(void) | |
169 | { | |
170 | void *ptr_1, *ptr_2; | |
171 | void **ptr; | |
172 | int rv = -1; | |
173 | int i; | |
174 | ||
175 | ptr = vmalloc(sizeof(void *) * 15000); | |
176 | if (!ptr) | |
177 | return rv; | |
178 | ||
179 | for (i = 0; i < 15000; i++) | |
180 | ptr[i] = vmalloc(1 * PAGE_SIZE); | |
181 | ||
182 | for (i = 0; i < test_loop_count; i++) { | |
183 | ptr_1 = vmalloc(100 * PAGE_SIZE); | |
184 | if (!ptr_1) | |
185 | goto leave; | |
186 | ||
187 | ptr_2 = vmalloc(1 * PAGE_SIZE); | |
188 | if (!ptr_2) { | |
189 | vfree(ptr_1); | |
190 | goto leave; | |
191 | } | |
192 | ||
193 | *((__u8 *)ptr_1) = 0; | |
194 | *((__u8 *)ptr_2) = 1; | |
195 | ||
196 | vfree(ptr_1); | |
197 | vfree(ptr_2); | |
198 | } | |
199 | ||
200 | /* Success */ | |
201 | rv = 0; | |
202 | ||
203 | leave: | |
204 | for (i = 0; i < 15000; i++) | |
205 | vfree(ptr[i]); | |
206 | ||
207 | vfree(ptr); | |
208 | return rv; | |
209 | } | |
210 | ||
211 | static int full_fit_alloc_test(void) | |
212 | { | |
213 | void **ptr, **junk_ptr, *tmp; | |
214 | int junk_length; | |
215 | int rv = -1; | |
216 | int i; | |
217 | ||
218 | junk_length = fls(num_online_cpus()); | |
219 | junk_length *= (32 * 1024 * 1024 / PAGE_SIZE); | |
220 | ||
221 | ptr = vmalloc(sizeof(void *) * junk_length); | |
222 | if (!ptr) | |
223 | return rv; | |
224 | ||
225 | junk_ptr = vmalloc(sizeof(void *) * junk_length); | |
226 | if (!junk_ptr) { | |
227 | vfree(ptr); | |
228 | return rv; | |
229 | } | |
230 | ||
231 | for (i = 0; i < junk_length; i++) { | |
232 | ptr[i] = vmalloc(1 * PAGE_SIZE); | |
233 | junk_ptr[i] = vmalloc(1 * PAGE_SIZE); | |
234 | } | |
235 | ||
236 | for (i = 0; i < junk_length; i++) | |
237 | vfree(junk_ptr[i]); | |
238 | ||
239 | for (i = 0; i < test_loop_count; i++) { | |
240 | tmp = vmalloc(1 * PAGE_SIZE); | |
241 | ||
242 | if (!tmp) | |
243 | goto error; | |
244 | ||
245 | *((__u8 *)tmp) = 1; | |
246 | vfree(tmp); | |
247 | } | |
248 | ||
249 | /* Success */ | |
250 | rv = 0; | |
251 | ||
252 | error: | |
253 | for (i = 0; i < junk_length; i++) | |
254 | vfree(ptr[i]); | |
255 | ||
256 | vfree(ptr); | |
257 | vfree(junk_ptr); | |
258 | ||
259 | return rv; | |
260 | } | |
261 | ||
262 | static int fix_size_alloc_test(void) | |
263 | { | |
264 | void *ptr; | |
265 | int i; | |
266 | ||
267 | for (i = 0; i < test_loop_count; i++) { | |
f8bcbecf | 268 | ptr = vmalloc((nr_pages > 0 ? nr_pages:1) * PAGE_SIZE); |
3f21a6b7 URS |
269 | |
270 | if (!ptr) | |
271 | return -1; | |
272 | ||
273 | *((__u8 *)ptr) = 0; | |
274 | ||
275 | vfree(ptr); | |
276 | } | |
277 | ||
278 | return 0; | |
279 | } | |
280 | ||
281 | static int | |
282 | pcpu_alloc_test(void) | |
283 | { | |
284 | int rv = 0; | |
285 | #ifndef CONFIG_NEED_PER_CPU_KM | |
286 | void __percpu **pcpu; | |
287 | size_t size, align; | |
288 | int i; | |
289 | ||
290 | pcpu = vmalloc(sizeof(void __percpu *) * 35000); | |
291 | if (!pcpu) | |
292 | return -1; | |
