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 | ||
6b1ead59 QP |
41 | __param(bool, use_huge, false, |
42 | "Use vmalloc_huge in fix_size_alloc_test"); | |
43 | ||
3f21a6b7 URS |
44 | __param(int, run_test_mask, INT_MAX, |
45 | "Set tests specified in the mask.\n\n" | |
da4fc00a URS |
46 | "\t\tid: 1, name: fix_size_alloc_test\n" |
47 | "\t\tid: 2, name: full_fit_alloc_test\n" | |
48 | "\t\tid: 4, name: long_busy_list_alloc_test\n" | |
49 | "\t\tid: 8, name: random_size_alloc_test\n" | |
50 | "\t\tid: 16, name: fix_align_alloc_test\n" | |
51 | "\t\tid: 32, name: random_size_align_alloc_test\n" | |
52 | "\t\tid: 64, name: align_shift_alloc_test\n" | |
53 | "\t\tid: 128, name: pcpu_alloc_test\n" | |
54 | "\t\tid: 256, name: kvfree_rcu_1_arg_vmalloc_test\n" | |
55 | "\t\tid: 512, name: kvfree_rcu_2_arg_vmalloc_test\n" | |
3f21a6b7 URS |
56 | /* Add a new test case description here. */ |
57 | ); | |
58 | ||
3f21a6b7 URS |
59 | /* |
60 | * Read write semaphore for synchronization of setup | |
61 | * phase that is done in main thread and workers. | |
62 | */ | |
63 | static DECLARE_RWSEM(prepare_for_test_rwsem); | |
64 | ||
65 | /* | |
66 | * Completion tracking for worker threads. | |
67 | */ | |
68 | static DECLARE_COMPLETION(test_all_done_comp); | |
69 | static atomic_t test_n_undone = ATOMIC_INIT(0); | |
70 | ||
71 | static inline void | |
72 | test_report_one_done(void) | |
73 | { | |
74 | if (atomic_dec_and_test(&test_n_undone)) | |
75 | complete(&test_all_done_comp); | |
76 | } | |
77 | ||
78 | static int random_size_align_alloc_test(void) | |
79 | { | |
5e21f2d5 URS |
80 | unsigned long size, align; |
81 | unsigned int rnd; | |
3f21a6b7 URS |
82 | void *ptr; |
83 | int i; | |
84 | ||
85 | for (i = 0; i < test_loop_count; i++) { | |
f743f16c | 86 | rnd = get_random_u8(); |
3f21a6b7 URS |
87 | |
88 | /* | |
89 | * Maximum 1024 pages, if PAGE_SIZE is 4096. | |
90 | */ | |
91 | align = 1 << (rnd % 23); | |
92 | ||
93 | /* | |
94 | * Maximum 10 pages. | |
95 | */ | |
96 | size = ((rnd % 10) + 1) * PAGE_SIZE; | |
97 | ||
c3f896dc CH |
98 | ptr = __vmalloc_node(size, align, GFP_KERNEL | __GFP_ZERO, 0, |
99 | __builtin_return_address(0)); | |
3f21a6b7 URS |
100 | if (!ptr) |
101 | return -1; | |
102 | ||
103 | vfree(ptr); | |
104 | } | |
105 | ||
106 | return 0; | |
107 | } | |
108 | ||
109 | /* | |
110 | * This test case is supposed to be failed. | |
111 | */ | |
112 | static int align_shift_alloc_test(void) | |
113 | { | |
114 | unsigned long align; | |
115 | void *ptr; | |
116 | int i; | |
117 | ||
118 | for (i = 0; i < BITS_PER_LONG; i++) { | |
119 | align = ((unsigned long) 1) << i; | |
120 | ||
c3f896dc CH |
121 | ptr = __vmalloc_node(PAGE_SIZE, align, GFP_KERNEL|__GFP_ZERO, 0, |
122 | __builtin_return_address(0)); | |
3f21a6b7 URS |
123 | if (!ptr) |
124 | return -1; | |
125 | ||
126 | vfree(ptr); | |
127 | } | |
128 | ||
129 | return 0; | |
130 | } | |
131 | ||
132 | static int fix_align_alloc_test(void) | |
133 | { | |
134 | void *ptr; | |
135 | int i; | |
136 | ||
137 | for (i = 0; i < test_loop_count; i++) { | |
c3f896dc CH |
138 | ptr = __vmalloc_node(5 * PAGE_SIZE, THREAD_ALIGN << 1, |
139 | GFP_KERNEL | __GFP_ZERO, 0, | |
140 | __builtin_return_address(0)); | |
3f21a6b7 URS |
141 | if (!ptr) |
142 | return -1; | |
143 | ||
