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 210 #ifdef CONFIG_TASKS_RCU
216 211
217 // Definitions for RCU Tasks ref scale testing 212 // Definitions for RCU Tasks ref scale testing: Empty read markers.
218 // These definitions also work for RCU Rude re 213 // These definitions also work for RCU Rude readers.
219 static void rcu_tasks_ref_scale_read_section(c 214 static void rcu_tasks_ref_scale_read_section(const int nloops)
220 { 215 {
221 int i; 216 int i;
222 217
223 for (i = nloops; i >= 0; i--) 218 for (i = nloops; i >= 0; i--)
224 continue; 219 continue;
225 } 220 }
226 221
227 static void rcu_tasks_ref_scale_delay_section( 222 static void rcu_tasks_ref_scale_delay_section(const int nloops, const int udl, const int ndl)
228 { 223 {
229 int i; 224 int i;
230 225
231 for (i = nloops; i >= 0; i--) 226 for (i = nloops; i >= 0; i--)
232 un_delay(udl, ndl); 227 un_delay(udl, ndl);
233 } 228 }
234 229
235 static const struct ref_scale_ops rcu_tasks_op !! 230 static struct ref_scale_ops rcu_tasks_ops = {
236 .init = rcu_sync_scale_init, 231 .init = rcu_sync_scale_init,
237 .readsection = rcu_tasks_ref_scale_ 232 .readsection = rcu_tasks_ref_scale_read_section,
238 .delaysection = rcu_tasks_ref_scale_ 233 .delaysection = rcu_tasks_ref_scale_delay_section,
239 .name = "rcu-tasks" 234 .name = "rcu-tasks"
240 }; 235 };
241 236
242 #define RCU_TASKS_OPS &rcu_tasks_ops, 237 #define RCU_TASKS_OPS &rcu_tasks_ops,
243 238
244 #else // #ifdef CONFIG_TASKS_RCU 239 #else // #ifdef CONFIG_TASKS_RCU
245 240
246 #define RCU_TASKS_OPS 241 #define RCU_TASKS_OPS
247 242
248 #endif // #else // #ifdef CONFIG_TASKS_RCU 243 #endif // #else // #ifdef CONFIG_TASKS_RCU
249 244
250 #ifdef CONFIG_TASKS_TRACE_RCU 245 #ifdef CONFIG_TASKS_TRACE_RCU
251 246
252 // Definitions for RCU Tasks Trace ref scale t 247 // Definitions for RCU Tasks Trace ref scale testing.
253 static void rcu_trace_ref_scale_read_section(c 248 static void rcu_trace_ref_scale_read_section(const int nloops)
254 { 249 {
255 int i; 250 int i;
256 251
257 for (i = nloops; i >= 0; i--) { 252 for (i = nloops; i >= 0; i--) {
258 rcu_read_lock_trace(); 253 rcu_read_lock_trace();
259 rcu_read_unlock_trace(); 254 rcu_read_unlock_trace();
260 } 255 }
261 } 256 }
262 257
263 static void rcu_trace_ref_scale_delay_section( 258 static void rcu_trace_ref_scale_delay_section(const int nloops, const int udl, const int ndl)
264 { 259 {
265 int i; 260 int i;
266 261
267 for (i = nloops; i >= 0; i--) { 262 for (i = nloops; i >= 0; i--) {
268 rcu_read_lock_trace(); 263 rcu_read_lock_trace();
269 un_delay(udl, ndl); 264 un_delay(udl, ndl);
270 rcu_read_unlock_trace(); 265 rcu_read_unlock_trace();
271 } 266 }
272 } 267 }
273 268
274 static const struct ref_scale_ops rcu_trace_op !! 269 static struct ref_scale_ops rcu_trace_ops = {
275 .init = rcu_sync_scale_init, 270 .init = rcu_sync_scale_init,
276 .readsection = rcu_trace_ref_scale_ 271 .readsection = rcu_trace_ref_scale_read_section,
277 .delaysection = rcu_trace_ref_scale_ 272 .delaysection = rcu_trace_ref_scale_delay_section,
278 .name = "rcu-trace" 273 .name = "rcu-trace"
279 }; 274 };
280 275
281 #define RCU_TRACE_OPS &rcu_trace_ops, 276 #define RCU_TRACE_OPS &rcu_trace_ops,
282 277
283 #else // #ifdef CONFIG_TASKS_TRACE_RCU 278 #else // #ifdef CONFIG_TASKS_TRACE_RCU
284 279
285 #define RCU_TRACE_OPS 280 #define RCU_TRACE_OPS
286 281
287 #endif // #else // #ifdef CONFIG_TASKS_TRACE_R 282 #endif // #else // #ifdef CONFIG_TASKS_TRACE_RCU
288 283
289 // Definitions for reference count 284 // Definitions for reference count
290 static atomic_t refcnt; 285 static atomic_t refcnt;
291 286
292 static void ref_refcnt_section(const int nloop 287 static void ref_refcnt_section(const int nloops)
