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 | { | |
5e21f2d5 URS |
77 | unsigned long size, align; |
78 | unsigned int rnd; | |
3f21a6b7 URS |
79 | void *ptr; |
80 | int i; | |
81 | ||
82 | for (i = 0; i < test_loop_count; i++) { | |
5e21f2d5 | 83 | rnd = prandom_u32(); |
3f21a6b7 URS |
84 | |
85 | /* | |
86 | * Maximum 1024 pages, if PAGE_SIZE is 4096. | |
87 | */ | |
88 | align = 1 << (rnd % 23); | |
89 | ||
90 | /* | |
91 | * Maximum 10 pages. | |
92 | */ | |
93 | size = ((rnd % 10) + 1) * PAGE_SIZE; | |
94 | ||
c3f896dc CH |
95 | ptr = __vmalloc_node(size, align, GFP_KERNEL | __GFP_ZERO, 0, |
96 | __builtin_return_address(0)); | |
3f21a6b7 URS |
97 | if (!ptr) |
98 | return -1; | |
99 | ||
100 | vfree(ptr); | |
101 | } | |
102 | ||
103 | return 0; | |
104 | } | |
105 | ||
106 | /* | |
107 | * This test case is supposed to be failed. | |
108 | */ | |
109 | static int align_shift_alloc_test(void) | |
110 | { | |
111 | unsigned long align; | |
112 | void *ptr; | |
113 | int i; | |
114 | ||
115 | for (i = 0; i < BITS_PER_LONG; i++) { | |
116 | align = ((unsigned long) 1) << i; | |
117 | ||
c3f896dc CH |
118 | ptr = __vmalloc_node(PAGE_SIZE, align, GFP_KERNEL|__GFP_ZERO, 0, |
119 | __builtin_return_address(0)); | |
3f21a6b7 URS |
120 | if (!ptr) |
121 | return -1; | |
122 | ||
123 | vfree(ptr); | |
124 | } | |
125 | ||
126 | return 0; | |
127 | } | |
128 | ||
129 | static int fix_align_alloc_test(void) | |
130 | { | |
131 | void *ptr; | |
132 | int i; | |
133 | ||
134 | for (i = 0; i < test_loop_count; i++) { | |
c3f896dc CH |
135 | ptr = __vmalloc_node(5 * PAGE_SIZE, THREAD_ALIGN << 1, |
136 | GFP_KERNEL | __GFP_ZERO, 0, | |
137 | __builtin_return_address(0)); | |
3f21a6b7 URS |
138 | if (!ptr) |
139 | return -1; | |
140 | ||
141 | vfree(ptr); | |
142 | } | |
143 | ||
144 | return 0; | |
145 | } | |
146 | ||
147 | static int random_size_alloc_test(void) | |
148 | { | |
149 | unsigned int n; | |
150 | void *p; | |
151 | int i; | |
152 | ||
153 | for (i = 0; i < test_loop_count; i++) { | |
5e21f2d5 | 154 | n = prandom_u32(); |
3f21a6b7 URS |
155 | n = (n % 100) + 1; |
156 | ||
157 | p = vmalloc(n * PAGE_SIZE); | |
158 | ||
159 | if (!p) | |
160 | return -1; | |
161 | ||
162 | *((__u8 *)p) = 1; | |
163 | vfree(p); | |
164 | } | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
169 | static int long_busy_list_alloc_test(void) | |
170 | { | |
171 | void *ptr_1, *ptr_2; | |
172 | void **ptr; | |
173 | int rv = -1; | |
174 | int i; | |
175 | ||
176 | ptr = vmalloc(sizeof(void *) * 15000); | |
177 | if (!ptr) | |
178 | return rv; | |
179 | ||
180 | for (i = 0; i < 15000; i++) | |
181 | ptr[i] = vmalloc(1 * PAGE_SIZE); | |
182 | ||
183 | for (i = 0; i < test_loop_count; i++) { | |
184 | ptr_1 = vmalloc(100 * PAGE_SIZE); | |
185 | if (!ptr_1) | |
186 | goto leave; | |
187 | ||