144 | vfree(ptr); | |
145 | } | |
146 | ||
147 | return 0; | |
148 | } | |
149 | ||
150 | static int random_size_alloc_test(void) | |
151 | { | |
152 | unsigned int n; | |
153 | void *p; | |
154 | int i; | |
155 | ||
156 | for (i = 0; i < test_loop_count; i++) { | |
e8a533cb | 157 | n = get_random_u32_inclusive(1, 100); |
3f21a6b7 URS |
158 | p = vmalloc(n * PAGE_SIZE); |
159 | ||
160 | if (!p) | |
161 | return -1; | |
162 | ||
163 | *((__u8 *)p) = 1; | |
164 | vfree(p); | |
165 | } | |
166 | ||
167 | return 0; | |
168 | } | |
169 | ||
170 | static int long_busy_list_alloc_test(void) | |
171 | { | |
172 | void *ptr_1, *ptr_2; | |
173 | void **ptr; | |
174 | int rv = -1; | |
175 | int i; | |
176 | ||
177 | ptr = vmalloc(sizeof(void *) * 15000); | |
178 | if (!ptr) | |
179 | return rv; | |
180 | ||
181 | for (i = 0; i < 15000; i++) | |
182 | ptr[i] = vmalloc(1 * PAGE_SIZE); | |
183 | ||
184 | for (i = 0; i < test_loop_count; i++) { | |
185 | ptr_1 = vmalloc(100 * PAGE_SIZE); | |
186 | if (!ptr_1) | |
187 | goto leave; | |
188 | ||
189 | ptr_2 = vmalloc(1 * PAGE_SIZE); | |
190 | if (!ptr_2) { | |
191 | vfree(ptr_1); | |
192 | goto leave; | |
193 | } | |
194 | ||
195 | *((__u8 *)ptr_1) = 0; | |
196 | *((__u8 *)ptr_2) = 1; | |
197 | ||
198 | vfree(ptr_1); | |
199 | vfree(ptr_2); | |
200 | } | |
201 | ||
202 | /* Success */ | |
203 | rv = 0; | |
204 | ||
205 | leave: | |
206 | for (i = 0; i < 15000; i++) | |
207 | vfree(ptr[i]); | |
208 | ||
209 | vfree(ptr); | |
210 | return rv; | |
211 | } | |
212 | ||
213 | static int full_fit_alloc_test(void) | |
214 | { | |
215 | void **ptr, **junk_ptr, *tmp; | |
216 | int junk_length; | |
217 | int rv = -1; | |
218 | int i; | |
219 | ||
220 | junk_length = fls(num_online_cpus()); | |
221 | junk_length *= (32 * 1024 * 1024 / PAGE_SIZE); | |
222 | ||
223 | ptr = vmalloc(sizeof(void *) * junk_length); | |
224 | if (!ptr) | |
225 | return rv; | |
226 | ||
227 | junk_ptr = vmalloc(sizeof(void *) * junk_length); | |
228 | if (!junk_ptr) { | |
229 | vfree(ptr); | |
230 | return rv; | |
231 | } | |
232 | ||
233 | for (i = 0; i < junk_length; i++) { | |
234 | ptr[i] = vmalloc(1 * PAGE_SIZE); | |
235 | junk_ptr[i] = vmalloc(1 * PAGE_SIZE); | |
236 | } | |
237 | ||
238 | for (i = 0; i < junk_length; i++) | |
239 | vfree(junk_ptr[i]); | |
240 | ||
241 | for (i = 0; i < test_loop_count; i++) { | |
242 | tmp = vmalloc(1 * PAGE_SIZE); | |
243 | ||
244 | if (!tmp) | |
245 | goto error; | |
246 | ||
247 | *((__u8 *)tmp) = 1; | |
248 | vfree(tmp); | |
249 | } | |
250 | ||
251 | /* Success */ | |
252 | rv = 0; | |
253 | ||
254 | error: | |
255 | for (i = 0; i < junk_length; i++) | |
256 | vfree(ptr[i]); | |
257 | ||
258 | vfree(ptr); | |
259 | vfree(junk_ptr); | |
260 | ||
261 | return rv; | |
262 | } | |
263 | ||
264 | static int fix_size_alloc_test(void) | |
265 | { | |
266 | void *ptr; | |
267 | int i; | |
268 | ||
269 | for (i = 0; i < test_loop_count; i++) { | |
6b1ead59 QP |
270 | if (use_huge) |
271 | ptr = vmalloc_huge((nr_pages > 0 ? nr_pages:1) * PAGE_SIZE, GFP_KERNEL); | |
272 | else | |
273 | ptr = vmalloc((nr_pages > 0 ? nr_pages:1) * PAGE_SIZE); | |
3f21a6b7 URS |
274 | |
275 | if (!ptr) | |
276 | return -1; | |
277 | ||
278 | *((__u8 *)ptr) = 0; | |
279 | ||
280 | vfree(ptr); | |
281 | } | |
282 | ||