293 | ||
294 | for (i = 0; i < 35000; i++) { | |
295 | unsigned int r; | |
296 | ||
297 | get_random_bytes(&r, sizeof(i)); | |
298 | size = (r % (PAGE_SIZE / 4)) + 1; | |
299 | ||
300 | /* | |
301 | * Maximum PAGE_SIZE | |
302 | */ | |
303 | get_random_bytes(&r, sizeof(i)); | |
304 | align = 1 << ((i % 11) + 1); | |
305 | ||
306 | pcpu[i] = __alloc_percpu(size, align); | |
307 | if (!pcpu[i]) | |
308 | rv = -1; | |
309 | } | |
310 | ||
311 | for (i = 0; i < 35000; i++) | |
312 | free_percpu(pcpu[i]); | |
313 | ||
314 | vfree(pcpu); | |
315 | #endif | |
316 | return rv; | |
317 | } | |
318 | ||
da4fc00a URS |
319 | struct test_kvfree_rcu { |
320 | struct rcu_head rcu; | |
321 | unsigned char array[20]; | |
322 | }; | |
323 | ||
324 | static int | |
325 | kvfree_rcu_1_arg_vmalloc_test(void) | |
326 | { | |
327 | struct test_kvfree_rcu *p; | |
328 | int i; | |
329 | ||
330 | for (i = 0; i < test_loop_count; i++) { | |
331 | p = vmalloc(1 * PAGE_SIZE); | |
332 | if (!p) | |
333 | return -1; | |
334 | ||
335 | p->array[0] = 'a'; | |
336 | kvfree_rcu(p); | |
337 | } | |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
342 | static int | |
343 | kvfree_rcu_2_arg_vmalloc_test(void) | |
344 | { | |
345 | struct test_kvfree_rcu *p; | |
346 | int i; | |
347 | ||
348 | for (i = 0; i < test_loop_count; i++) { | |
349 | p = vmalloc(1 * PAGE_SIZE); | |
350 | if (!p) | |
351 | return -1; | |
352 | ||
353 | p->array[0] = 'a'; | |
354 | kvfree_rcu(p, rcu); | |
355 | } | |
356 | ||
357 | return 0; | |
358 | } | |
359 | ||
3f21a6b7 URS |
360 | struct test_case_desc { |
361 | const char *test_name; | |
362 | int (*test_func)(void); | |
363 | }; | |
364 | ||
365 | static struct test_case_desc test_case_array[] = { | |
366 | { "fix_size_alloc_test", fix_size_alloc_test }, | |
367 | { "full_fit_alloc_test", full_fit_alloc_test }, | |
368 | { "long_busy_list_alloc_test", long_busy_list_alloc_test }, | |
369 | { "random_size_alloc_test", random_size_alloc_test }, | |
370 | { "fix_align_alloc_test", fix_align_alloc_test }, | |
371 | { "random_size_align_alloc_test", random_size_align_alloc_test }, | |
372 | { "align_shift_alloc_test", align_shift_alloc_test }, | |
373 | { "pcpu_alloc_test", pcpu_alloc_test }, | |
da4fc00a URS |
374 | { "kvfree_rcu_1_arg_vmalloc_test", kvfree_rcu_1_arg_vmalloc_test }, |
375 | { "kvfree_rcu_2_arg_vmalloc_test", kvfree_rcu_2_arg_vmalloc_test }, | |
3f21a6b7 URS |
376 | /* Add a new test case here. */ |
377 | }; | |
378 | ||
379 | struct test_case_data { | |
380 | int test_failed; | |
381 | int test_passed; | |
382 | u64 time; | |
383 | }; | |
384 | ||
3f21a6b7 URS |
385 | static struct test_driver { |
386 | struct task_struct *task; | |
80f47599 URS |
387 | struct test_case_data data[ARRAY_SIZE(test_case_array)]; |
388 | ||
3f21a6b7 URS |
389 | unsigned long start; |
390 | unsigned long stop; | |
80f47599 | 391 | } *tdriver; |
3f21a6b7 URS |
392 | |
393 | static void shuffle_array(int *arr, int n) | |