293 { 288 {
294 int i; 289 int i;
295 290
296 for (i = nloops; i >= 0; i--) { 291 for (i = nloops; i >= 0; i--) {
297 atomic_inc(&refcnt); 292 atomic_inc(&refcnt);
298 atomic_dec(&refcnt); 293 atomic_dec(&refcnt);
299 } 294 }
300 } 295 }
301 296
302 static void ref_refcnt_delay_section(const int 297 static void ref_refcnt_delay_section(const int nloops, const int udl, const int ndl)
303 { 298 {
304 int i; 299 int i;
305 300
306 for (i = nloops; i >= 0; i--) { 301 for (i = nloops; i >= 0; i--) {
307 atomic_inc(&refcnt); 302 atomic_inc(&refcnt);
308 un_delay(udl, ndl); 303 un_delay(udl, ndl);
309 atomic_dec(&refcnt); 304 atomic_dec(&refcnt);
310 } 305 }
311 } 306 }
312 307
313 static const struct ref_scale_ops refcnt_ops = !! 308 static struct ref_scale_ops refcnt_ops = {
314 .init = rcu_sync_scale_init, 309 .init = rcu_sync_scale_init,
315 .readsection = ref_refcnt_section, 310 .readsection = ref_refcnt_section,
316 .delaysection = ref_refcnt_delay_sec 311 .delaysection = ref_refcnt_delay_section,
317 .name = "refcnt" 312 .name = "refcnt"
318 }; 313 };
319 314
320 // Definitions for rwlock 315 // Definitions for rwlock
321 static rwlock_t test_rwlock; 316 static rwlock_t test_rwlock;
322 317
323 static bool ref_rwlock_init(void) !! 318 static void ref_rwlock_init(void)
324 { 319 {
325 rwlock_init(&test_rwlock); 320 rwlock_init(&test_rwlock);
326 return true; <<
327 } 321 }
328 322
329 static void ref_rwlock_section(const int nloop 323 static void ref_rwlock_section(const int nloops)
330 { 324 {
331 int i; 325 int i;
332 326
333 for (i = nloops; i >= 0; i--) { 327 for (i = nloops; i >= 0; i--) {
334 read_lock(&test_rwlock); 328 read_lock(&test_rwlock);
335 read_unlock(&test_rwlock); 329 read_unlock(&test_rwlock);
336 } 330 }
337 } 331 }
338 332
339 static void ref_rwlock_delay_section(const int 333 static void ref_rwlock_delay_section(const int nloops, const int udl, const int ndl)
340 { 334 {
341 int i; 335 int i;
342 336
343 for (i = nloops; i >= 0; i--) { 337 for (i = nloops; i >= 0; i--) {
344 read_lock(&test_rwlock); 338 read_lock(&test_rwlock);
345 un_delay(udl, ndl); 339 un_delay(udl, ndl);
346 read_unlock(&test_rwlock); 340 read_unlock(&test_rwlock);
347 } 341 }
348 } 342 }
349 343
350 static const struct ref_scale_ops rwlock_ops = !! 344 static struct ref_scale_ops rwlock_ops = {
351 .init = ref_rwlock_init, 345 .init = ref_rwlock_init,
352 .readsection = ref_rwlock_section, 346 .readsection = ref_rwlock_section,
353 .delaysection = ref_rwlock_delay_sec 347 .delaysection = ref_rwlock_delay_section,
354 .name = "rwlock" 348 .name = "rwlock"
355 }; 349 };
356 350
357 // Definitions for rwsem 351 // Definitions for rwsem
358 static struct rw_semaphore test_rwsem; 352 static struct rw_semaphore test_rwsem;
359 353
360 static bool ref_rwsem_init(void) !! 354 static void ref_rwsem_init(void)
361 { 355 {
362 init_rwsem(&test_rwsem); 356 init_rwsem(&test_rwsem);
363 return true; <<
364 } 357 }
365 358
366 static void ref_rwsem_section(const int nloops 359 static void ref_rwsem_section(const int nloops)
367 { 360 {
368 int i; 361 int i;
369 362
370 for (i = nloops; i >= 0; i--) { 363 for (i = nloops; i >= 0; i--) {
371 down_read(&test_rwsem); 364 down_read(&test_rwsem);
372 up_read(&test_rwsem); 365 up_read(&test_rwsem);
373 } 366 }
374 } 367 }
375 368
376 static void ref_rwsem_delay_section(const int 369 static void ref_rwsem_delay_section(const int nloops, const int udl, const int ndl)
377 { 370 {
378 int i; 371 int i;
379 372
380 for (i = nloops; i >= 0; i--) { 373 for (i = nloops; i >= 0; i--) {
381 down_read(&test_rwsem); 374 down_read(&test_rwsem);
382 un_delay(udl, ndl); 375 un_delay(udl, ndl);