283 | return 0; | |
284 | } | |
285 | ||
286 | static int | |
287 | pcpu_alloc_test(void) | |
288 | { | |
289 | int rv = 0; | |
290 | #ifndef CONFIG_NEED_PER_CPU_KM | |
291 | void __percpu **pcpu; | |
292 | size_t size, align; | |
293 | int i; | |
294 | ||
295 | pcpu = vmalloc(sizeof(void __percpu *) * 35000); | |
296 | if (!pcpu) | |
297 | return -1; | |
298 | ||
299 | for (i = 0; i < 35000; i++) { | |
e8a533cb | 300 | size = get_random_u32_inclusive(1, PAGE_SIZE / 4); |
3f21a6b7 URS |
301 | |
302 | /* | |
303 | * Maximum PAGE_SIZE | |
304 | */ | |
e8a533cb | 305 | align = 1 << get_random_u32_inclusive(1, 11); |
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'; | |
c779b972 | 337 | kvfree_rcu_mightsleep(p); |
da4fc00a URS |
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 | { | |
34b46efd | 396 | int i, j; |
3f21a6b7 URS |
397 | |
398 | for (i = n - 1; i > 0; i--) { | |
3f21a6b7 | 399 | /* Cut the range. */ |
8032bf12 | 400 | j = get_random_u32_below(i); |
3f21a6b7 URS |
401 | |
402 | /* Swap indexes. */ | |
34b46efd | 403 | swap(arr[i], arr[j]); |
3f21a6b7 URS |
404 | } |
405 | } | |
406 | ||
407 | static int test_func(void *private) | |
408 | { | |
409 | struct test_driver *t = private; | |
3f21a6b7 | 410 | int random_array[ARRAY_SIZE(test_case_array)]; |
7507c402 | 411 | int index, i, j; |
3f21a6b7 URS |
412 | ktime_t kt; |
413 | u64 delta; | |
414 | ||
3f21a6b7 URS |
415 | for (i = 0; i < ARRAY_SIZE(test_case_array); i++) |
416 | random_array[i] = i; | |
417 | ||
418 | if (!sequential_test_order) | |
419 | shuffle_array(random_array, ARRAY_SIZE(test_case_array)); | |
420 | ||
421 | /* | |
422 | * Block until initialization is done. | |
423 | */ | |
424 | down_read(&prepare_for_test_rwsem); | |
425 | ||
426 | t->start = get_cycles(); | |
427 | for (i = 0; i < ARRAY_SIZE(test_case_array); i++) { | |
428 | index = random_array[i]; | |
429 | ||
430 | /* | |
431 | * Skip tests if run_test_mask has been specified. | |
432 | */ | |
433 | if (!((run_test_mask & (1 << index)) >> index)) | |
434 | continue; | |
435 | ||
436 | kt = ktime_get(); | |
437 | for (j = 0; j < test_repeat_count; j++) { | |
7507c402 | 438 | if (!test_case_array[index].test_func()) |
80f47599 | 439 | t->data[index].test_passed++; |
3f21a6b7 | 440 | else |
80f47599 | 441 | t->data[index].test_failed++; |
3f21a6b7 URS |
442 | } |
443 | ||
444 | /* | |
445 | * Take an average time that test took. | |
446 | */ | |
447 | delta = (u64) ktime_us_delta(ktime_get(), kt); | |
448 | do_div(delta, (u32) test_repeat_count); | |
449 | ||
80f47599 | 450 | t->data[index].time = delta; |
3f21a6b7 URS |
451 | } |
452 | t->stop = get_cycles(); | |
453 | ||
454 | up_read(&prepare_for_test_rwsem); | |
455 | test_report_one_done(); | |
456 | ||
457 | /* | |
458 | * Wait for the kthread_stop() call. | |
459 | */ | |
460 | while (!kthread_should_stop()) | |
461 | msleep(10); | |
462 | ||
463 | return 0; | |
464 | } | |
465 | ||
80f47599 | 466 | static int |
3f21a6b7 URS |
467 | init_test_configurtion(void) |
468 | { | |
469 | /* | |
80f47599 URS |
470 | * A maximum number of workers is defined as hard-coded |
471 | * value and set to USHRT_MAX. We add such gap just in | |
472 | * case and for potential heavy stressing. | |
3f21a6b7 | 473 | */ |
80f47599 | 474 | nr_threads = clamp(nr_threads, 1, (int) USHRT_MAX); |
