1 // SPDX-License-Identifier: GPL-2.0
4 * Test module for stress and analyze performance of vmalloc allocator.
5 * (C) 2018 Uladzislau Rezki (Sony) <urezki@gmail.com>
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>
18 #include <linux/rcupdate.h>
19 #include <linux/slab.h>
21 #define __param(type, name, init, msg) \
22 static type name = init; \
23 module_param(name, type, 0444); \
24 MODULE_PARM_DESC(name, msg) \
26 __param(bool, single_cpu_test, false,
27 "Use single first online CPU to run tests");
29 __param(bool, sequential_test_order, false,
30 "Use sequential stress tests order");
32 __param(int, test_repeat_count, 1,
33 "Set test repeat counter");
35 __param(int, test_loop_count, 1000000,
36 "Set test loop counter");
38 __param(int, run_test_mask, INT_MAX,
39 "Set tests specified in the mask.\n\n"
40 "\t\tid: 1, name: fix_size_alloc_test\n"
41 "\t\tid: 2, name: full_fit_alloc_test\n"
42 "\t\tid: 4, name: long_busy_list_alloc_test\n"
43 "\t\tid: 8, name: random_size_alloc_test\n"
44 "\t\tid: 16, name: fix_align_alloc_test\n"
45 "\t\tid: 32, name: random_size_align_alloc_test\n"
46 "\t\tid: 64, name: align_shift_alloc_test\n"
47 "\t\tid: 128, name: pcpu_alloc_test\n"
48 "\t\tid: 256, name: kvfree_rcu_1_arg_vmalloc_test\n"
49 "\t\tid: 512, name: kvfree_rcu_2_arg_vmalloc_test\n"
50 "\t\tid: 1024, name: kvfree_rcu_1_arg_slab_test\n"
51 "\t\tid: 2048, name: kvfree_rcu_2_arg_slab_test\n"
52 /* Add a new test case description here. */
56 * Depends on single_cpu_test parameter. If it is true, then
57 * use first online CPU to trigger a test on, otherwise go with
60 static cpumask_t cpus_run_test_mask = CPU_MASK_NONE;
63 * Read write semaphore for synchronization of setup
64 * phase that is done in main thread and workers.
66 static DECLARE_RWSEM(prepare_for_test_rwsem);
69 * Completion tracking for worker threads.
71 static DECLARE_COMPLETION(test_all_done_comp);
72 static atomic_t test_n_undone = ATOMIC_INIT(0);
75 test_report_one_done(void)
77 if (atomic_dec_and_test(&test_n_undone))
78 complete(&test_all_done_comp);
81 static int random_size_align_alloc_test(void)
83 unsigned long size, align, rnd;
87 for (i = 0; i < test_loop_count; i++) {
88 get_random_bytes(&rnd, sizeof(rnd));
91 * Maximum 1024 pages, if PAGE_SIZE is 4096.
93 align = 1 << (rnd % 23);
98 size = ((rnd % 10) + 1) * PAGE_SIZE;
100 ptr = __vmalloc_node(size, align, GFP_KERNEL | __GFP_ZERO, 0,
101 __builtin_return_address(0));
112 * This test case is supposed to be failed.