383 up_read(&test_rwsem); 376 up_read(&test_rwsem);
384 } 377 }
385 } 378 }
386 379
387 static const struct ref_scale_ops rwsem_ops = !! 380 static struct ref_scale_ops rwsem_ops = {
388 .init = ref_rwsem_init, 381 .init = ref_rwsem_init,
389 .readsection = ref_rwsem_section, 382 .readsection = ref_rwsem_section,
390 .delaysection = ref_rwsem_delay_sect 383 .delaysection = ref_rwsem_delay_section,
391 .name = "rwsem" 384 .name = "rwsem"
392 }; 385 };
393 386
394 // Definitions for global spinlock 387 // Definitions for global spinlock
395 static DEFINE_RAW_SPINLOCK(test_lock); 388 static DEFINE_RAW_SPINLOCK(test_lock);
396 389
397 static void ref_lock_section(const int nloops) 390 static void ref_lock_section(const int nloops)
398 { 391 {
399 int i; 392 int i;
400 393
401 preempt_disable(); 394 preempt_disable();
402 for (i = nloops; i >= 0; i--) { 395 for (i = nloops; i >= 0; i--) {
403 raw_spin_lock(&test_lock); 396 raw_spin_lock(&test_lock);
404 raw_spin_unlock(&test_lock); 397 raw_spin_unlock(&test_lock);
405 } 398 }
406 preempt_enable(); 399 preempt_enable();
407 } 400 }
408 401
409 static void ref_lock_delay_section(const int n 402 static void ref_lock_delay_section(const int nloops, const int udl, const int ndl)
410 { 403 {
411 int i; 404 int i;
412 405
413 preempt_disable(); 406 preempt_disable();
414 for (i = nloops; i >= 0; i--) { 407 for (i = nloops; i >= 0; i--) {
415 raw_spin_lock(&test_lock); 408 raw_spin_lock(&test_lock);
416 un_delay(udl, ndl); 409 un_delay(udl, ndl);
417 raw_spin_unlock(&test_lock); 410 raw_spin_unlock(&test_lock);
418 } 411 }
419 preempt_enable(); 412 preempt_enable();
420 } 413 }
421 414
422 static const struct ref_scale_ops lock_ops = { !! 415 static struct ref_scale_ops lock_ops = {
423 .readsection = ref_lock_section, 416 .readsection = ref_lock_section,
424 .delaysection = ref_lock_delay_secti 417 .delaysection = ref_lock_delay_section,
425 .name = "lock" 418 .name = "lock"
426 }; 419 };
427 420
428 // Definitions for global irq-save spinlock 421 // Definitions for global irq-save spinlock
429 422
430 static void ref_lock_irq_section(const int nlo 423 static void ref_lock_irq_section(const int nloops)
431 { 424 {
432 unsigned long flags; 425 unsigned long flags;
433 int i; 426 int i;
434 427
435 preempt_disable(); 428 preempt_disable();
436 for (i = nloops; i >= 0; i--) { 429 for (i = nloops; i >= 0; i--) {
437 raw_spin_lock_irqsave(&test_lo 430 raw_spin_lock_irqsave(&test_lock, flags);
438 raw_spin_unlock_irqrestore(&te 431 raw_spin_unlock_irqrestore(&test_lock, flags);
439 } 432 }
440 preempt_enable(); 433 preempt_enable();
441 } 434 }
442 435
443 static void ref_lock_irq_delay_section(const i 436 static void ref_lock_irq_delay_section(const int nloops, const int udl, const int ndl)
444 { 437 {
445 unsigned long flags; 438 unsigned long flags;
446 int i; 439 int i;
447 440
448 preempt_disable(); 441 preempt_disable();
449 for (i = nloops; i >= 0; i--) { 442 for (i = nloops; i >= 0; i--) {
450 raw_spin_lock_irqsave(&test_lo 443 raw_spin_lock_irqsave(&test_lock, flags);
451 un_delay(udl, ndl); 444 un_delay(udl, ndl);
452 raw_spin_unlock_irqrestore(&te 445 raw_spin_unlock_irqrestore(&test_lock, flags);
453 } 446 }
454 preempt_enable(); 447 preempt_enable();
455 } 448 }
456 449
457 static const struct ref_scale_ops lock_irq_ops !! 450 static struct ref_scale_ops lock_irq_ops = {
458 .readsection = ref_lock_irq_section 451 .readsection = ref_lock_irq_section,
459 .delaysection = ref_lock_irq_delay_s 452 .delaysection = ref_lock_irq_delay_section,
460 .name = "lock-irq" 453 .name = "lock-irq"
461 }; 454 };
462 455
463 // Definitions acquire-release. 456 // Definitions acquire-release.