188 | ptr_2 = vmalloc(1 * PAGE_SIZE); | |
189 | if (!ptr_2) { | |
190 | vfree(ptr_1); | |
191 | goto leave; | |
192 | } | |
193 | ||
194 | *((__u8 *)ptr_1) = 0; | |
195 | *((__u8 *)ptr_2) = 1; | |
196 | ||
197 | vfree(ptr_1); | |
198 | vfree(ptr_2); | |
199 | } | |
200 | ||
201 | /* Success */ | |
202 | rv = 0; | |
203 | ||
204 | leave: | |
205 | for (i = 0; i < 15000; i++) | |
206 | vfree(ptr[i]); | |
207 | ||
208 | vfree(ptr); | |
209 | return rv; | |
210 | } | |
211 | ||
212 | static int full_fit_alloc_test(void) | |
213 | { | |
214 | void **ptr, **junk_ptr, *tmp; | |
215 | int junk_length; | |
216 | int rv = -1; | |
217 | int i; | |
218 | ||
219 | junk_length = fls(num_online_cpus()); | |
220 | junk_length *= (32 * 1024 * 1024 / PAGE_SIZE); | |
221 | ||
222 | ptr = vmalloc(sizeof(void *) * junk_length); | |
223 | if (!ptr) | |
224 | return rv; | |
225 | ||
226 | junk_ptr = vmalloc(sizeof(void *) * junk_length); | |
227 | if (!junk_ptr) { | |
228 | vfree(ptr); | |
229 | return rv; | |
230 | } | |
231 | ||
232 | for (i = 0; i < junk_length; i++) { | |
233 | ptr[i] = vmalloc(1 * PAGE_SIZE); | |
234 | junk_ptr[i] = vmalloc(1 * PAGE_SIZE); | |
235 | } | |
236 | ||
237 | for (i = 0; i < junk_length; i++) | |
238 | vfree(junk_ptr[i]); | |
239 | ||
240 | for (i = 0; i < test_loop_count; i++) { | |
241 | tmp = vmalloc(1 * PAGE_SIZE); | |
242 | ||
243 | if (!tmp) | |
244 | goto error; | |
245 | ||
246 | *((__u8 *)tmp) = 1; | |
247 | vfree(tmp); | |
248 | } | |
249 | ||
250 | /* Success */ | |
251 | rv = 0; | |
252 | ||
253 | error: | |
254 | for (i = 0; i < junk_length; i++) | |
255 | vfree(ptr[i]); | |
256 | ||
257 | vfree(ptr); | |
258 | vfree(junk_ptr); | |
259 | ||
260 | return rv; | |
261 | } | |
262 | ||
263 | static int fix_size_alloc_test(void) | |
264 | { | |
265 | void *ptr; | |
266 | int i; | |
267 | ||
268 | for (i = 0; i < test_loop_count; i++) { | |
f8bcbecf | 269 | ptr = vmalloc((nr_pages > 0 ? nr_pages:1) * PAGE_SIZE); |
3f21a6b7 URS |
270 | |
271 | if (!ptr) | |
272 | return -1; | |
273 | ||
274 | *((__u8 *)ptr) = 0; | |
275 | ||
276 | vfree(ptr); | |
277 | } | |
278 | ||
279 | return 0; | |
280 | } | |
281 | ||
282 | static int | |
283 | pcpu_alloc_test(void) | |
284 | { | |
285 | int rv = 0; | |
286 | #ifndef CONFIG_NEED_PER_CPU_KM | |
287 | void __percpu **pcpu; | |
288 | size_t size, align; | |
289 | int i; | |
290 | ||
291 | pcpu = vmalloc(sizeof(void __percpu *) * 35000); | |
292 | if (!pcpu) | |
293 | return -1; | |
294 | ||
295 | for (i = 0; i < 35000; i++) { | |
296 | unsigned int r; | |
297 | ||
5e21f2d5 | 298 | r = prandom_u32(); |
3f21a6b7 URS |
299 | size = (r % (PAGE_SIZE / 4)) + 1; |
300 | ||
301 | /* | |
302 | * Maximum PAGE_SIZE | |
303 | */ | |
5e21f2d5 URS |
304 | r = prandom_u32(); |
305 | align = 1 << ((r % 11) + 1); | |
3f21a6b7 URS |
306 | |
