1 // SPDX-License-Identifier: GPL-2.0+ 1 // SPDX-License-Identifier: GPL-2.0+
2 // 2 //
3 // Scalability test comparing RCU vs other mec 3 // Scalability test comparing RCU vs other mechanisms
4 // for acquiring references on objects. 4 // for acquiring references on objects.
5 // 5 //
6 // Copyright (C) Google, 2020. 6 // Copyright (C) Google, 2020.
7 // 7 //
8 // Author: Joel Fernandes <joel@joelfernandes. 8 // Author: Joel Fernandes <joel@joelfernandes.org>
9 9
10 #define pr_fmt(fmt) fmt 10 #define pr_fmt(fmt) fmt
11 11
12 #include <linux/atomic.h> 12 #include <linux/atomic.h>
13 #include <linux/bitops.h> 13 #include <linux/bitops.h>
14 #include <linux/completion.h> 14 #include <linux/completion.h>
15 #include <linux/cpu.h> 15 #include <linux/cpu.h>
16 #include <linux/delay.h> 16 #include <linux/delay.h>
17 #include <linux/err.h> 17 #include <linux/err.h>
18 #include <linux/init.h> 18 #include <linux/init.h>
19 #include <linux/interrupt.h> 19 #include <linux/interrupt.h>
20 #include <linux/kthread.h> 20 #include <linux/kthread.h>
21 #include <linux/kernel.h> 21 #include <linux/kernel.h>
22 #include <linux/mm.h> 22 #include <linux/mm.h>
23 #include <linux/module.h> 23 #include <linux/module.h>
24 #include <linux/moduleparam.h> 24 #include <linux/moduleparam.h>
25 #include <linux/notifier.h> 25 #include <linux/notifier.h>
26 #include <linux/percpu.h> 26 #include <linux/percpu.h>
27 #include <linux/rcupdate.h> 27 #include <linux/rcupdate.h>
28 #include <linux/rcupdate_trace.h> 28 #include <linux/rcupdate_trace.h>
29 #include <linux/reboot.h> 29 #include <linux/reboot.h>
30 #include <linux/sched.h> 30 #include <linux/sched.h>
31 #include <linux/seq_buf.h> <<
32 #include <linux/spinlock.h> 31 #include <linux/spinlock.h>
33 #include <linux/smp.h> 32 #include <linux/smp.h>
34 #include <linux/stat.h> 33 #include <linux/stat.h>
35 #include <linux/srcu.h> 34 #include <linux/srcu.h>
36 #include <linux/slab.h> 35 #include <linux/slab.h>
37 #include <linux/torture.h> 36 #include <linux/torture.h>
38 #include <linux/types.h> 37 #include <linux/types.h>
39 38
40 #include "rcu.h" 39 #include "rcu.h"
41 40
42 #define SCALE_FLAG "-ref-scale: " 41 #define SCALE_FLAG "-ref-scale: "
43 42
44 #define SCALEOUT(s, x...) \ 43 #define SCALEOUT(s, x...) \
45 pr_alert("%s" SCALE_FLAG s, scale_type 44 pr_alert("%s" SCALE_FLAG s, scale_type, ## x)
46 45
47 #define VERBOSE_SCALEOUT(s, x...) \ 46 #define VERBOSE_SCALEOUT(s, x...) \
48 do { \ 47 do { \
49 if (verbose) \ 48 if (verbose) \
50 pr_alert("%s" SCALE_FL 49 pr_alert("%s" SCALE_FLAG s "\n", scale_type, ## x); \
51 } while (0) 50 } while (0)
52 51
53 static atomic_t verbose_batch_ctr; 52 static atomic_t verbose_batch_ctr;
54 53
55 #define VERBOSE_SCALEOUT_BATCH(s, x...) 54 #define VERBOSE_SCALEOUT_BATCH(s, x...) \
56 do { 55 do { \
57 if (verbose && 56 if (verbose && \
58 (verbose_batched <= 0 || 57 (verbose_batched <= 0 || \
59 !(atomic_inc_return(&verbose_batc 58 !(atomic_inc_return(&verbose_batch_ctr) % verbose_batched))) { \
60 schedule_timeout_uninterruptib 59 schedule_timeout_uninterruptible(1); \
61 pr_alert("%s" SCALE_FLAG s "\n 60 pr_alert("%s" SCALE_FLAG s "\n", scale_type, ## x); \
62 } 61 } \
63 } while (0) 62 } while (0)
64 63
65 #define SCALEOUT_ERRSTRING(s, x...) pr_alert(" 64 #define SCALEOUT_ERRSTRING(s, x...) pr_alert("%s" SCALE_FLAG "!!! " s "\n", scale_type, ## x)
66 65
67 MODULE_DESCRIPTION("Scalability test for objec <<
68 MODULE_LICENSE("GPL"); 66 MODULE_LICENSE("GPL");
69 MODULE_AUTHOR("Joel Fernandes (Google) <joel@j 67 MODULE_AUTHOR("Joel Fernandes (Google) <joel@joelfernandes.org>");
70 68
71 static char *scale_type = "rcu"; 69 static char *scale_type = "rcu";
72 module_param(scale_type, charp, 0444); 70 module_param(scale_type, charp, 0444);
73 MODULE_PARM_DESC(scale_type, "Type of test (rc 71 MODULE_PARM_DESC(scale_type, "Type of test (rcu, srcu, refcnt, rwsem, rwlock.");
74 72
75 torture_param(int, verbose, 0, "Enable verbose 73 torture_param(int, verbose, 0, "Enable verbose debugging printk()s");
76 torture_param(int, verbose_batched, 0, "Batch 74 torture_param(int, verbose_batched, 0, "Batch verbose debugging printk()s");
77 75
78 // Wait until there are multiple CPUs before s 76 // Wait until there are multiple CPUs before starting test.
79 torture_param(int, holdoff, IS_BUILTIN(CONFIG_ 77 torture_param(int, holdoff, IS_BUILTIN(CONFIG_RCU_REF_SCALE_TEST) ? 10 : 0,
80 "Holdoff time before test start 78 "Holdoff time before test start (s)");
81 // Number of typesafe_lookup structures, that <<
82 torture_param(long, lookup_instances, 0, "Numb <<
83 // Number of loops per experiment, all readers 79 // Number of loops per experiment, all readers execute operations concurrently.
84 torture_param(long, loops, 10000, "Number of l 80 torture_param(long, loops, 10000, "Number of loops per experiment.");
85 // Number of readers, with -1 defaulting to ab 81 // Number of readers, with -1 defaulting to about 75% of the CPUs.
86 torture_param(int, nreaders, -1, "Number of re 82 torture_param(int, nreaders, -1, "Number of readers, -1 for 75% of CPUs.");
87 // Number of runs. 83 // Number of runs.
88 torture_param(int, nruns, 30, "Number of exper 84 torture_param(int, nruns, 30, "Number of experiments to run.");
89 // Reader delay in nanoseconds, 0 for no delay 85 // Reader delay in nanoseconds, 0 for no delay.
90 torture_param(int, readdelay, 0, "Read-side de 86 torture_param(int, readdelay, 0, "Read-side delay in nanoseconds.");
91 87
92 #ifdef MODULE 88 #ifdef MODULE
93 # define REFSCALE_SHUTDOWN 0 89 # define REFSCALE_SHUTDOWN 0
94 #else 90 #else
95 # define REFSCALE_SHUTDOWN 1 91 # define REFSCALE_SHUTDOWN 1
96 #endif 92 #endif
97 93
98 torture_param(bool, shutdown, REFSCALE_SHUTDOW 94 torture_param(bool, shutdown, REFSCALE_SHUTDOWN,
99 "Shutdown at end of scalability 95 "Shutdown at end of scalability tests.");
100 96
101 struct reader_task { 97 struct reader_task {
102 struct task_struct *task; 98 struct task_struct *task;
103 int start_reader; 99 int start_reader;
104 wait_queue_head_t wq; 100 wait_queue_head_t wq;
105 u64 last_duration_ns; 101 u64 last_duration_ns;
106 }; 102 };
107 103
108 static struct task_struct *shutdown_task; 104 static struct task_struct *shutdown_task;
109 static wait_queue_head_t shutdown_wq; 105 static wait_queue_head_t shutdown_wq;
110 106
111 static struct task_struct *main_task; 107 static struct task_struct *main_task;
112 static wait_queue_head_t main_wq; 108 static wait_queue_head_t main_wq;
113 static int shutdown_start; 109 static int shutdown_start;
114 110
115 static struct reader_task *reader_tasks; 111 static struct reader_task *reader_tasks;
116 112
117 // Number of readers that are part of the curr 113 // Number of readers that are part of the current experiment.