464 static DEFINE_PER_CPU(unsigned long, test_acqr 457 static DEFINE_PER_CPU(unsigned long, test_acqrel);
465 458
466 static void ref_acqrel_section(const int nloop 459 static void ref_acqrel_section(const int nloops)
467 { 460 {
468 unsigned long x; 461 unsigned long x;
469 int i; 462 int i;
470 463
471 preempt_disable(); 464 preempt_disable();
472 for (i = nloops; i >= 0; i--) { 465 for (i = nloops; i >= 0; i--) {
473 x = smp_load_acquire(this_cpu_ 466 x = smp_load_acquire(this_cpu_ptr(&test_acqrel));
474 smp_store_release(this_cpu_ptr 467 smp_store_release(this_cpu_ptr(&test_acqrel), x + 1);
475 } 468 }
476 preempt_enable(); 469 preempt_enable();
477 } 470 }
478 471
479 static void ref_acqrel_delay_section(const int 472 static void ref_acqrel_delay_section(const int nloops, const int udl, const int ndl)
480 { 473 {
481 unsigned long x; 474 unsigned long x;
482 int i; 475 int i;
483 476
484 preempt_disable(); 477 preempt_disable();
485 for (i = nloops; i >= 0; i--) { 478 for (i = nloops; i >= 0; i--) {
486 x = smp_load_acquire(this_cpu_ 479 x = smp_load_acquire(this_cpu_ptr(&test_acqrel));
487 un_delay(udl, ndl); 480 un_delay(udl, ndl);
488 smp_store_release(this_cpu_ptr 481 smp_store_release(this_cpu_ptr(&test_acqrel), x + 1);
489 } 482 }
490 preempt_enable(); 483 preempt_enable();
491 } 484 }
492 485
493 static const struct ref_scale_ops acqrel_ops = !! 486 static struct ref_scale_ops acqrel_ops = {
494 .readsection = ref_acqrel_section, 487 .readsection = ref_acqrel_section,
495 .delaysection = ref_acqrel_delay_sec 488 .delaysection = ref_acqrel_delay_section,
496 .name = "acqrel" 489 .name = "acqrel"
497 }; 490 };
498 491
499 static volatile u64 stopopts; 492 static volatile u64 stopopts;
500 493
501 static void ref_clock_section(const int nloops 494 static void ref_clock_section(const int nloops)
502 { 495 {
503 u64 x = 0; 496 u64 x = 0;
504 int i; 497 int i;
505 498
506 preempt_disable(); 499 preempt_disable();
507 for (i = nloops; i >= 0; i--) 500 for (i = nloops; i >= 0; i--)
508 x += ktime_get_real_fast_ns(); 501 x += ktime_get_real_fast_ns();
509 preempt_enable(); 502 preempt_enable();
510 stopopts = x; 503 stopopts = x;
511 } 504 }
512 505
513 static void ref_clock_delay_section(const int 506 static void ref_clock_delay_section(const int nloops, const int udl, const int ndl)
514 { 507 {
515 u64 x = 0; 508 u64 x = 0;
516 int i; 509 int i;
517 510
518 preempt_disable(); 511 preempt_disable();
519 for (i = nloops; i >= 0; i--) { 512 for (i = nloops; i >= 0; i--) {
520 x += ktime_get_real_fast_ns(); 513 x += ktime_get_real_fast_ns();
521 un_delay(udl, ndl); 514 un_delay(udl, ndl);
522 } 515 }
523 preempt_enable(); 516 preempt_enable();
524 stopopts = x; 517 stopopts = x;
525 } 518 }
526 519
527 static const struct ref_scale_ops clock_ops = !! 520 static struct ref_scale_ops clock_ops = {
528 .readsection = ref_clock_section, 521 .readsection = ref_clock_section,
529 .delaysection = ref_clock_delay_sect 522 .delaysection = ref_clock_delay_section,
530 .name = "clock" 523 .name = "clock"
531 }; 524 };
532 525
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) 526 static void rcu_scale_one_reader(void)
797 { 527 {
798 if (readdelay <= 0) 528 if (readdelay <= 0)
799 cur_ops->readsection(loops); 529 cur_ops->readsection(loops);
800 else 530 else
801 cur_ops->delaysection(loops, r 531 cur_ops->delaysection(loops, readdelay / 1000, readdelay % 1000);
802 } 532 }
803 533
804 // Reader kthread. Repeatedly does empty RCU 534 // Reader kthread. Repeatedly does empty RCU read-side
805 // critical section, minimizing update-side in 535 // critical section, minimizing update-side interference.
806 static int 536 static int
807 ref_scale_reader(void *arg) 537 ref_scale_reader(void *arg)
808 { 538 {
809 unsigned long flags; 539 unsigned long flags;
810 long me = (long)arg; 540 long me = (long)arg;
811 struct reader_task *rt = &(reader_task 541 struct reader_task *rt = &(reader_tasks[me]);
812 u64 start; 542 u64 start;
813 s64 duration; 543 s64 duration;
814 544
815 VERBOSE_SCALEOUT_BATCH("ref_scale_read 545 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: task started", me);
816 WARN_ON_ONCE(set_cpus_allowed_ptr(curr 546 WARN_ON_ONCE(set_cpus_allowed_ptr(current, cpumask_of(me % nr_cpu_ids)));
817 set_user_nice(current, MAX_NICE); 547 set_user_nice(current, MAX_NICE);
818 atomic_inc(&n_init); 548 atomic_inc(&n_init);
819 if (holdoff) 549 if (holdoff)
820 schedule_timeout_interruptible 550 schedule_timeout_interruptible(holdoff * HZ);
821 repeat: 551 repeat:
822 VERBOSE_SCALEOUT_BATCH("ref_scale_read 552 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: waiting to start next experiment on cpu %d", me, raw_smp_processor_id());
823 553
824 // Wait for signal that this reader ca 554 // Wait for signal that this reader can start.
825 wait_event(rt->wq, (atomic_read(&nread 555 wait_event(rt->wq, (atomic_read(&nreaders_exp) && smp_load_acquire(&rt->start_reader)) ||
826 torture_must_stop() 556 torture_must_stop());
827 557
828 if (torture_must_stop()) 558 if (torture_must_stop())
829 goto end; 559 goto end;
830 560
831 // Make sure that the CPU is affinitiz 561 // Make sure that the CPU is affinitized appropriately during testing.