307 | pcpu[i] = __alloc_percpu(size, align); | |
308 | if (!pcpu[i]) | |
309 | rv = -1; | |
310 | } | |
311 | ||
312 | for (i = 0; i < 35000; i++) | |
313 | free_percpu(pcpu[i]); | |
314 | ||
315 | vfree(pcpu); | |
316 | #endif | |
317 | return rv; | |
318 | } | |
319 | ||
da4fc00a URS |
320 | struct test_kvfree_rcu { |
321 | struct rcu_head rcu; | |
322 | unsigned char array[20]; | |
323 | }; | |
324 | ||
325 | static int | |
326 | kvfree_rcu_1_arg_vmalloc_test(void) | |
327 | { | |
328 | struct test_kvfree_rcu *p; | |
329 | int i; | |
330 | ||
331 | for (i = 0; i < test_loop_count; i++) { | |
332 | p = vmalloc(1 * PAGE_SIZE); | |
333 | if (!p) | |
334 | return -1; | |
335 | ||
336 | p->array[0] = 'a'; | |
337 | kvfree_rcu(p); | |
338 | } | |
339 | ||
340 | return 0; | |
341 | } | |
342 | ||
343 | static int | |
344 | kvfree_rcu_2_arg_vmalloc_test(void) | |
345 | { | |
346 | struct test_kvfree_rcu *p; | |
347 | int i; | |
348 | ||
349 | for (i = 0; i < test_loop_count; i++) { | |
350 | p = vmalloc(1 * PAGE_SIZE); | |
351 | if (!p) | |
352 | return -1; | |
353 | ||
354 | p->array[0] = 'a'; | |
355 | kvfree_rcu(p, rcu); | |
356 | } | |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
3f21a6b7 URS |
361 | struct test_case_desc { |
362 | const char *test_name; | |
363 | int (*test_func)(void); | |
364 | }; | |
365 | ||
366 | static struct test_case_desc test_case_array[] = { | |
367 | { "fix_size_alloc_test", fix_size_alloc_test }, | |
368 | { "full_fit_alloc_test", full_fit_alloc_test }, | |
369 | { "long_busy_list_alloc_test", long_busy_list_alloc_test }, | |
370 | { "random_size_alloc_test", random_size_alloc_test }, | |
371 | { "fix_align_alloc_test", fix_align_alloc_test }, | |
372 | { "random_size_align_alloc_test", random_size_align_alloc_test }, | |
373 | { "align_shift_alloc_test", align_shift_alloc_test }, | |
374 | { "pcpu_alloc_test", pcpu_alloc_test }, | |
da4fc00a URS |
375 | { "kvfree_rcu_1_arg_vmalloc_test", kvfree_rcu_1_arg_vmalloc_test }, |
376 | { "kvfree_rcu_2_arg_vmalloc_test", kvfree_rcu_2_arg_vmalloc_test }, | |
3f21a6b7 URS |
377 | /* Add a new test case here. */ |
378 | }; | |
379 | ||
380 | struct test_case_data { | |
381 | int test_failed; | |
382 | int test_passed; | |
383 | u64 time; | |
384 | }; | |
385 | ||
3f21a6b7 URS |
386 | static struct test_driver { |
387 | struct task_struct *task; | |
80f47599 URS |
388 | struct test_case_data data[ARRAY_SIZE(test_case_array)]; |
389 | ||
3f21a6b7 URS |
390 | unsigned long start; |
391 | unsigned long stop; | |
80f47599 | 392 | } *tdriver; |
3f21a6b7 URS |
393 | |
394 | static void shuffle_array(int *arr, int n) | |
395 | { | |
396 | unsigned int rnd; | |
34b46efd | 397 | int i, j; |
3f21a6b7 URS |
398 | |
399 | for (i = n - 1; i > 0; i--) { | |
5e21f2d5 | 400 | rnd = prandom_u32(); |
3f21a6b7 URS |
401 | |