114 static int align_shift_alloc_test(void)
120 for (i = 0; i < BITS_PER_LONG; i++) {
121 align = ((unsigned long) 1) << i;
123 ptr = __vmalloc_node(PAGE_SIZE, align, GFP_KERNEL|__GFP_ZERO, 0,
124 __builtin_return_address(0));
134 static int fix_align_alloc_test(void)
139 for (i = 0; i < test_loop_count; i++) {
140 ptr = __vmalloc_node(5 * PAGE_SIZE, THREAD_ALIGN << 1,
141 GFP_KERNEL | __GFP_ZERO, 0,
142 __builtin_return_address(0));
152 static int random_size_alloc_test(void)
158 for (i = 0; i < test_loop_count; i++) {
159 get_random_bytes(&n, sizeof(i));
162 p = vmalloc(n * PAGE_SIZE);
174 static int long_busy_list_alloc_test(void)
181 ptr = vmalloc(sizeof(void *) * 15000);
185 for (i = 0; i < 15000; i++)
186 ptr[i] = vmalloc(1 * PAGE_SIZE);
188 for (i = 0; i < test_loop_count; i++) {
189 ptr_1 = vmalloc(100 * PAGE_SIZE);
193 ptr_2 = vmalloc(1 * PAGE_SIZE);
199 *((__u8 *)ptr_1) = 0;
200 *((__u8 *)ptr_2) = 1;
210 for (i = 0; i < 15000; i++)
217 static int full_fit_alloc_test(void)
219 void **ptr, **junk_ptr, *tmp;
224 junk_length = fls(num_online_cpus());
225 junk_length *= (32 * 1024 * 1024 / PAGE_SIZE);
227 ptr = vmalloc(sizeof(void *) * junk_length);
231 junk_ptr = vmalloc(sizeof(void *) * junk_length);
237 for (i = 0; i < junk_length; i++) {
238 ptr[i] = vmalloc(1 * PAGE_SIZE);
239 junk_ptr[i] = vmalloc(1 * PAGE_SIZE);
242 for (i = 0; i < junk_length; i++)
245 for (i = 0; i < test_loop_count; i++) {
246 tmp = vmalloc(1 * PAGE_SIZE);
259 for (i = 0; i < junk_length; i++)
268 static int fix_size_alloc_test(void)
273 for (i = 0; i < test_loop_count; i++) {
274 ptr = vmalloc(3 * PAGE_SIZE);
288 pcpu_alloc_test(void)
291 #ifndef CONFIG_NEED_PER_CPU_KM
292 void __percpu **pcpu;
296 pcpu = vmalloc(sizeof(void __percpu *) * 35000);
300 for (i = 0; i < 35000; i++) {
303 get_random_bytes(&r, sizeof(i));
304 size = (r % (PAGE_SIZE / 4)) + 1;
309 get_random_bytes(&r, sizeof(i));
310 align = 1 << ((i % 11) + 1);
312 pcpu[i] = __alloc_percpu(size, align);
317 for (i = 0; i < 35000; i++)
318 free_percpu(pcpu[i]);
325 struct test_kvfree_rcu {
327 unsigned char array[20];
331 kvfree_rcu_1_arg_vmalloc_test(void)
333 struct test_kvfree_rcu *p;
336 for (i = 0; i < test_loop_count; i++) {
337 p = vmalloc(1 * PAGE_SIZE);
349 kvfree_rcu_2_arg_vmalloc_test(void)
351 struct test_kvfree_rcu *p;
354 for (i = 0; i < test_loop_count; i++) {
355 p = vmalloc(1 * PAGE_SIZE);
367 kvfree_rcu_1_arg_slab_test(void)
369 struct test_kvfree_rcu *p;
372 for (i = 0; i < test_loop_count; i++) {
373 p = kmalloc(sizeof(*p), GFP_KERNEL);
385 kvfree_rcu_2_arg_slab_test(void)
387 struct test_kvfree_rcu *p;
390 for (i = 0; i < test_loop_count; i++) {
391 p = kmalloc(sizeof(*p), GFP_KERNEL);
402 struct test_case_desc {
403 const char *test_name;
404 int (*test_func)(void);
407 static struct test_case_desc test_case_array[] = {
408 { "fix_size_alloc_test", fix_size_alloc_test },
409 { "full_fit_alloc_test", full_fit_alloc_test },
410 { "long_busy_list_alloc_test", long_busy_list_alloc_test },
411 { "random_size_alloc_test", random_size_alloc_test },
412 { "fix_align_alloc_test", fix_align_alloc_test },
413 { "random_size_align_alloc_test", random_size_align_alloc_test },
414 { "align_shift_alloc_test", align_shift_alloc_test },
415 { "pcpu_alloc_test", pcpu_alloc_test },
416 { "kvfree_rcu_1_arg_vmalloc_test", kvfree_rcu_1_arg_vmalloc_test },
417 { "kvfree_rcu_2_arg_vmalloc_test", kvfree_rcu_2_arg_vmalloc_test },
418 { "kvfree_rcu_1_arg_slab_test", kvfree_rcu_1_arg_slab_test },
419 { "kvfree_rcu_2_arg_slab_test", kvfree_rcu_2_arg_slab_test },
420 /* Add a new test case here. */
423 struct test_case_data {
429 /* Split it to get rid of: WARNING: line over 80 characters */
430 static struct test_case_data
431 per_cpu_test_data[NR_CPUS][ARRAY_SIZE(test_case_array)];
433 static struct test_driver {
434 struct task_struct *task;
438 } per_cpu_test_driver[NR_CPUS];
440 static void shuffle_array(int *arr, int n)
445 for (i = n - 1; i > 0; i--) {
446 get_random_bytes(&rnd, sizeof(rnd));
458 static int test_func(void *private)
460 struct test_driver *t = private;
461 int random_array[ARRAY_SIZE(test_case_array)];
466 if (set_cpus_allowed_ptr(current, cpumask_of(t->cpu)) < 0)
467 pr_err("Failed to set affinity to %d CPU\n", t->cpu);
469 for (i = 0; i < ARRAY_SIZE(test_case_array); i++)
472 if (!sequential_test_order)
473 shuffle_array(random_array, ARRAY_SIZE(test_case_array));
476 * Block until initialization is done.