832 WARN_ON_ONCE(raw_smp_processor_id() != 562 WARN_ON_ONCE(raw_smp_processor_id() != me);
833 563
834 WRITE_ONCE(rt->start_reader, 0); 564 WRITE_ONCE(rt->start_reader, 0);
835 if (!atomic_dec_return(&n_started)) 565 if (!atomic_dec_return(&n_started))
836 while (atomic_read_acquire(&n_ 566 while (atomic_read_acquire(&n_started))
837 cpu_relax(); 567 cpu_relax();
838 568
839 VERBOSE_SCALEOUT_BATCH("ref_scale_read 569 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: experiment %d started", me, exp_idx);
840 570
841 571
842 // To reduce noise, do an initial cach 572 // To reduce noise, do an initial cache-warming invocation, check
843 // in, and then keep warming until eve 573 // in, and then keep warming until everyone has checked in.
844 rcu_scale_one_reader(); 574 rcu_scale_one_reader();
845 if (!atomic_dec_return(&n_warmedup)) 575 if (!atomic_dec_return(&n_warmedup))
846 while (atomic_read_acquire(&n_ 576 while (atomic_read_acquire(&n_warmedup))
847 rcu_scale_one_reader() 577 rcu_scale_one_reader();
848 // Also keep interrupts disabled. Thi 578 // Also keep interrupts disabled. This also has the effect
849 // of preventing entries into slow pat 579 // of preventing entries into slow path for rcu_read_unlock().
850 local_irq_save(flags); 580 local_irq_save(flags);
851 start = ktime_get_mono_fast_ns(); 581 start = ktime_get_mono_fast_ns();
852 582
853 rcu_scale_one_reader(); 583 rcu_scale_one_reader();
854 584
855 duration = ktime_get_mono_fast_ns() - 585 duration = ktime_get_mono_fast_ns() - start;
856 local_irq_restore(flags); 586 local_irq_restore(flags);
857 587
858 rt->last_duration_ns = WARN_ON_ONCE(du 588 rt->last_duration_ns = WARN_ON_ONCE(duration < 0) ? 0 : duration;
859 // To reduce runtime-skew noise, do ma 589 // To reduce runtime-skew noise, do maintain-load invocations until
860 // everyone is done. 590 // everyone is done.
861 if (!atomic_dec_return(&n_cooleddown)) 591 if (!atomic_dec_return(&n_cooleddown))
862 while (atomic_read_acquire(&n_ 592 while (atomic_read_acquire(&n_cooleddown))
863 rcu_scale_one_reader() 593 rcu_scale_one_reader();
864 594
865 if (atomic_dec_and_test(&nreaders_exp) 595 if (atomic_dec_and_test(&nreaders_exp))
866 wake_up(&main_wq); 596 wake_up(&main_wq);
867 597
868 VERBOSE_SCALEOUT_BATCH("ref_scale_read 598 VERBOSE_SCALEOUT_BATCH("ref_scale_reader %ld: experiment %d ended, (readers remaining=%d)",
869 me, exp_idx, a 599 me, exp_idx, atomic_read(&nreaders_exp));
870 600
871 if (!torture_must_stop()) 601 if (!torture_must_stop())
872 goto repeat; 602 goto repeat;
873 end: 603 end:
874 torture_kthread_stopping("ref_scale_re 604 torture_kthread_stopping("ref_scale_reader");
875 return 0; 605 return 0;
876 } 606 }
877 607
878 static void reset_readers(void) 608 static void reset_readers(void)
879 { 609 {
880 int i; 610 int i;
881 struct reader_task *rt; 611 struct reader_task *rt;
882 612
883 for (i = 0; i < nreaders; i++) { 613 for (i = 0; i < nreaders; i++) {
884 rt = &(reader_tasks[i]); 614 rt = &(reader_tasks[i]);
885 615
886 rt->last_duration_ns = 0; 616 rt->last_duration_ns = 0;
887 } 617 }
888 } 618 }
889 619
890 // Print the results of each reader and return 620 // Print the results of each reader and return the sum of all their durations.