3f21a6b7 | 475 | |
80f47599 URS |
476 | /* Allocate the space for test instances. */ |
477 | tdriver = kvcalloc(nr_threads, sizeof(*tdriver), GFP_KERNEL); | |
478 | if (tdriver == NULL) | |
479 | return -1; | |
3f21a6b7 URS |
480 | |
481 | if (test_repeat_count <= 0) | |
482 | test_repeat_count = 1; | |
483 | ||
484 | if (test_loop_count <= 0) | |
485 | test_loop_count = 1; | |
80f47599 URS |
486 | |
487 | return 0; | |
3f21a6b7 URS |
488 | } |
489 | ||
490 | static void do_concurrent_test(void) | |
491 | { | |
80f47599 | 492 | int i, ret; |
3f21a6b7 URS |
493 | |
494 | /* | |
495 | * Set some basic configurations plus sanity check. | |
496 | */ | |
80f47599 URS |
497 | ret = init_test_configurtion(); |
498 | if (ret < 0) | |
499 | return; | |
3f21a6b7 URS |
500 | |
501 | /* | |
502 | * Put on hold all workers. | |
503 | */ | |
504 | down_write(&prepare_for_test_rwsem); | |
505 | ||
80f47599 URS |
506 | for (i = 0; i < nr_threads; i++) { |
507 | struct test_driver *t = &tdriver[i]; | |
3f21a6b7 | 508 | |
80f47599 | 509 | t->task = kthread_run(test_func, t, "vmalloc_test/%d", i); |
3f21a6b7 URS |
510 | |
511 | if (!IS_ERR(t->task)) | |
512 | /* Success. */ | |
513 | atomic_inc(&test_n_undone); | |
514 | else | |
80f47599 | 515 | pr_err("Failed to start %d kthread\n", i); |
3f21a6b7 URS |
516 | } |
517 | ||
518 | /* | |
519 | * Now let the workers do their job. | |
520 | */ | |
521 | up_write(&prepare_for_test_rwsem); | |
522 | ||
523 | /* | |
524 | * Sleep quiet until all workers are done with 1 second | |
525 | * interval. Since the test can take a lot of time we | |
526 | * can run into a stack trace of the hung task. That is | |
527 | * why we go with completion_timeout and HZ value. | |
528 | */ | |
529 | do { | |
530 | ret = wait_for_completion_timeout(&test_all_done_comp, HZ); | |
531 | } while (!ret); | |
532 | ||
80f47599 URS |
533 | for (i = 0; i < nr_threads; i++) { |
534 | struct test_driver *t = &tdriver[i]; | |
535 | int j; | |
3f21a6b7 URS |
536 | |
537 | if (!IS_ERR(t->task)) | |
538 | kthread_stop(t->task); | |
539 | ||
80f47599 URS |
540 | for (j = 0; j < ARRAY_SIZE(test_case_array); j++) { |
541 | if (!((run_test_mask & (1 << j)) >> j)) | |
3f21a6b7 URS |
542 | continue; |
543 | ||
544 | pr_info( | |
545 | "Summary: %s passed: %d failed: %d repeat: %d loops: %d avg: %llu usec\n", | |
80f47599 URS |
546 | test_case_array[j].test_name, |
547 | t->data[j].test_passed, | |
548 | t->data[j].test_failed, | |
3f21a6b7 | 549 | test_repeat_count, test_loop_count, |
80f47599 | 550 | t->data[j].time); |
3f21a6b7 URS |
551 | } |
552 | ||
80f47599 URS |
553 | pr_info("All test took worker%d=%lu cycles\n", |
554 | i, t->stop - t->start); | |
3f21a6b7 | 555 | } |
80f47599 URS |
556 | |
557 | kvfree(tdriver); | |
3f21a6b7 URS |
558 | } |
559 | ||
560 | static int vmalloc_test_init(void) | |
561 | { | |
562 | do_concurrent_test(); | |
563 | return -EAGAIN; /* Fail will directly unload the module */ | |
564 | } | |
565 | ||
566 | static void vmalloc_test_exit(void) | |
567 | { | |
568 | } | |
569 | ||
570 | module_init(vmalloc_test_init) | |
571 | module_exit(vmalloc_test_exit) | |
572 | ||
573 | MODULE_LICENSE("GPL"); | |
574 | MODULE_AUTHOR("Uladzislau Rezki"); | |
575 | MODULE_DESCRIPTION("vmalloc test module"); |