402 | /* Cut the range. */ | |
403 | j = rnd % i; | |
404 | ||
405 | /* Swap indexes. */ | |
34b46efd | 406 | swap(arr[i], arr[j]); |
3f21a6b7 URS |
407 | } |
408 | } | |
409 | ||
410 | static int test_func(void *private) | |
411 | { | |
412 | struct test_driver *t = private; | |
3f21a6b7 | 413 | int random_array[ARRAY_SIZE(test_case_array)]; |
7507c402 | 414 | int index, i, j; |
3f21a6b7 URS |
415 | ktime_t kt; |
416 | u64 delta; | |
417 | ||
3f21a6b7 URS |
418 | for (i = 0; i < ARRAY_SIZE(test_case_array); i++) |
419 | random_array[i] = i; | |
420 | ||
421 | if (!sequential_test_order) | |
422 | shuffle_array(random_array, ARRAY_SIZE(test_case_array)); | |
423 | ||
424 | /* | |
425 | * Block until initialization is done. | |
426 | */ | |
427 | down_read(&prepare_for_test_rwsem); | |
428 | ||
429 | t->start = get_cycles(); | |
430 | for (i = 0; i < ARRAY_SIZE(test_case_array); i++) { | |
431 | index = random_array[i]; | |
432 | ||
433 | /* | |
434 | * Skip tests if run_test_mask has been specified. | |
435 | */ | |
436 | if (!((run_test_mask & (1 << index)) >> index)) | |
437 | continue; | |
438 | ||
439 | kt = ktime_get(); | |
440 | for (j = 0; j < test_repeat_count; j++) { | |
7507c402 | 441 | if (!test_case_array[index].test_func()) |
80f47599 | 442 | t->data[index].test_passed++; |
3f21a6b7 | 443 | else |
80f47599 | 444 | t->data[index].test_failed++; |
3f21a6b7 URS |
445 | } |
446 | ||
447 | /* | |
448 | * Take an average time that test took. | |
449 | */ | |
450 | delta = (u64) ktime_us_delta(ktime_get(), kt); | |
451 | do_div(delta, (u32) test_repeat_count); | |
452 | ||
80f47599 | 453 | t->data[index].time = delta; |
3f21a6b7 URS |
454 | } |
455 | t->stop = get_cycles(); | |
456 | ||
457 | up_read(&prepare_for_test_rwsem); | |
458 | test_report_one_done(); | |
459 | ||
460 | /* | |
461 | * Wait for the kthread_stop() call. | |
462 | */ | |
463 | while (!kthread_should_stop()) | |
464 | msleep(10); | |
465 | ||
466 | return 0; | |
467 | } | |
468 | ||
80f47599 | 469 | static int |
3f21a6b7 URS |
470 | init_test_configurtion(void) |
471 | { | |
472 | /* | |
80f47599 URS |
473 | * A maximum number of workers is defined as hard-coded |
474 | * value and set to USHRT_MAX. We add such gap just in | |
475 | * case and for potential heavy stressing. | |
3f21a6b7 | 476 | */ |
80f47599 | 477 | nr_threads = clamp(nr_threads, 1, (int) USHRT_MAX); |
3f21a6b7 | 478 | |
80f47599 URS |
479 | /* Allocate the space for test instances. */ |
480 | tdriver = kvcalloc(nr_threads, sizeof(*tdriver), GFP_KERNEL); | |
481 | if (tdriver == NULL) | |
482 | return -1; | |
3f21a6b7 URS |
483 | |
484 | if (test_repeat_count <= 0) | |
485 | test_repeat_count = 1; | |
486 | ||
487 | if (test_loop_count <= 0) | |
488 | test_loop_count = 1; | |
80f47599 URS |
489 | |
490 | return 0; | |
3f21a6b7 URS |
491 | } |