891 static u64 process_durations(int n) 621 static u64 process_durations(int n)
892 { 622 {
893 int i; 623 int i;
894 struct reader_task *rt; 624 struct reader_task *rt;
895 struct seq_buf s; !! 625 char buf1[64];
896 char *buf; 626 char *buf;
897 u64 sum = 0; 627 u64 sum = 0;
898 628
899 buf = kmalloc(800 + 64, GFP_KERNEL); 629 buf = kmalloc(800 + 64, GFP_KERNEL);
900 if (!buf) 630 if (!buf)
901 return 0; 631 return 0;
902 seq_buf_init(&s, buf, 800 + 64); !! 632 buf[0] = 0;
903 !! 633 sprintf(buf, "Experiment #%d (Format: <THREAD-NUM>:<Total loop time in ns>)",
904 seq_buf_printf(&s, "Experiment #%d (Fo !! 634 exp_idx);
905 exp_idx); <<
906 635
907 for (i = 0; i < n && !torture_must_sto 636 for (i = 0; i < n && !torture_must_stop(); i++) {
908 rt = &(reader_tasks[i]); 637 rt = &(reader_tasks[i]);
>> 638 sprintf(buf1, "%d: %llu\t", i, rt->last_duration_ns);
909 639
910 if (i % 5 == 0) 640 if (i % 5 == 0)
911 seq_buf_putc(&s, '\n') !! 641 strcat(buf, "\n");
912 !! 642 if (strlen(buf) >= 800) {
913 if (seq_buf_used(&s) >= 800) { !! 643 pr_alert("%s", buf);
914 pr_alert("%s", seq_buf !! 644 buf[0] = 0;
915 seq_buf_clear(&s); <<
916 } 645 }
917 !! 646 strcat(buf, buf1);
918 seq_buf_printf(&s, "%d: %llu\t <<
919 647
920 sum += rt->last_duration_ns; 648 sum += rt->last_duration_ns;
921 } 649 }
922 pr_alert("%s\n", seq_buf_str(&s)); !! 650 pr_alert("%s\n", buf);
923 651
924 kfree(buf); 652 kfree(buf);
925 return sum; 653 return sum;
926 } 654 }
927 655
928 // The main_func is the main orchestrator, it 656 // The main_func is the main orchestrator, it performs a bunch of
929 // experiments. For every experiment, it orde 657 // experiments. For every experiment, it orders all the readers
930 // involved to start and waits for them to fin 658 // involved to start and waits for them to finish the experiment. It
931 // then reads their timestamps and starts the 659 // then reads their timestamps and starts the next experiment. Each
932 // experiment progresses from 1 concurrent rea 660 // experiment progresses from 1 concurrent reader to N of them at which
933 // point all the timestamps are printed. 661 // point all the timestamps are printed.
934 static int main_func(void *arg) 662 static int main_func(void *arg)
935 { 663 {
936 int exp, r; 664 int exp, r;
937 char buf1[64]; 665 char buf1[64];
938 char *buf; 666 char *buf;
939 u64 *result_avg; 667 u64 *result_avg;
940 668
941 set_cpus_allowed_ptr(current, cpumask_ 669 set_cpus_allowed_ptr(current, cpumask_of(nreaders % nr_cpu_ids));
942 set_user_nice(current, MAX_NICE); 670 set_user_nice(current, MAX_NICE);
943 671
944 VERBOSE_SCALEOUT("main_func task start 672 VERBOSE_SCALEOUT("main_func task started");
945 result_avg = kzalloc(nruns * sizeof(*r 673 result_avg = kzalloc(nruns * sizeof(*result_avg), GFP_KERNEL);
946 buf = kzalloc(800 + 64, GFP_KERNEL); 674 buf = kzalloc(800 + 64, GFP_KERNEL);
947 if (!result_avg || !buf) { 675 if (!result_avg || !buf) {
948 SCALEOUT_ERRSTRING("out of mem 676 SCALEOUT_ERRSTRING("out of memory");
949 goto oom_exit; 677 goto oom_exit;
950 } 678 }
951 if (holdoff) 679 if (holdoff)
952 schedule_timeout_interruptible 680 schedule_timeout_interruptible(holdoff * HZ);
953 681
954 // Wait for all threads to start. 682 // Wait for all threads to start.
955 atomic_inc(&n_init); 683 atomic_inc(&n_init);
956 while (atomic_read(&n_init) < nreaders 684 while (atomic_read(&n_init) < nreaders + 1)
957 schedule_timeout_uninterruptib 685 schedule_timeout_uninterruptible(1);
958 686
959 // Start exp readers up per experiment 687 // Start exp readers up per experiment
960 for (exp = 0; exp < nruns && !torture_ 688 for (exp = 0; exp < nruns && !torture_must_stop(); exp++) {
961 if (torture_must_stop()) 689 if (torture_must_stop())
962 goto end; 690 goto end;
963 691
964 reset_readers(); 692 reset_readers();
965 atomic_set(&nreaders_exp, nrea 693 atomic_set(&nreaders_exp, nreaders);
966 atomic_set(&n_started, nreader 694 atomic_set(&n_started, nreaders);
967 atomic_set(&n_warmedup, nreade 695 atomic_set(&n_warmedup, nreaders);
968 atomic_set(&n_cooleddown, nrea 696 atomic_set(&n_cooleddown, nreaders);
969 697
970 exp_idx = exp; 698 exp_idx = exp;
971 699
972 for (r = 0; r < nreaders; r++) 700 for (r = 0; r < nreaders; r++) {
973 smp_store_release(&rea 701 smp_store_release(&reader_tasks[r].start_reader, 1);
974 wake_up(&reader_tasks[ 702 wake_up(&reader_tasks[r].wq);
975 } 703 }
976 704
977 VERBOSE_SCALEOUT("main_func: e 705 VERBOSE_SCALEOUT("main_func: experiment started, waiting for %d readers",
978 nreaders); 706 nreaders);
979 707
980 wait_event(main_wq, 708 wait_event(main_wq,
981 !atomic_read(&nread 709 !atomic_read(&nreaders_exp) || torture_must_stop());
982 710
983 VERBOSE_SCALEOUT("main_func: e 711 VERBOSE_SCALEOUT("main_func: experiment ended");
984 712
985 if (torture_must_stop()) 713 if (torture_must_stop())
986 goto end; 714 goto end;
987 715
988 result_avg[exp] = div_u64(1000 716 result_avg[exp] = div_u64(1000 * process_durations(nreaders), nreaders * loops);
989 } 717 }
990 718
991 // Print the average of all experiment 719 // Print the average of all experiments
992 SCALEOUT("END OF TEST. Calculating ave 720 SCALEOUT("END OF TEST. Calculating average duration per loop (nanoseconds)...\n");
993 721
994 pr_alert("Runs\tTime(ns)\n"); 722 pr_alert("Runs\tTime(ns)\n");
995 for (exp = 0; exp < nruns; exp++) { 723 for (exp = 0; exp < nruns; exp++) {
996 u64 avg; 724 u64 avg;
997 u32 rem; 725 u32 rem;
998 726
999 avg = div_u64_rem(result_avg[e 727 avg = div_u64_rem(result_avg[exp], 1000, &rem);
1000 sprintf(buf1, "%d\t%llu.%03u\ 728 sprintf(buf1, "%d\t%llu.%03u\n", exp + 1, avg, rem);
1001 strcat(buf, buf1); 729 strcat(buf, buf1);
1002 if (strlen(buf) >= 800) { 730 if (strlen(buf) >= 800) {
1003 pr_alert("%s", buf); 731 pr_alert("%s", buf);
1004 buf[0] = 0; 732 buf[0] = 0;
1005 } 733 }
1006 } 734 }
1007 735
1008 pr_alert("%s", buf); 736 pr_alert("%s", buf);
1009 737
1010 oom_exit: 738 oom_exit:
1011 // This will shutdown everything incl 739 // This will shutdown everything including us.