118 static atomic_t nreaders_exp; 114 static atomic_t nreaders_exp;
119 115
120 // Use to wait for all threads to start. 116 // Use to wait for all threads to start.
121 static atomic_t n_init; 117 static atomic_t n_init;
122 static atomic_t n_started; 118 static atomic_t n_started;
123 static atomic_t n_warmedup; 119 static atomic_t n_warmedup;
124 static atomic_t n_cooleddown; 120 static atomic_t n_cooleddown;
125 121
126 // Track which experiment is currently running 122 // Track which experiment is currently running.
127 static int exp_idx; 123 static int exp_idx;
128 124
129 // Operations vector for selecting different t 125 // Operations vector for selecting different types of tests.
130 struct ref_scale_ops { 126 struct ref_scale_ops {
131 bool (*init)(void); !! 127 void (*init)(void);
132 void (*cleanup)(void); 128 void (*cleanup)(void);
133 void (*readsection)(const int nloops); 129 void (*readsection)(const int nloops);
134 void (*delaysection)(const int nloops, 130 void (*delaysection)(const int nloops, const int udl, const int ndl);
135 const char *name; 131 const char *name;
136 }; 132 };
137 133
138 static const struct ref_scale_ops *cur_ops; !! 134 static struct ref_scale_ops *cur_ops;
139 135
140 static void un_delay(const int udl, const int 136 static void un_delay(const int udl, const int ndl)
141 { 137 {
142 if (udl) 138 if (udl)
143 udelay(udl); 139 udelay(udl);
144 if (ndl) 140 if (ndl)
145 ndelay(ndl); 141 ndelay(ndl);
146 } 142 }
147 143
148 static void ref_rcu_read_section(const int nlo 144 static void ref_rcu_read_section(const int nloops)
149 { 145 {
150 int i; 146 int i;
151 147
152 for (i = nloops; i >= 0; i--) { 148 for (i = nloops; i >= 0; i--) {
153 rcu_read_lock(); 149 rcu_read_lock();
154 rcu_read_unlock(); 150 rcu_read_unlock();
155 } 151 }
156 } 152 }
157 153
158 static void ref_rcu_delay_section(const int nl 154 static void ref_rcu_delay_section(const int nloops, const int udl, const int ndl)
159 { 155 {
160 int i; 156 int i;
161 157
162 for (i = nloops; i >= 0; i--) { 158 for (i = nloops; i >= 0; i--) {
163 rcu_read_lock(); 159 rcu_read_lock();
164 un_delay(udl, ndl); 160 un_delay(udl, ndl);
165 rcu_read_unlock(); 161 rcu_read_unlock();
166 } 162 }
167 } 163 }
168 164
169 static bool rcu_sync_scale_init(void) !! 165 static void rcu_sync_scale_init(void)
170 { 166 {
171 return true; <<
172 } 167 }
173 168
174 static const struct ref_scale_ops rcu_ops = { !! 169 static struct ref_scale_ops rcu_ops = {
175 .init = rcu_sync_scale_init, 170 .init = rcu_sync_scale_init,
176 .readsection = ref_rcu_read_section 171 .readsection = ref_rcu_read_section,
177 .delaysection = ref_rcu_delay_sectio 172 .delaysection = ref_rcu_delay_section,
178 .name = "rcu" 173 .name = "rcu"
179 }; 174 };
180 175
181 // Definitions for SRCU ref scale testing. 176 // Definitions for SRCU ref scale testing.
182 DEFINE_STATIC_SRCU(srcu_refctl_scale); 177 DEFINE_STATIC_SRCU(srcu_refctl_scale);
183 static struct srcu_struct *srcu_ctlp = &srcu_r 178 static struct srcu_struct *srcu_ctlp = &srcu_refctl_scale;
184 179
185 static void srcu_ref_scale_read_section(const 180 static void srcu_ref_scale_read_section(const int nloops)
186 { 181 {
187 int i; 182 int i;
188 int idx; 183 int idx;
189 184
190 for (i = nloops; i >= 0; i--) { 185 for (i = nloops; i >= 0; i--) {
191 idx = srcu_read_lock(srcu_ctlp 186 idx = srcu_read_lock(srcu_ctlp);
192 srcu_read_unlock(srcu_ctlp, id 187 srcu_read_unlock(srcu_ctlp, idx);
193 } 188 }
194 } 189 }
195 190
196 static void srcu_ref_scale_delay_section(const 191 static void srcu_ref_scale_delay_section(const int nloops, const int udl, const int ndl)
197 { 192 {
198 int i; 193 int i;
199 int idx; 194 int idx;
200 195
201 for (i = nloops; i >= 0; i--) { 196 for (i = nloops; i >= 0; i--) {
202 idx = srcu_read_lock(srcu_ctlp 197 idx = srcu_read_lock(srcu_ctlp);
203 un_delay(udl, ndl); 198 un_delay(udl, ndl);
204 srcu_read_unlock(srcu_ctlp, id 199 srcu_read_unlock(srcu_ctlp, idx);
205 } 200 }
206 } 201 }
207 202
208 static const struct ref_scale_ops srcu_ops = { !! 203 static struct ref_scale_ops srcu_ops = {
209 .init = rcu_sync_scale_init, 204 .init = rcu_sync_scale_init,
210 .readsection = srcu_ref_scale_read_ 205 .readsection = srcu_ref_scale_read_section,
211 .delaysection = srcu_ref_scale_delay 206 .delaysection = srcu_ref_scale_delay_section,
212 .name = "srcu" 207 .name = "srcu"
213 }; 208 };
214 209
215 #ifdef CONFIG_TASKS_RCU <<
216 <<
217 // Definitions for RCU Tasks ref scale testing 210 // Definitions for RCU Tasks ref scale testing: Empty read markers.
218 // These definitions also work for RCU Rude re 211 // These definitions also work for RCU Rude readers.
219 static void rcu_tasks_ref_scale_read_section(c 212 static void rcu_tasks_ref_scale_read_section(const int nloops)
220 { 213 {
221 int i; 214 int i;
222 215
223 for (i = nloops; i >= 0; i--) 216 for (i = nloops; i >= 0; i--)
224 continue; 217 continue;
225 } 218 }
226 219
227 static void rcu_tasks_ref_scale_delay_section( 220 static void rcu_tasks_ref_scale_delay_section(const int nloops, const int udl, const int ndl)
228 { 221 {
229 int i; 222 int i;
230 223
231 for (i = nloops; i >= 0; i--) 224 for (i = nloops; i >= 0; i--)
232 un_delay(udl, ndl); 225 un_delay(udl, ndl);
233 } 226 }
234 227
235 static const struct ref_scale_ops rcu_tasks_op !! 228 static struct ref_scale_ops rcu_tasks_ops = {
236 .init = rcu_sync_scale_init, 229 .init = rcu_sync_scale_init,
237 .readsection = rcu_tasks_ref_scale_ 230 .readsection = rcu_tasks_ref_scale_read_section,
238 .delaysection = rcu_tasks_ref_scale_ 231 .delaysection = rcu_tasks_ref_scale_delay_section,
239 .name = "rcu-tasks" 232 .name = "rcu-tasks"
240 }; 233 };
241 234
242 #define RCU_TASKS_OPS &rcu_tasks_ops, <<
243 <<
244 #else // #ifdef CONFIG_TASKS_RCU <<
245 <<
246 #define RCU_TASKS_OPS <<
247 <<
248 #endif // #else // #ifdef CONFIG_TASKS_RCU <<
249 <<
250 #ifdef CONFIG_TASKS_TRACE_RCU <<
251 <<
252 // Definitions for RCU Tasks Trace ref scale t 235 // Definitions for RCU Tasks Trace ref scale testing.