478 down_read(&prepare_for_test_rwsem);
480 t->start = get_cycles();
481 for (i = 0; i < ARRAY_SIZE(test_case_array); i++) {
482 index = random_array[i];
485 * Skip tests if run_test_mask has been specified.
487 if (!((run_test_mask & (1 << index)) >> index))
491 for (j = 0; j < test_repeat_count; j++) {
492 if (!test_case_array[index].test_func())
493 per_cpu_test_data[t->cpu][index].test_passed++;
495 per_cpu_test_data[t->cpu][index].test_failed++;
499 * Take an average time that test took.
501 delta = (u64) ktime_us_delta(ktime_get(), kt);
502 do_div(delta, (u32) test_repeat_count);
504 per_cpu_test_data[t->cpu][index].time = delta;
506 t->stop = get_cycles();
508 up_read(&prepare_for_test_rwsem);
509 test_report_one_done();
512 * Wait for the kthread_stop() call.
514 while (!kthread_should_stop())
521 init_test_configurtion(void)
524 * Reset all data of all CPUs.
526 memset(per_cpu_test_data, 0, sizeof(per_cpu_test_data));
529 cpumask_set_cpu(cpumask_first(cpu_online_mask),
530 &cpus_run_test_mask);
532 cpumask_and(&cpus_run_test_mask, cpu_online_mask,
535 if (test_repeat_count <= 0)
536 test_repeat_count = 1;
538 if (test_loop_count <= 0)
542 static void do_concurrent_test(void)
547 * Set some basic configurations plus sanity check.
549 init_test_configurtion();
552 * Put on hold all workers.
554 down_write(&prepare_for_test_rwsem);
556 for_each_cpu(cpu, &cpus_run_test_mask) {
557 struct test_driver *t = &per_cpu_test_driver[cpu];
560 t->task = kthread_run(test_func, t, "vmalloc_test/%d", cpu);
562 if (!IS_ERR(t->task))
564 atomic_inc(&test_n_undone);
566 pr_err("Failed to start kthread for %d CPU\n", cpu);
570 * Now let the workers do their job.
572 up_write(&prepare_for_test_rwsem);
575 * Sleep quiet until all workers are done with 1 second
576 * interval. Since the test can take a lot of time we
577 * can run into a stack trace of the hung task. That is
578 * why we go with completion_timeout and HZ value.
581 ret = wait_for_completion_timeout(&test_all_done_comp, HZ);
584 for_each_cpu(cpu, &cpus_run_test_mask) {
585 struct test_driver *t = &per_cpu_test_driver[cpu];
588 if (!IS_ERR(t->task))
589 kthread_stop(t->task);
591 for (i = 0; i < ARRAY_SIZE(test_case_array); i++) {
592 if (!((run_test_mask & (1 << i)) >> i))
596 "Summary: %s passed: %d failed: %d repeat: %d loops: %d avg: %llu usec\n",
597 test_case_array[i].test_name,
598 per_cpu_test_data[cpu][i].test_passed,
599 per_cpu_test_data[cpu][i].test_failed,
600 test_repeat_count, test_loop_count,
601 per_cpu_test_data[cpu][i].time);
604 pr_info("All test took CPU%d=%lu cycles\n",
605 cpu, t->stop - t->start);
609 static int vmalloc_test_init(void)
611 do_concurrent_test();
612 return -EAGAIN; /* Fail will directly unload the module */
615 static void vmalloc_test_exit(void)
619 module_init(vmalloc_test_init)
620 module_exit(vmalloc_test_exit)
622 MODULE_LICENSE("GPL");
623 MODULE_AUTHOR("Uladzislau Rezki");
624 MODULE_DESCRIPTION("vmalloc test module");