1012 if (shutdown) { 740 if (shutdown) {
1013 shutdown_start = 1; 741 shutdown_start = 1;
1014 wake_up(&shutdown_wq); 742 wake_up(&shutdown_wq);
1015 } 743 }
1016 744
1017 // Wait for torture to stop us 745 // Wait for torture to stop us
1018 while (!torture_must_stop()) 746 while (!torture_must_stop())
1019 schedule_timeout_uninterrupti 747 schedule_timeout_uninterruptible(1);
1020 748
1021 end: 749 end:
1022 torture_kthread_stopping("main_func") 750 torture_kthread_stopping("main_func");
1023 kfree(result_avg); 751 kfree(result_avg);
1024 kfree(buf); 752 kfree(buf);
1025 return 0; 753 return 0;
1026 } 754 }
1027 755
1028 static void 756 static void
1029 ref_scale_print_module_parms(const struct ref !! 757 ref_scale_print_module_parms(struct ref_scale_ops *cur_ops, const char *tag)
1030 { 758 {
1031 pr_alert("%s" SCALE_FLAG 759 pr_alert("%s" SCALE_FLAG
1032 "--- %s: verbose=%d verbose !! 760 "--- %s: verbose=%d shutdown=%d holdoff=%d loops=%ld nreaders=%d nruns=%d readdelay=%d\n", scale_type, tag,
1033 verbose, verbose_batched, sh !! 761 verbose, shutdown, holdoff, loops, nreaders, nruns, readdelay);
1034 } 762 }
1035 763
1036 static void 764 static void
1037 ref_scale_cleanup(void) 765 ref_scale_cleanup(void)
1038 { 766 {
1039 int i; 767 int i;
1040 768
1041 if (torture_cleanup_begin()) 769 if (torture_cleanup_begin())
1042 return; 770 return;
1043 771
1044 if (!cur_ops) { 772 if (!cur_ops) {
1045 torture_cleanup_end(); 773 torture_cleanup_end();
1046 return; 774 return;
1047 } 775 }
1048 776
1049 if (reader_tasks) { 777 if (reader_tasks) {
1050 for (i = 0; i < nreaders; i++ 778 for (i = 0; i < nreaders; i++)
1051 torture_stop_kthread( 779 torture_stop_kthread("ref_scale_reader",
1052 780 reader_tasks[i].task);
1053 } 781 }
1054 kfree(reader_tasks); 782 kfree(reader_tasks);
1055 783
1056 torture_stop_kthread("main_task", mai 784 torture_stop_kthread("main_task", main_task);
1057 kfree(main_task); 785 kfree(main_task);
1058 786
1059 // Do scale-type-specific cleanup ope 787 // Do scale-type-specific cleanup operations.
1060 if (cur_ops->cleanup != NULL) 788 if (cur_ops->cleanup != NULL)
1061 cur_ops->cleanup(); 789 cur_ops->cleanup();
1062 790
1063 torture_cleanup_end(); 791 torture_cleanup_end();
1064 } 792 }
1065 793
1066 // Shutdown kthread. Just waits to be awaken 794 // Shutdown kthread. Just waits to be awakened, then shuts down system.
1067 static int 795 static int
1068 ref_scale_shutdown(void *arg) 796 ref_scale_shutdown(void *arg)
1069 { 797 {
1070 wait_event_idle(shutdown_wq, shutdown 798 wait_event_idle(shutdown_wq, shutdown_start);
1071 799
1072 smp_mb(); // Wake before output. 800 smp_mb(); // Wake before output.