253 static void rcu_trace_ref_scale_read_section(c 236 static void rcu_trace_ref_scale_read_section(const int nloops)
254 { 237 {
255 int i; 238 int i;
256 239
257 for (i = nloops; i >= 0; i--) { 240 for (i = nloops; i >= 0; i--) {
258 rcu_read_lock_trace(); 241 rcu_read_lock_trace();
259 rcu_read_unlock_trace(); 242 rcu_read_unlock_trace();
260 } 243 }
261 } 244 }
262 245
263 static void rcu_trace_ref_scale_delay_section( 246 static void rcu_trace_ref_scale_delay_section(const int nloops, const int udl, const int ndl)
264 { 247 {
265 int i; 248 int i;
266 249
267 for (i = nloops; i >= 0; i--) { 250 for (i = nloops; i >= 0; i--) {
268 rcu_read_lock_trace(); 251 rcu_read_lock_trace();
269 un_delay(udl, ndl); 252 un_delay(udl, ndl);
270 rcu_read_unlock_trace(); 253 rcu_read_unlock_trace();
271 } 254 }
272 } 255 }
273 256
274 static const struct ref_scale_ops rcu_trace_op !! 257 static struct ref_scale_ops rcu_trace_ops = {
275 .init = rcu_sync_scale_init, 258 .init = rcu_sync_scale_init,
276 .readsection = rcu_trace_ref_scale_ 259 .readsection = rcu_trace_ref_scale_read_section,
277 .delaysection = rcu_trace_ref_scale_ 260 .delaysection = rcu_trace_ref_scale_delay_section,
278 .name = "rcu-trace" 261 .name = "rcu-trace"
279 }; 262 };
280 263
281 #define RCU_TRACE_OPS &rcu_trace_ops, <<
282 <<
283 #else // #ifdef CONFIG_TASKS_TRACE_RCU <<
284 <<
285 #define RCU_TRACE_OPS <<
286 <<
287 #endif // #else // #ifdef CONFIG_TASKS_TRACE_R <<
288 <<
289 // Definitions for reference count 264 // Definitions for reference count
290 static atomic_t refcnt; 265 static atomic_t refcnt;
291 266
292 static void ref_refcnt_section(const int nloop 267 static void ref_refcnt_section(const int nloops)
293 { 268 {
294 int i; 269 int i;
295 270
296 for (i = nloops; i >= 0; i--) { 271 for (i = nloops; i >= 0; i--) {
297 atomic_inc(&refcnt); 272 atomic_inc(&refcnt);
298 atomic_dec(&refcnt); 273 atomic_dec(&refcnt);
299 } 274 }
300 } 275 }
301 276
302 static void ref_refcnt_delay_section(const int 277 static void ref_refcnt_delay_section(const int nloops, const int udl, const int ndl)
303 { 278 {
304 int i; 279 int i;
305 280
306 for (i = nloops; i >= 0; i--) { 281 for (i = nloops; i >= 0; i--) {
307 atomic_inc(&refcnt); 282 atomic_inc(&refcnt);
308 un_delay(udl, ndl); 283 un_delay(udl, ndl);
309 atomic_dec(&refcnt); 284 atomic_dec(&refcnt);
310 } 285 }
311 } 286 }
312 287
313 static const struct ref_scale_ops refcnt_ops = !! 288 static struct ref_scale_ops refcnt_ops = {
314 .init = rcu_sync_scale_init, 289 .init = rcu_sync_scale_init,
315 .readsection = ref_refcnt_section, 290 .readsection = ref_refcnt_section,
316 .delaysection = ref_refcnt_delay_sec 291 .delaysection = ref_refcnt_delay_section,
317 .name = "refcnt" 292 .name = "refcnt"
318 }; 293 };
319 294
320 // Definitions for rwlock 295 // Definitions for rwlock
321 static rwlock_t test_rwlock; 296 static rwlock_t test_rwlock;
322 297
323 static bool ref_rwlock_init(void) !! 298 static void ref_rwlock_init(void)
324 { 299 {
325 rwlock_init(&test_rwlock); 300 rwlock_init(&test_rwlock);
326 return true; <<
327 } 301 }
328 302
329 static void ref_rwlock_section(const int nloop 303 static void ref_rwlock_section(const int nloops)
330 { 304 {
331 int i; 305 int i;
332 306
333 for (i = nloops; i >= 0; i--) { 307 for (i = nloops; i >= 0; i--) {
334 read_lock(&test_rwlock); 308 read_lock(&test_rwlock);
335 read_unlock(&test_rwlock); 309 read_unlock(&test_rwlock);
336 } 310 }
337 } 311 }
338 312
339 static void ref_rwlock_delay_section(const int 313 static void ref_rwlock_delay_section(const int nloops, const int udl, const int ndl)
340 { 314 {
341 int i; 315 int i;
342 316
343 for (i = nloops; i >= 0; i--) { 317 for (i = nloops; i >= 0; i--) {
344 read_lock(&test_rwlock); 318 read_lock(&test_rwlock);
345 un_delay(udl, ndl); 319 un_delay(udl, ndl);
346 read_unlock(&test_rwlock); 320 read_unlock(&test_rwlock);
347 } 321 }
348 } 322 }
349 323
350 static const struct ref_scale_ops rwlock_ops = !! 324 static struct ref_scale_ops rwlock_ops = {
351 .init = ref_rwlock_init, 325 .init = ref_rwlock_init,
352 .readsection = ref_rwlock_section, 326 .readsection = ref_rwlock_section,
353 .delaysection = ref_rwlock_delay_sec 327 .delaysection = ref_rwlock_delay_section,
354 .name = "rwlock" 328 .name = "rwlock"
355 }; 329 };
356 330
357 // Definitions for rwsem 331 // Definitions for rwsem
358 static struct rw_semaphore test_rwsem; 332 static struct rw_semaphore test_rwsem;
359 333
360 static bool ref_rwsem_init(void) !! 334 static void ref_rwsem_init(void)
361 { 335 {
362 init_rwsem(&test_rwsem); 336 init_rwsem(&test_rwsem);
363 return true; <<
364 } 337 }
365 338
366 static void ref_rwsem_section(const int nloops 339 static void ref_rwsem_section(const int nloops)
367 { 340 {
368 int i; 341 int i;
369 342
370 for (i = nloops; i >= 0; i--) { 343 for (i = nloops; i >= 0; i--) {
371 down_read(&test_rwsem); 344 down_read(&test_rwsem);
372 up_read(&test_rwsem); 345 up_read(&test_rwsem);
373 } 346 }
374 } 347 }
375 348
376 static void ref_rwsem_delay_section(const int 349 static void ref_rwsem_delay_section(const int nloops, const int udl, const int ndl)
377 { 350 {
378 int i; 351 int i;
379 352
380 for (i = nloops; i >= 0; i--) { 353 for (i = nloops; i >= 0; i--) {
381 down_read(&test_rwsem); 354 down_read(&test_rwsem);
382 un_delay(udl, ndl); 355 un_delay(udl, ndl);
383 up_read(&test_rwsem); 356 up_read(&test_rwsem);
384 } 357 }
385 } 358 }
386 359
387 static const struct ref_scale_ops rwsem_ops = !! 360 static struct ref_scale_ops rwsem_ops = {
388 .init = ref_rwsem_init, 361 .init = ref_rwsem_init,
389 .readsection = ref_rwsem_section, 362 .readsection = ref_rwsem_section,
390 .delaysection = ref_rwsem_delay_sect 363 .delaysection = ref_rwsem_delay_section,
391 .name = "rwsem" 364 .name = "rwsem"
392 }; 365 };
393 366
394 // Definitions for global spinlock 367 // Definitions for global spinlock
395 static DEFINE_RAW_SPINLOCK(test_lock); !! 368 static DEFINE_SPINLOCK(test_lock);
396 369
397 static void ref_lock_section(const int nloops) 370 static void ref_lock_section(const int nloops)
398 { 371 {
399 int i; 372 int i;
400 373
401 preempt_disable(); 374 preempt_disable();
402 for (i = nloops; i >= 0; i--) { 375 for (i = nloops; i >= 0; i--) {
403 raw_spin_lock(&test_lock); !! 376 spin_lock(&test_lock);
404 raw_spin_unlock(&test_lock); !! 377 spin_unlock(&test_lock);
405 } 378 }
406 preempt_enable(); 379 preempt_enable();
407 } 380 }
408 381
409 static void ref_lock_delay_section(const int n 382 static void ref_lock_delay_section(const int nloops, const int udl, const int ndl)
410 { 383 {
411 int i; 384 int i;
412 385
413 preempt_disable(); 386 preempt_disable();