394 | { | |
395 | unsigned int rnd; | |
396 | int i, j, x; | |
397 | ||
398 | for (i = n - 1; i > 0; i--) { | |
399 | get_random_bytes(&rnd, sizeof(rnd)); | |
400 | ||
401 | /* Cut the range. */ | |
402 | j = rnd % i; | |
403 | ||
404 | /* Swap indexes. */ | |
405 | x = arr[i]; | |
406 | arr[i] = arr[j]; | |
407 | arr[j] = x; | |
408 | } | |
409 | } | |
410 | ||
411 | static int test_func(void *private) | |
412 | { | |
413 | struct test_driver *t = private; | |
3f21a6b7 | 414 | int random_array[ARRAY_SIZE(test_case_array)]; |
7507c402 | 415 | int index, i, j; |
3f21a6b7 URS |
416 | ktime_t kt; |
417 | u64 delta; | |
418 | ||
3f21a6b7 URS |
419 | for (i = 0; i < ARRAY_SIZE(test_case_array); i++) |
420 | random_array[i] = i; | |
421 | ||
422 | if (!sequential_test_order) | |
423 | shuffle_array(random_array, ARRAY_SIZE(test_case_array)); | |
424 | ||
425 | /* | |
426 | * Block until initialization is done. | |
427 | */ | |
428 | down_read(&prepare_for_test_rwsem); | |
429 | ||
430 | t->start = get_cycles(); | |
431 | for (i = 0; i < ARRAY_SIZE(test_case_array); i++) { | |
432 | index = random_array[i]; | |
433 | ||
434 | /* | |
435 | * Skip tests if run_test_mask has been specified. | |
436 | */ | |
437 | if (!((run_test_mask & (1 << index)) >> index)) | |
438 | continue; | |
439 | ||
440 | kt = ktime_get(); | |
441 | for (j = 0; j < test_repeat_count; j++) { | |
7507c402 | 442 | if (!test_case_array[index].test_func()) |
80f47599 | 443 | t->data[index].test_passed++; |
3f21a6b7 | 444 | else |
80f47599 | 445 | t->data[index].test_failed++; |
3f21a6b7 URS |
446 | } |
447 | ||
448 | /* | |
449 | * Take an average time that test took. | |
450 | */ | |
451 | delta = (u64) ktime_us_delta(ktime_get(), kt); | |
452 | do_div(delta, (u32) test_repeat_count); | |
453 | ||
80f47599 | 454 | t->data[index].time = delta; |
3f21a6b7 URS |
455 | } |
456 | t->stop = get_cycles(); | |
457 | ||
458 | up_read(&prepare_for_test_rwsem); | |
459 | test_report_one_done(); | |
460 | ||
461 | /* | |
462 | * Wait for the kthread_stop() call. | |
463 | */ | |
464 | while (!kthread_should_stop()) | |
465 | msleep(10); | |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
80f47599 | 470 | static int |
3f21a6b7 URS |
471 | init_test_configurtion(void) |
472 | { | |
473 | /* | |
80f47599 URS |
474 | * A maximum number of workers is defined as hard-coded |
475 | * value and set to USHRT_MAX. We add such gap just in | |
476 | * case and for potential heavy stressing. | |
3f21a6b7 | 477 | */ |
80f47599 | 478 | nr_threads = clamp(nr_threads, 1, (int) USHRT_MAX); |
3f21a6b7 | 479 | |
80f47599 URS |
480 | /* Allocate the space for test instances. */ |
481 | tdriver = kvcalloc(nr_threads, sizeof(*tdriver), GFP_KERNEL); | |
482 | if (tdriver == NULL) | |
483 | return -1; | |
3f21a6b7 URS |
484 | |
485 | if (test_repeat_count <= 0) | |
486 | test_repeat_count = 1; | |
487 | ||
488 | if (test_loop_count <= 0) | |