492 | ||
493 | static void do_concurrent_test(void) | |
494 | { | |
80f47599 | 495 | int i, ret; |
3f21a6b7 URS |
496 | |
497 | /* | |
498 | * Set some basic configurations plus sanity check. | |
499 | */ | |
80f47599 URS |
500 | ret = init_test_configurtion(); |
501 | if (ret < 0) | |
502 | return; | |
3f21a6b7 URS |
503 | |
504 | /* | |
505 | * Put on hold all workers. | |
506 | */ | |
507 | down_write(&prepare_for_test_rwsem); | |
508 | ||
80f47599 URS |
509 | for (i = 0; i < nr_threads; i++) { |
510 | struct test_driver *t = &tdriver[i]; | |
3f21a6b7 | 511 | |
80f47599 | 512 | t->task = kthread_run(test_func, t, "vmalloc_test/%d", i); |
3f21a6b7 URS |
513 | |
514 | if (!IS_ERR(t->task)) | |
515 | /* Success. */ | |
516 | atomic_inc(&test_n_undone); | |
517 | else | |
80f47599 | 518 | pr_err("Failed to start %d kthread\n", i); |
3f21a6b7 URS |
519 | } |
520 | ||
521 | /* | |
522 | * Now let the workers do their job. | |
523 | */ | |
524 | up_write(&prepare_for_test_rwsem); | |
525 | ||
526 | /* | |
527 | * Sleep quiet until all workers are done with 1 second | |
528 | * interval. Since the test can take a lot of time we | |
529 | * can run into a stack trace of the hung task. That is | |
530 | * why we go with completion_timeout and HZ value. | |
531 | */ | |
532 | do { | |
533 | ret = wait_for_completion_timeout(&test_all_done_comp, HZ); | |
534 | } while (!ret); | |
535 | ||
80f47599 URS |
536 | for (i = 0; i < nr_threads; i++) { |
537 | struct test_driver *t = &tdriver[i]; | |
538 | int j; | |
3f21a6b7 URS |
539 | |
540 | if (!IS_ERR(t->task)) | |
541 | kthread_stop(t->task); | |
542 | ||
80f47599 URS |
543 | for (j = 0; j < ARRAY_SIZE(test_case_array); j++) { |
544 | if (!((run_test_mask & (1 << j)) >> j)) | |
3f21a6b7 URS |
545 | continue; |
546 | ||
547 | pr_info( | |
548 | "Summary: %s passed: %d failed: %d repeat: %d loops: %d avg: %llu usec\n", | |
80f47599 URS |
549 | test_case_array[j].test_name, |
550 | t->data[j].test_passed, | |
551 | t->data[j].test_failed, | |
3f21a6b7 | 552 | test_repeat_count, test_loop_count, |
80f47599 | 553 | t->data[j].time); |
3f21a6b7 URS |
554 | } |
555 | ||
80f47599 URS |
556 | pr_info("All test took worker%d=%lu cycles\n", |
557 | i, t->stop - t->start); | |
3f21a6b7 | 558 | } |
80f47599 URS |
559 | |
560 | kvfree(tdriver); | |
3f21a6b7 URS |
561 | } |
562 | ||
563 | static int vmalloc_test_init(void) | |
564 | { | |
565 | do_concurrent_test(); | |
566 | return -EAGAIN; /* Fail will directly unload the module */ | |
567 | } | |
568 | ||
569 | static void vmalloc_test_exit(void) | |
570 | { | |
571 | } | |
572 | ||
573 | module_init(vmalloc_test_init) | |
574 | module_exit(vmalloc_test_exit) | |
575 | ||
576 | MODULE_LICENSE("GPL"); | |
577 | MODULE_AUTHOR("Uladzislau Rezki"); | |
578 | MODULE_DESCRIPTION("vmalloc test module"); |