414 for (i = nloops; i >= 0; i--) { 387 for (i = nloops; i >= 0; i--) {
415 raw_spin_lock(&test_lock); !! 388 spin_lock(&test_lock);
416 un_delay(udl, ndl); 389 un_delay(udl, ndl);
417 raw_spin_unlock(&test_lock); !! 390 spin_unlock(&test_lock);
418 } 391 }
419 preempt_enable(); 392 preempt_enable();
420 } 393 }
421 394
422 static const struct ref_scale_ops lock_ops = { !! 395 static struct ref_scale_ops lock_ops = {
423 .readsection = ref_lock_section, 396 .readsection = ref_lock_section,
424 .delaysection = ref_lock_delay_secti 397 .delaysection = ref_lock_delay_section,
425 .name = "lock" 398 .name = "lock"
426 }; 399 };
427 400
428 // Definitions for global irq-save spinlock 401 // Definitions for global irq-save spinlock
429 402
430 static void ref_lock_irq_section(const int nlo 403 static void ref_lock_irq_section(const int nloops)
431 { 404 {
432 unsigned long flags; 405 unsigned long flags;
433 int i; 406 int i;
434 407
435 preempt_disable(); 408 preempt_disable();
436 for (i = nloops; i >= 0; i--) { 409 for (i = nloops; i >= 0; i--) {
437 raw_spin_lock_irqsave(&test_lo !! 410 spin_lock_irqsave(&test_lock, flags);
438 raw_spin_unlock_irqrestore(&te !! 411 spin_unlock_irqrestore(&test_lock, flags);
439 } 412 }
440 preempt_enable(); 413 preempt_enable();
441 } 414 }
442 415
443 static void ref_lock_irq_delay_section(const i 416 static void ref_lock_irq_delay_section(const int nloops, const int udl, const int ndl)
444 { 417 {
445 unsigned long flags; 418 unsigned long flags;
446 int i; 419 int i;
447 420
448 preempt_disable(); 421 preempt_disable();
449 for (i = nloops; i >= 0; i--) { 422 for (i = nloops; i >= 0; i--) {
450 raw_spin_lock_irqsave(&test_lo !! 423 spin_lock_irqsave(&test_lock, flags);
451 un_delay(udl, ndl); 424 un_delay(udl, ndl);
452 raw_spin_unlock_irqrestore(&te !! 425 spin_unlock_irqrestore(&test_lock, flags);
453 } 426 }
454 preempt_enable(); 427 preempt_enable();
455 } 428 }
456 429
457 static const struct ref_scale_ops lock_irq_ops !! 430 static struct ref_scale_ops lock_irq_ops = {
458 .readsection = ref_lock_irq_section 431 .readsection = ref_lock_irq_section,
459 .delaysection = ref_lock_irq_delay_s 432 .delaysection = ref_lock_irq_delay_section,
460 .name = "lock-irq" 433 .name = "lock-irq"
461 }; 434 };
462 435
463 // Definitions acquire-release. 436 // Definitions acquire-release.
464 static DEFINE_PER_CPU(unsigned long, test_acqr 437 static DEFINE_PER_CPU(unsigned long, test_acqrel);
465 438
466 static void ref_acqrel_section(const int nloop 439 static void ref_acqrel_section(const int nloops)
467 { 440 {
468 unsigned long x; 441 unsigned long x;
469 int i; 442 int i;
470 443
471 preempt_disable(); 444 preempt_disable();
472 for (i = nloops; i >= 0; i--) { 445 for (i = nloops; i >= 0; i--) {
473 x = smp_load_acquire(this_cpu_ 446 x = smp_load_acquire(this_cpu_ptr(&test_acqrel));
474 smp_store_release(this_cpu_ptr 447 smp_store_release(this_cpu_ptr(&test_acqrel), x + 1);
475 } 448 }
476 preempt_enable(); 449 preempt_enable();
477 } 450 }
478 451
479 static void ref_acqrel_delay_section(const int 452 static void ref_acqrel_delay_section(const int nloops, const int udl, const int ndl)
480 { 453 {
481 unsigned long x; 454 unsigned long x;
482 int i; 455 int i;
483 456
484 preempt_disable(); 457 preempt_disable();
485 for (i = nloops; i >= 0; i--) { 458 for (i = nloops; i >= 0; i--) {
486 x = smp_load_acquire(this_cpu_ 459 x = smp_load_acquire(this_cpu_ptr(&test_acqrel));
487 un_delay(udl, ndl); 460 un_delay(udl, ndl);
488 smp_store_release(this_cpu_ptr 461 smp_store_release(this_cpu_ptr(&test_acqrel), x + 1);
489 } 462 }
490 preempt_enable(); 463 preempt_enable();
491 } 464 }
492 465
493 static const struct ref_scale_ops acqrel_ops = !! 466 static struct ref_scale_ops acqrel_ops = {
494 .readsection = ref_acqrel_section, 467 .readsection = ref_acqrel_section,
495 .delaysection = ref_acqrel_delay_sec 468 .delaysection = ref_acqrel_delay_section,
496 .name = "acqrel" 469 .name = "acqrel"
497 }; 470 };
498 471
499 static volatile u64 stopopts; 472 static volatile u64 stopopts;
500 473
501 static void ref_clock_section(const int nloops 474 static void ref_clock_section(const int nloops)
502 { 475 {
503 u64 x = 0; 476 u64 x = 0;
504 int i; 477 int i;
505 478
506 preempt_disable(); 479 preempt_disable();
507 for (i = nloops; i >= 0; i--) 480 for (i = nloops; i >= 0; i--)
508 x += ktime_get_real_fast_ns(); 481 x += ktime_get_real_fast_ns();
509 preempt_enable(); 482 preempt_enable();
510 stopopts = x; 483 stopopts = x;
511 } 484 }
512 485
513 static void ref_clock_delay_section(const int 486 static void ref_clock_delay_section(const int nloops, const int udl, const int ndl)
514 { 487 {
515 u64 x = 0; 488 u64 x = 0;
516 int i; 489 int i;
517 490
518 preempt_disable(); 491 preempt_disable();
519 for (i = nloops; i >= 0; i--) { 492 for (i = nloops; i >= 0; i--) {
520 x += ktime_get_real_fast_ns(); 493 x += ktime_get_real_fast_ns();
521 un_delay(udl, ndl); 494 un_delay(udl, ndl);
522 } 495 }
523 preempt_enable(); 496 preempt_enable();
524 stopopts = x; 497 stopopts = x;
525 } 498 }
526 499
527 static const struct ref_scale_ops clock_ops = !! 500 static struct ref_scale_ops clock_ops = {
528 .readsection = ref_clock_section, 501 .readsection = ref_clock_section,
529 .delaysection = ref_clock_delay_sect 502 .delaysection = ref_clock_delay_section,
530 .name = "clock" 503 .name = "clock"
531 }; 504 };
532 505
533 static void ref_jiffies_section(const int nloo <<
534 { <<
535 u64 x = 0; <<
536 int i; <<
537 <<
538 preempt_disable(); <<
539 for (i = nloops; i >= 0; i--) <<
540 x += jiffies; <<
541 preempt_enable(); <<
542 stopopts = x; <<
543 } <<
544 <<
545 static void ref_jiffies_delay_section(const in <<
546 { <<
547 u64 x = 0; <<
548 int i; <<
549 <<
550 preempt_disable(); <<
551 for (i = nloops; i >= 0; i--) { <<
552 x += jiffies; <<
553 un_delay(udl, ndl); <<
554 } <<
555 preempt_enable(); <<
556 stopopts = x; <<
557 } <<
558 <<
559 static const struct ref_scale_ops jiffies_ops <<
560 .readsection = ref_jiffies_section, <<
561 .delaysection = ref_jiffies_delay_se <<
562 .name = "jiffies" <<
563 }; <<
564 <<
565 ////////////////////////////////////////////// <<
566 // <<
567 // Methods leveraging SLAB_TYPESAFE_BY_RCU. <<
568 // <<
569 <<
570 // Item to look up in a typesafe manner. Arra <<
571 struct refscale_typesafe { <<
572 atomic_t rts_refctr; // Used by all f <<
573 spinlock_t rts_lock; <<
574 seqlock_t rts_seqlock; <<
575 unsigned int a; <<
576 unsigned int b; <<
577 }; <<
578 <<
579 static struct kmem_cache *typesafe_kmem_cachep <<
580 static struct refscale_typesafe **rtsarray; <<
581 static long rtsarray_size; <<
582 static DEFINE_TORTURE_RANDOM_PERCPU(refscale_r <<