489 | test_loop_count = 1; | |
80f47599 URS |
490 | |
491 | return 0; | |
3f21a6b7 URS |
492 | } |
493 | ||
494 | static void do_concurrent_test(void) | |
495 | { | |
80f47599 | 496 | int i, ret; |
3f21a6b7 URS |
497 | |
498 | /* | |
499 | * Set some basic configurations plus sanity check. | |
500 | */ | |
80f47599 URS |
501 | ret = init_test_configurtion(); |
502 | if (ret < 0) | |
503 | return; | |
3f21a6b7 URS |
504 | |
505 | /* | |
506 | * Put on hold all workers. | |
507 | */ | |
508 | down_write(&prepare_for_test_rwsem); | |
509 | ||
80f47599 URS |
510 | for (i = 0; i < nr_threads; i++) { |
511 | struct test_driver *t = &tdriver[i]; | |
3f21a6b7 | 512 | |
80f47599 | 513 | t->task = kthread_run(test_func, t, "vmalloc_test/%d", i); |
3f21a6b7 URS |
514 | |
515 | if (!IS_ERR(t->task)) | |
516 | /* Success. */ | |
517 | atomic_inc(&test_n_undone); | |
518 | else | |
80f47599 | 519 | pr_err("Failed to start %d kthread\n", i); |
3f21a6b7 URS |
520 | } |
521 | ||
522 | /* | |
523 | * Now let the workers do their job. | |
524 | */ | |
525 | up_write(&prepare_for_test_rwsem); | |
526 | ||
527 | /* | |
528 | * Sleep quiet until all workers are done with 1 second | |
529 | * interval. Since the test can take a lot of time we | |
530 | * can run into a stack trace of the hung task. That is | |
531 | * why we go with completion_timeout and HZ value. | |
532 | */ | |
533 | do { | |
534 | ret = wait_for_completion_timeout(&test_all_done_comp, HZ); | |
535 | } while (!ret); | |
536 | ||
80f47599 URS |
537 | for (i = 0; i < nr_threads; i++) { |
538 | struct test_driver *t = &tdriver[i]; | |
539 | int j; | |
3f21a6b7 URS |
540 | |
541 | if (!IS_ERR(t->task)) | |
542 | kthread_stop(t->task); | |
543 | ||
80f47599 URS |
544 | for (j = 0; j < ARRAY_SIZE(test_case_array); j++) { |
545 | if (!((run_test_mask & (1 << j)) >> j)) | |
3f21a6b7 URS |
546 | continue; |
547 | ||
548 | pr_info( | |
549 | "Summary: %s passed: %d failed: %d repeat: %d loops: %d avg: %llu usec\n", | |
80f47599 URS |
550 | test_case_array[j].test_name, |
551 | t->data[j].test_passed, | |
552 | t->data[j].test_failed, | |
3f21a6b7 | 553 | test_repeat_count, test_loop_count, |
80f47599 | 554 | t->data[j].time); |
3f21a6b7 URS |
555 | } |
556 | ||
80f47599 URS |
557 | pr_info("All test took worker%d=%lu cycles\n", |
558 | i, t->stop - t->start); | |
3f21a6b7 | 559 | } |
80f47599 URS |
560 | |
561 | kvfree(tdriver); | |
3f21a6b7 URS |
562 | } |
563 | ||
564 | static int vmalloc_test_init(void) | |
565 | { | |
566 | do_concurrent_test(); | |
567 | return -EAGAIN; /* Fail will directly unload the module */ | |
568 | } | |
569 | ||
570 | static void vmalloc_test_exit(void) | |
571 | { | |
572 | } | |
573 | ||
574 | module_init(vmalloc_test_init) | |
575 | module_exit(vmalloc_test_exit) | |
576 | ||
577 | MODULE_LICENSE("GPL"); | |
578 | MODULE_AUTHOR("Uladzislau Rezki"); | |
579 | MODULE_DESCRIPTION("vmalloc test module"); |