583 static bool (*rts_acquire)(struct refscale_typ <<
584 static bool (*rts_release)(struct refscale_typ <<
585 <<
586 // Conditionally acquire an explicit in-struct <<
587 static bool typesafe_ref_acquire(struct refsca <<
588 { <<
589 return atomic_inc_not_zero(&rtsp->rts_ <<
590 } <<
591 <<
592 // Unconditionally release an explicit in-stru <<
593 static bool typesafe_ref_release(struct refsca <<
594 { <<
595 if (!atomic_dec_return(&rtsp->rts_refc <<
596 WRITE_ONCE(rtsp->a, rtsp->a + <<
597 kmem_cache_free(typesafe_kmem_ <<
598 } <<
599 return true; <<
600 } <<
601 <<
602 // Unconditionally acquire an explicit in-stru <<
603 static bool typesafe_lock_acquire(struct refsc <<
604 { <<
605 spin_lock(&rtsp->rts_lock); <<
606 return true; <<
607 } <<
608 <<
609 // Unconditionally release an explicit in-stru <<
610 static bool typesafe_lock_release(struct refsc <<
611 { <<
612 spin_unlock(&rtsp->rts_lock); <<
613 return true; <<
614 } <<
615 <<
616 // Unconditionally acquire an explicit in-stru <<
617 static bool typesafe_seqlock_acquire(struct re <<
618 { <<
619 *start = read_seqbegin(&rtsp->rts_seql <<
620 return true; <<
621 } <<
622 <<
623 // Conditionally release an explicit in-struct <<
624 // true if this release was successful, that i <<
625 static bool typesafe_seqlock_release(struct re <<
626 { <<
627 return !read_seqretry(&rtsp->rts_seqlo <<
628 } <<
629 <<
630 // Do a read-side critical section with the sp <<
631 // microseconds and nanoseconds inserted so as <<
632 // of failure. <<
633 static void typesafe_delay_section(const int n <<
634 { <<
635 unsigned int a; <<
636 unsigned int b; <<
637 int i; <<
638 long idx; <<
639 struct refscale_typesafe *rtsp; <<
640 unsigned int start; <<
641 <<
642 for (i = nloops; i >= 0; i--) { <<
643 preempt_disable(); <<
644 idx = torture_random(this_cpu_ <<
645 preempt_enable(); <<
646 retry: <<
647 rcu_read_lock(); <<
648 rtsp = rcu_dereference(rtsarra <<
649 a = READ_ONCE(rtsp->a); <<
650 if (!rts_acquire(rtsp, &start) <<
651 rcu_read_unlock(); <<
652 goto retry; <<
653 } <<
654 if (a != READ_ONCE(rtsp->a)) { <<
655 (void)rts_release(rtsp <<
656 rcu_read_unlock(); <<
657 goto retry; <<
658 } <<
659 un_delay(udl, ndl); <<
660 b = READ_ONCE(rtsp->a); <<
661 // Remember, seqlock read-side <<
662 if (!rts_release(rtsp, start)) <<
663 rcu_read_unlock(); <<
664 goto retry; <<
665 } <<
666 WARN_ONCE(a != b, "Re-read of <<
667 b = rtsp->b; <<
668 rcu_read_unlock(); <<
669 WARN_ON_ONCE(a * a != b); <<
670 } <<
671 } <<
672 <<
673 // Because the acquisition and release methods <<
674 // is no point in optimizing away the un_delay <<
675 // Thus simply define typesafe_read_section() <<
676 // typesafe_delay_section(). <<
677 static void typesafe_read_section(const int nl <<
678 { <<
679 typesafe_delay_section(nloops, 0, 0); <<
680 } <<
681 <<
682 // Allocate and initialize one refscale_typesa <<
683 static struct refscale_typesafe *typesafe_allo <<
684 { <<
685 struct refscale_typesafe *rtsp; <<
686 <<
687 rtsp = kmem_cache_alloc(typesafe_kmem_ <<
688 if (!rtsp) <<
689 return NULL; <<
690 atomic_set(&rtsp->rts_refctr, 1); <<
691 WRITE_ONCE(rtsp->a, rtsp->a + 1); <<
692 WRITE_ONCE(rtsp->b, rtsp->a * rtsp->a) <<
693 return rtsp; <<
694 } <<
695 <<
696 // Slab-allocator constructor for refscale_typ <<
697 // out of a new slab of system memory. <<
698 static void refscale_typesafe_ctor(void *rtsp_ <<
699 { <<
700 struct refscale_typesafe *rtsp = rtsp_ <<
701 <<
702 spin_lock_init(&rtsp->rts_lock); <<
703 seqlock_init(&rtsp->rts_seqlock); <<
704 preempt_disable(); <<
705 rtsp->a = torture_random(this_cpu_ptr( <<
706 preempt_enable(); <<
707 } <<
708 <<
709 static const struct ref_scale_ops typesafe_ref <<
710 static const struct ref_scale_ops typesafe_loc <<
711 static const struct ref_scale_ops typesafe_seq <<
712 <<
713 // Initialize for a typesafe test. <<
714 static bool typesafe_init(void) <<
715 { <<
716 long idx; <<
717 long si = lookup_instances; <<
718 <<
719 typesafe_kmem_cachep = kmem_cache_crea <<
720 <<
721 <<
722 if (!typesafe_kmem_cachep) <<
723 return false; <<
724 if (si < 0) <<
725 si = -si * nr_cpu_ids; <<
726 else if (si == 0) <<
727 si = nr_cpu_ids; <<
728 rtsarray_size = si; <<
729 rtsarray = kcalloc(si, sizeof(*rtsarra <<
730 if (!rtsarray) <<
731 return false; <<
732 for (idx = 0; idx < rtsarray_size; idx <<
733 rtsarray[idx] = typesafe_alloc <<
734 if (!rtsarray[idx]) <<
735 return false; <<
736 } <<
737 if (cur_ops == &typesafe_ref_ops) { <<
738 rts_acquire = typesafe_ref_acq <<
739 rts_release = typesafe_ref_rel <<
740 } else if (cur_ops == &typesafe_lock_o <<
741 rts_acquire = typesafe_lock_ac <<
742 rts_release = typesafe_lock_re <<
743 } else if (cur_ops == &typesafe_seqloc <<
744 rts_acquire = typesafe_seqlock <<
745 rts_release = typesafe_seqlock <<
746 } else { <<
747 WARN_ON_ONCE(1); <<
748 return false; <<
749 } <<
750 return true; <<
751 } <<
752 <<
753 // Clean up after a typesafe test. <<
754 static void typesafe_cleanup(void) <<
755 { <<
756 long idx; <<
757 <<
758 if (rtsarray) { <<
759 for (idx = 0; idx < rtsarray_s <<
760 kmem_cache_free(typesa <<
761 kfree(rtsarray); <<
762 rtsarray = NULL; <<
763 rtsarray_size = 0; <<
764 } <<
765 kmem_cache_destroy(typesafe_kmem_cache <<
766 typesafe_kmem_cachep = NULL; <<
767 rts_acquire = NULL; <<
768 rts_release = NULL; <<
769 } <<
770 <<
771 // The typesafe_init() function distinguishes <<
772 static const struct ref_scale_ops typesafe_ref <<
773 .init = typesafe_init, <<
774 .cleanup = typesafe_cleanup, <<
775 .readsection = typesafe_read_sectio <<
776 .delaysection = typesafe_delay_secti <<
777 .name = "typesafe_ref" <<
778 }; <<
779 <<
780 static const struct ref_scale_ops typesafe_loc <<
781 .init = typesafe_init, <<
782 .cleanup = typesafe_cleanup, <<
783 .readsection = typesafe_read_sectio <<
784 .delaysection = typesafe_delay_secti <<
785 .name = "typesafe_lock" <<
786 }; <<
787 <<
788 static const struct ref_scale_ops typesafe_seq <<
789 .init = typesafe_init, <<
790 .cleanup = typesafe_cleanup, <<
791 .readsection = typesafe_read_sectio <<
792 .delaysection = typesafe_delay_secti <<
793 .name = "typesafe_seqlock" <<
794 }; <<
795 <<
796 static void rcu_scale_one_reader(void) 506 static void rcu_scale_one_reader(void)
797 { 507 {
798 if (readdelay <= 0) 508 if (readdelay <= 0)
799 cur_ops->readsection(loops); 509 cur_ops->readsection(loops);
800 else 510 else
801 cur_ops->delaysection(loops, r 511 cur_ops->delaysection(loops, readdelay / 1000, readdelay % 1000);
802 } 512 }
803 513
804 // Reader kthread. Repeatedly does empty RCU 514 // Reader kthread. Repeatedly does empty RCU read-side
805 // critical section, minimizing update-side in 515 // critical section, minimizing update-side interference.
806 static int 516 static int
807 ref_scale_reader(void *arg) 517 ref_scale_reader(void *arg)
808 { 518 {
809 unsigned long flags; 519 unsigned long flags;
810 long me = (long)arg; 520 long me = (long)arg;
811 struct reader_task *rt = &(reader_task 521 struct reader_task *rt = &(reader_tasks[me]);
812 u64 start; 522 u64 start;
813 s64 duration; 523 s64 duration;
814 524
815 VERBOSE_SCALEOUT_BATCH("ref_scale_read 525 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: task started", me);
816 WARN_ON_ONCE(set_cpus_allowed_ptr(curr 526 WARN_ON_ONCE(set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids)));
817 set_user_nice(current, MAX_NICE); 527 set_user_nice(current, MAX_NICE);
818 atomic_inc(&n_init); 528 atomic_inc(&n_init);
819 if (holdoff) 529 if (holdoff)
820 schedule_timeout_interruptible 530 schedule_timeout_interruptible(holdoff * HZ);
821 repeat: 531 repeat:
822 VERBOSE_SCALEOUT_BATCH("ref_scale_read 532 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: waiting to start next experiment on cpu %d", me, raw_smp_processor_id());
823 533
824 // Wait for signal that this reader ca 534 // Wait for signal that this reader can start.
825 wait_event(rt->wq, (atomic_read(&nread 535 wait_event(rt->wq, (atomic_read(&nreaders_exp) && smp_load_acquire(&rt->start_reader)) ||
826 torture_must_stop() 536 torture_must_stop());
827 537
828 if (torture_must_stop()) 538 if (torture_must_stop())
829 goto end; 539 goto end;
830 540
831 // Make sure that the CPU is affinitiz 541 // Make sure that the CPU is affinitized appropriately during testing.
832 WARN_ON_ONCE(raw_smp_processor_id() != 542 WARN_ON_ONCE(raw_smp_processor_id() != me);
833 543
834 WRITE_ONCE(rt->start_reader, 0); 544 WRITE_ONCE(rt->start_reader, 0);
835 if (!atomic_dec_return(&n_started)) 545 if (!atomic_dec_return(&n_started))
836 while (atomic_read_acquire(&n_ 546 while (atomic_read_acquire(&n_started))
837 cpu_relax(); 547 cpu_relax();
838 548
839 VERBOSE_SCALEOUT_BATCH("ref_scale_read 549 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: experiment %d started", me, exp_idx);
840 550
841 551
842 // To reduce noise, do an initial cach 552 // To reduce noise, do an initial cache-warming invocation, check
843 // in, and then keep warming until eve 553 // in, and then keep warming until everyone has checked in.
844 rcu_scale_one_reader(); 554 rcu_scale_one_reader();
845 if (!atomic_dec_return(&n_warmedup)) 555 if (!atomic_dec_return(&n_warmedup))
846 while (atomic_read_acquire(&n_ 556 while (atomic_read_acquire(&n_warmedup))
847 rcu_scale_one_reader() 557 rcu_scale_one_reader();
848 // Also keep interrupts disabled. Thi 558 // Also keep interrupts disabled. This also has the effect
849 // of preventing entries into slow pat 559 // of preventing entries into slow path for rcu_read_unlock().
850 local_irq_save(flags); 560 local_irq_save(flags);
851 start = ktime_get_mono_fast_ns(); 561 start = ktime_get_mono_fast_ns();
852 562
853 rcu_scale_one_reader(); 563 rcu_scale_one_reader();
854 564
855 duration = ktime_get_mono_fast_ns() - 565 duration = ktime_get_mono_fast_ns() - start;
856 local_irq_restore(flags); 566 local_irq_restore(flags);
857 567
858 rt->last_duration_ns = WARN_ON_ONCE(du 568 rt->last_duration_ns = WARN_ON_ONCE(duration < 0) ? 0 : duration;
859 // To reduce runtime-skew noise, do ma 569 // To reduce runtime-skew noise, do maintain-load invocations until
860 // everyone is done. 570 // everyone is done.
861 if (!atomic_dec_return(&n_cooleddown)) 571 if (!atomic_dec_return(&n_cooleddown))
862 while (atomic_read_acquire(&n_ 572 while (atomic_read_acquire(&n_cooleddown))
863 rcu_scale_one_reader() 573 rcu_scale_one_reader();
864 574
865 if (atomic_dec_and_test(&nreaders_exp) 575 if (atomic_dec_and_test(&nreaders_exp))
866 wake_up(&main_wq); 576 wake_up(&main_wq);
867 577
868 VERBOSE_SCALEOUT_BATCH("ref_scale_read 578 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: experiment %d ended, (readers remaining=%d)",
869 me, exp_idx, a 579 me, exp_idx, atomic_read(&nreaders_exp));
870 580
871 if (!torture_must_stop()) 581 if (!torture_must_stop())
872 goto repeat; 582 goto repeat;
873 end: 583 end:
874 torture_kthread_stopping("ref_scale_re 584 torture_kthread_stopping("ref_scale_reader");
875 return 0; 585 return 0;
876 } 586 }
877 587
878 static void reset_readers(void) 588 static void reset_readers(void)
879 { 589 {
880 int i; 590 int i;
881 struct reader_task *rt; 591 struct reader_task *rt;
882 592
883 for (i = 0; i < nreaders; i++) { 593 for (i = 0; i < nreaders; i++) {
884 rt = &(reader_tasks[i]); 594 rt = &(reader_tasks[i]);
885 595
886 rt->last_duration_ns = 0; 596 rt->last_duration_ns = 0;
887 } 597 }
888 } 598 }
889 599
890 // Print the results of each reader and return 600 // Print the results of each reader and return the sum of all their durations.
891 static u64 process_durations(int n) 601 static u64 process_durations(int n)
892 { 602 {
893 int i; 603 int i;
894 struct reader_task *rt; 604 struct reader_task *rt;
895 struct seq_buf s; !! 605 char buf1[64];
896 char *buf; 606 char *buf;
897 u64 sum = 0; 607 u64 sum = 0;
898 608
899 buf = kmalloc(800 + 64, GFP_KERNEL); 609 buf = kmalloc(800 + 64, GFP_KERNEL);
900 if (!buf) 610 if (!buf)
901 return 0; 611 return 0;
902 seq_buf_init(&s, buf, 800 + 64); !! 612 buf[0] = 0;
903 !! 613 sprintf(buf, "Experiment #%d (Format: <THREAD-NUM>:<Total loop time in ns>)",
904 seq_buf_printf(&s, "Experiment #%d (Fo !! 614 exp_idx);
905 exp_idx); <<
906 615
907 for (i = 0; i < n && !torture_must_sto 616 for (i = 0; i < n && !torture_must_stop(); i++) {
908 rt = &(reader_tasks[i]); 617 rt = &(reader_tasks[i]);
>> 618 sprintf(buf1, "%d: %llu\t", i, rt->last_duration_ns);
909 619
910 if (i % 5 == 0) 620 if (i % 5 == 0)
911 seq_buf_putc(&s, '\n') !! 621 strcat(buf, "\n");
912 !! 622 if (strlen(buf) >= 800) {
913 if (seq_buf_used(&s) >= 800) { !! 623 pr_alert("%s", buf);
914 pr_alert("%s", seq_buf !! 624 buf[0] = 0;
915 seq_buf_clear(&s); <<
916 } 625 }
917 !! 626 strcat(buf, buf1);
918 seq_buf_printf(&s, "%d: %llu\t <<
919 627
920 sum += rt->last_duration_ns; 628 sum += rt->last_duration_ns;
921 } 629 }
922 pr_alert("%s\n", seq_buf_str(&s)); !! 630 pr_alert("%s\n", buf);
923 631
924 kfree(buf); 632 kfree(buf);
925 return sum; 633 return sum;
926 } 634 }
927 635
928 // The main_func is the main orchestrator, it 636 // The main_func is the main orchestrator, it performs a bunch of
929 // experiments. For every experiment, it orde 637 // experiments. For every experiment, it orders all the readers
930 // involved to start and waits for them to fin 638 // involved to start and waits for them to finish the experiment. It
931 // then reads their timestamps and starts the 639 // then reads their timestamps and starts the next experiment. Each
932 // experiment progresses from 1 concurrent rea 640 // experiment progresses from 1 concurrent reader to N of them at which
933 // point all the timestamps are printed. 641 // point all the timestamps are printed.
934 static int main_func(void *arg) 642 static int main_func(void *arg)
935 { 643 {
936 int exp, r; 644 int exp, r;
937 char buf1[64]; 645 char buf1[64];
938 char *buf; 646 char *buf;
939 u64 *result_avg; 647 u64 *result_avg;
940 648
941 set_cpus_allowed_ptr(current, cpumask_ 649 set_cpus_allowed_ptr(current, cpumask_of(nreaders % nr_cpu_ids));
942 set_user_nice(current, MAX_NICE); 650 set_user_nice(current, MAX_NICE);
943 651
944 VERBOSE_SCALEOUT("main_func task start 652 VERBOSE_SCALEOUT("main_func task started");
945 result_avg = kzalloc(nruns * sizeof(*r 653 result_avg = kzalloc(nruns * sizeof(*result_avg), GFP_KERNEL);
946 buf = kzalloc(800 + 64, GFP_KERNEL); 654 buf = kzalloc(800 + 64, GFP_KERNEL);
947 if (!result_avg || !buf) { 655 if (!result_avg || !buf) {
948 SCALEOUT_ERRSTRING("out of mem 656 SCALEOUT_ERRSTRING("out of memory");
949 goto oom_exit; 657 goto oom_exit;
950 } 658 }
951 if (holdoff) 659 if (holdoff)
952 schedule_timeout_interruptible 660 schedule_timeout_interruptible(holdoff * HZ);
953 661
954 // Wait for all threads to start. 662 // Wait for all threads to start.
955 atomic_inc(&n_init); 663 atomic_inc(&n_init);
956 while (atomic_read(&n_init) < nreaders 664 while (atomic_read(&n_init) < nreaders + 1)
957 schedule_timeout_uninterruptib 665 schedule_timeout_uninterruptible(1);
958 666
959 // Start exp readers up per experiment 667 // Start exp readers up per experiment
960 for (exp = 0; exp < nruns && !torture_ 668 for (exp = 0; exp < nruns && !torture_must_stop(); exp++) {
961 if (torture_must_stop()) 669 if (torture_must_stop())
962 goto end; 670 goto end;
963 671
964 reset_readers(); 672 reset_readers();
965 atomic_set(&nreaders_exp, nrea 673 atomic_set(&nreaders_exp, nreaders);
966 atomic_set(&n_started, nreader 674 atomic_set(&n_started, nreaders);
967 atomic_set(&n_warmedup, nreade 675 atomic_set(&n_warmedup, nreaders);
968 atomic_set(&n_cooleddown, nrea 676 atomic_set(&n_cooleddown, nreaders);
969 677
970 exp_idx = exp; 678 exp_idx = exp;
971 679
972 for (r = 0; r < nreaders; r++) 680 for (r = 0; r < nreaders; r++) {
973 smp_store_release(&rea 681 smp_store_release(&reader_tasks[r].start_reader, 1);
974 wake_up(&reader_tasks[ 682 wake_up(&reader_tasks[r].wq);
975 } 683 }
976 684
977 VERBOSE_SCALEOUT("main_func: e 685 VERBOSE_SCALEOUT("main_func: experiment started, waiting for %d readers",
978 nreaders); 686 nreaders);
979 687
980 wait_event(main_wq, 688 wait_event(main_wq,
981 !atomic_read(&nread 689 !atomic_read(&nreaders_exp) || torture_must_stop());
982 690
983 VERBOSE_SCALEOUT("main_func: e 691 VERBOSE_SCALEOUT("main_func: experiment ended");
984 692
985 if (torture_must_stop()) 693 if (torture_must_stop())
986 goto end; 694 goto end;
987 695
988 result_avg[exp] = div_u64(1000 696 result_avg[exp] = div_u64(1000 * process_durations(nreaders), nreaders * loops);
989 } 697 }
990 698
991 // Print the average of all experiment 699 // Print the average of all experiments
992 SCALEOUT("END OF TEST. Calculating ave 700 SCALEOUT("END OF TEST. Calculating average duration per loop (nanoseconds)...\n");
993 701
994 pr_alert("Runs\tTime(ns)\n"); 702 pr_alert("Runs\tTime(ns)\n");
995 for (exp = 0; exp < nruns; exp++) { 703 for (exp = 0; exp < nruns; exp++) {
996 u64 avg; 704 u64 avg;
997 u32 rem; 705 u32 rem;
998 706
999 avg = div_u64_rem(result_avg[e 707 avg = div_u64_rem(result_avg[exp], 1000, &rem);
1000 sprintf(buf1, "%d\t%llu.%03u\ 708 sprintf(buf1, "%d\t%llu.%03u\n", exp + 1, avg, rem);
1001 strcat(buf, buf1); 709 strcat(buf, buf1);
1002 if (strlen(buf) >= 800) { 710 if (strlen(buf) >= 800) {
1003 pr_alert("%s", buf); 711 pr_alert("%s", buf);
1004 buf[0] = 0; 712 buf[0] = 0;
1005 } 713 }
1006 } 714 }
1007 715
1008 pr_alert("%s", buf); 716 pr_alert("%s", buf);
1009 717
1010 oom_exit: 718 oom_exit:
1011 // This will shutdown everything incl 719 // This will shutdown everything including us.
1012 if (shutdown) { 720 if (shutdown) {
1013 shutdown_start = 1; 721 shutdown_start = 1;
1014 wake_up(&shutdown_wq); 722 wake_up(&shutdown_wq);
1015 } 723 }
1016 724
1017 // Wait for torture to stop us 725 // Wait for torture to stop us
1018 while (!torture_must_stop()) 726 while (!torture_must_stop())
1019 schedule_timeout_uninterrupti 727 schedule_timeout_uninterruptible(1);
1020 728
1021 end: 729 end:
1022 torture_kthread_stopping("main_func") 730 torture_kthread_stopping("main_func");
1023 kfree(result_avg); 731 kfree(result_avg);
1024 kfree(buf); 732 kfree(buf);
1025 return 0; 733 return 0;
1026 } 734 }
1027 735
1028 static void 736 static void
1029 ref_scale_print_module_parms(const struct ref !! 737 ref_scale_print_module_parms(struct ref_scale_ops *cur_ops, const char *tag)
1030 { 738 {
1031 pr_alert("%s" SCALE_FLAG 739 pr_alert("%s" SCALE_FLAG
1032 "--- %s: verbose=%d verbose !! 740 "--- %s: verbose=%d shutdown=%d holdoff=%d loops=%ld nreaders=%d nruns=%d readdelay=%d\n", scale_type, tag,
1033 verbose, verbose_batched, sh !! 741 verbose, shutdown, holdoff, loops, nreaders, nruns, readdelay);
1034 } 742 }
1035 743
1036 static void 744 static void
1037 ref_scale_cleanup(void) 745 ref_scale_cleanup(void)
1038 { 746 {
1039 int i; 747 int i;
1040 748
1041 if (torture_cleanup_begin()) 749 if (torture_cleanup_begin())
1042 return; 750 return;
1043 751
1044 if (!cur_ops) { 752 if (!cur_ops) {
1045 torture_cleanup_end(); 753 torture_cleanup_end();
1046 return; 754 return;
1047 } 755 }
1048 756
1049 if (reader_tasks) { 757 if (reader_tasks) {
1050 for (i = 0; i < nreaders; i++ 758 for (i = 0; i < nreaders; i++)
1051 torture_stop_kthread( 759 torture_stop_kthread("ref_scale_reader",
1052 760 reader_tasks[i].task);
1053 } 761 }
1054 kfree(reader_tasks); 762 kfree(reader_tasks);
1055 763
1056 torture_stop_kthread("main_task", mai 764 torture_stop_kthread("main_task", main_task);
1057 kfree(main_task); 765 kfree(main_task);
1058 766
1059 // Do scale-type-specific cleanup ope 767 // Do scale-type-specific cleanup operations.
1060 if (cur_ops->cleanup != NULL) 768 if (cur_ops->cleanup != NULL)
1061 cur_ops->cleanup(); 769 cur_ops->cleanup();
1062 770
1063 torture_cleanup_end(); 771 torture_cleanup_end();
1064 } 772 }
1065 773
1066 // Shutdown kthread. Just waits to be awaken 774 // Shutdown kthread. Just waits to be awakened, then shuts down system.
1067 static int 775 static int
1068 ref_scale_shutdown(void *arg) 776 ref_scale_shutdown(void *arg)
1069 { 777 {
1070 wait_event_idle(shutdown_wq, shutdown !! 778 wait_event(shutdown_wq, shutdown_start);
1071 779
1072 smp_mb(); // Wake before output. 780 smp_mb(); // Wake before output.
1073 ref_scale_cleanup(); 781 ref_scale_cleanup();
1074 kernel_power_off(); 782 kernel_power_off();
1075 783
1076 return -EINVAL; 784 return -EINVAL;
1077 } 785 }
1078 786
1079 static int __init 787 static int __init
1080 ref_scale_init(void) 788 ref_scale_init(void)
1081 { 789 {
1082 long i; 790 long i;
1083 int firsterr = 0; 791 int firsterr = 0;
1084 static const struct ref_scale_ops *sc !! 792 static struct ref_scale_ops *scale_ops[] = {
1085 &rcu_ops, &srcu_ops, RCU_TRAC !! 793 &rcu_ops, &srcu_ops, &rcu_trace_ops, &rcu_tasks_ops, &refcnt_ops, &rwlock_ops,
1086 &rwsem_ops, &lock_ops, &lock_ !! 794 &rwsem_ops, &lock_ops, &lock_irq_ops, &acqrel_ops, &clock_ops,
1087 &typesafe_ref_ops, &typesafe_ <<
1088 }; 795 };
1089 796
1090 if (!torture_init_begin(scale_type, v 797 if (!torture_init_begin(scale_type, verbose))
1091 return -EBUSY; 798 return -EBUSY;
1092 799
1093 for (i = 0; i < ARRAY_SIZE(scale_ops) 800 for (i = 0; i < ARRAY_SIZE(scale_ops); i++) {
1094 cur_ops = scale_ops[i]; 801 cur_ops = scale_ops[i];
1095 if (strcmp(scale_type, cur_op 802 if (strcmp(scale_type, cur_ops->name) == 0)
1096 break; 803 break;
1097 } 804 }
1098 if (i == ARRAY_SIZE(scale_ops)) { 805 if (i == ARRAY_SIZE(scale_ops)) {
1099 pr_alert("rcu-scale: invalid 806 pr_alert("rcu-scale: invalid scale type: \"%s\"\n", scale_type);
1100 pr_alert("rcu-scale types:"); 807 pr_alert("rcu-scale types:");
1101 for (i = 0; i < ARRAY_SIZE(sc 808 for (i = 0; i < ARRAY_SIZE(scale_ops); i++)
1102 pr_cont(" %s", scale_ 809 pr_cont(" %s", scale_ops[i]->name);
1103 pr_cont("\n"); 810 pr_cont("\n");
1104 firsterr = -EINVAL; 811 firsterr = -EINVAL;
1105 cur_ops = NULL; 812 cur_ops = NULL;
1106 goto unwind; 813 goto unwind;
1107 } 814 }
1108 if (cur_ops->init) 815 if (cur_ops->init)
1109 if (!cur_ops->init()) { !! 816 cur_ops->init();
1110 firsterr = -EUCLEAN; <<
1111 goto unwind; <<
1112 } <<
1113 817
1114 ref_scale_print_module_parms(cur_ops, 818 ref_scale_print_module_parms(cur_ops, "Start of test");
1115 819
1116 // Shutdown task 820 // Shutdown task
1117 if (shutdown) { 821 if (shutdown) {
1118 init_waitqueue_head(&shutdown 822 init_waitqueue_head(&shutdown_wq);
1119 firsterr = torture_create_kth 823 firsterr = torture_create_kthread(ref_scale_shutdown, NULL,
1120 824 shutdown_task);
1121 if (torture_init_error(firste 825 if (torture_init_error(firsterr))
1122 goto unwind; 826 goto unwind;
1123 schedule_timeout_uninterrupti 827 schedule_timeout_uninterruptible(1);
1124 } 828 }
1125 829
1126 // Reader tasks (default to ~75% of o 830 // Reader tasks (default to ~75% of online CPUs).
1127 if (nreaders < 0) 831 if (nreaders < 0)
1128 nreaders = (num_online_cpus() 832 nreaders = (num_online_cpus() >> 1) + (num_online_cpus() >> 2);
1129 if (WARN_ONCE(loops <= 0, "%s: loops 833 if (WARN_ONCE(loops <= 0, "%s: loops = %ld, adjusted to 1\n", __func__, loops))
1130 loops = 1; 834 loops = 1;
1131 if (WARN_ONCE(nreaders <= 0, "%s: nre 835 if (WARN_ONCE(nreaders <= 0, "%s: nreaders = %d, adjusted to 1\n", __func__, nreaders))
1132 nreaders = 1; 836 nreaders = 1;
1133 if (WARN_ONCE(nruns <= 0, "%s: nruns 837 if (WARN_ONCE(nruns <= 0, "%s: nruns = %d, adjusted to 1\n", __func__, nruns))
1134 nruns = 1; 838 nruns = 1;
1135 reader_tasks = kcalloc(nreaders, size 839 reader_tasks = kcalloc(nreaders, sizeof(reader_tasks[0]),
1136 GFP_KERNEL); 840 GFP_KERNEL);
1137 if (!reader_tasks) { 841 if (!reader_tasks) {
1138 SCALEOUT_ERRSTRING("out of me 842 SCALEOUT_ERRSTRING("out of memory");
1139 firsterr = -ENOMEM; 843 firsterr = -ENOMEM;
1140 goto unwind; 844 goto unwind;
1141 } 845 }
1142 846
1143 VERBOSE_SCALEOUT("Starting %d reader 847 VERBOSE_SCALEOUT("Starting %d reader threads", nreaders);
1144 848
1145 for (i = 0; i < nreaders; i++) { 849 for (i = 0; i < nreaders; i++) {
1146 init_waitqueue_head(&reader_t <<
1147 firsterr = torture_create_kth 850 firsterr = torture_create_kthread(ref_scale_reader, (void *)i,
1148 851 reader_tasks[i].task);
1149 if (torture_init_error(firste 852 if (torture_init_error(firsterr))
1150 goto unwind; 853 goto unwind;
>> 854
>> 855 init_waitqueue_head(&(reader_tasks[i].wq));
1151 } 856 }
1152 857
1153 // Main Task 858 // Main Task
1154 init_waitqueue_head(&main_wq); 859 init_waitqueue_head(&main_wq);
1155 firsterr = torture_create_kthread(mai 860 firsterr = torture_create_kthread(main_func, NULL, main_task);
1156 if (torture_init_error(firsterr)) 861 if (torture_init_error(firsterr))
1157 goto unwind; 862 goto unwind;
1158 863
1159 torture_init_end(); 864 torture_init_end();
1160 return 0; 865 return 0;
1161 866
1162 unwind: 867 unwind:
1163 torture_init_end(); 868 torture_init_end();
1164 ref_scale_cleanup(); 869 ref_scale_cleanup();
1165 if (shutdown) { 870 if (shutdown) {
1166 WARN_ON(!IS_MODULE(CONFIG_RCU 871 WARN_ON(!IS_MODULE(CONFIG_RCU_REF_SCALE_TEST));
1167 kernel_power_off(); 872 kernel_power_off();
1168 } 873 }
1169 return firsterr; 874 return firsterr;
1170 } 875 }
1171 876
1172 module_init(ref_scale_init); 877 module_init(ref_scale_init);
1173 module_exit(ref_scale_cleanup); 878 module_exit(ref_scale_cleanup);
1174 879
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.