1073 ref_scale_cleanup(); 801 ref_scale_cleanup();
1074 kernel_power_off(); 802 kernel_power_off();
1075 803
1076 return -EINVAL; 804 return -EINVAL;
1077 } 805 }
1078 806
1079 static int __init 807 static int __init
1080 ref_scale_init(void) 808 ref_scale_init(void)
1081 { 809 {
1082 long i; 810 long i;
1083 int firsterr = 0; 811 int firsterr = 0;
1084 static const struct ref_scale_ops *sc !! 812 static struct ref_scale_ops *scale_ops[] = {
1085 &rcu_ops, &srcu_ops, RCU_TRAC 813 &rcu_ops, &srcu_ops, RCU_TRACE_OPS RCU_TASKS_OPS &refcnt_ops, &rwlock_ops,
1086 &rwsem_ops, &lock_ops, &lock_ !! 814 &rwsem_ops, &lock_ops, &lock_irq_ops, &acqrel_ops, &clock_ops,
1087 &typesafe_ref_ops, &typesafe_ <<
1088 }; 815 };
1089 816
1090 if (!torture_init_begin(scale_type, v 817 if (!torture_init_begin(scale_type, verbose))
1091 return -EBUSY; 818 return -EBUSY;
1092 819
1093 for (i = 0; i < ARRAY_SIZE(scale_ops) 820 for (i = 0; i < ARRAY_SIZE(scale_ops); i++) {
1094 cur_ops = scale_ops[i]; 821 cur_ops = scale_ops[i];
1095 if (strcmp(scale_type, cur_op 822 if (strcmp(scale_type, cur_ops->name) == 0)
1096 break; 823 break;
1097 } 824 }
1098 if (i == ARRAY_SIZE(scale_ops)) { 825 if (i == ARRAY_SIZE(scale_ops)) {
1099 pr_alert("rcu-scale: invalid 826 pr_alert("rcu-scale: invalid scale type: \"%s\"\n", scale_type);
1100 pr_alert("rcu-scale types:"); 827 pr_alert("rcu-scale types:");
1101 for (i = 0; i < ARRAY_SIZE(sc 828 for (i = 0; i < ARRAY_SIZE(scale_ops); i++)
1102 pr_cont(" %s", scale_ 829 pr_cont(" %s", scale_ops[i]->name);
1103 pr_cont("\n"); 830 pr_cont("\n");
1104 firsterr = -EINVAL; 831 firsterr = -EINVAL;
1105 cur_ops = NULL; 832 cur_ops = NULL;
1106 goto unwind; 833 goto unwind;
1107 } 834 }
1108 if (cur_ops->init) 835 if (cur_ops->init)
1109 if (!cur_ops->init()) { !! 836 cur_ops->init();
1110 firsterr = -EUCLEAN; <<
1111 goto unwind; <<
1112 } <<
1113 837
1114 ref_scale_print_module_parms(cur_ops, 838 ref_scale_print_module_parms(cur_ops, "Start of test");
1115 839
1116 // Shutdown task 840 // Shutdown task
1117 if (shutdown) { 841 if (shutdown) {
1118 init_waitqueue_head(&shutdown 842 init_waitqueue_head(&shutdown_wq);
1119 firsterr = torture_create_kth 843 firsterr = torture_create_kthread(ref_scale_shutdown, NULL,
1120 844 shutdown_task);
1121 if (torture_init_error(firste 845 if (torture_init_error(firsterr))
1122 goto unwind; 846 goto unwind;
1123 schedule_timeout_uninterrupti 847 schedule_timeout_uninterruptible(1);
1124 } 848 }
1125 849
1126 // Reader tasks (default to ~75% of o 850 // Reader tasks (default to ~75% of online CPUs).
1127 if (nreaders < 0) 851 if (nreaders < 0)
1128 nreaders = (num_online_cpus() 852 nreaders = (num_online_cpus() >> 1) + (num_online_cpus() >> 2);
1129 if (WARN_ONCE(loops <= 0, "%s: loops 853 if (WARN_ONCE(loops <= 0, "%s: loops = %ld, adjusted to 1\n", __func__, loops))
1130 loops = 1; 854 loops = 1;
1131 if (WARN_ONCE(nreaders <= 0, "%s: nre 855 if (WARN_ONCE(nreaders <= 0, "%s: nreaders = %d, adjusted to 1\n", __func__, nreaders))
1132 nreaders = 1; 856 nreaders = 1;
1133 if (WARN_ONCE(nruns <= 0, "%s: nruns 857 if (WARN_ONCE(nruns <= 0, "%s: nruns = %d, adjusted to 1\n", __func__, nruns))
1134 nruns = 1; 858 nruns = 1;
1135 reader_tasks = kcalloc(nreaders, size 859 reader_tasks = kcalloc(nreaders, sizeof(reader_tasks[0]),
1136 GFP_KERNEL); 860 GFP_KERNEL);
1137 if (!reader_tasks) { 861 if (!reader_tasks) {
1138 SCALEOUT_ERRSTRING("out of me 862 SCALEOUT_ERRSTRING("out of memory");
1139 firsterr = -ENOMEM; 863 firsterr = -ENOMEM;
1140 goto unwind; 864 goto unwind;
1141 } 865 }
1142 866
1143 VERBOSE_SCALEOUT("Starting %d reader 867 VERBOSE_SCALEOUT("Starting %d reader threads", nreaders);
1144 868
1145 for (i = 0; i < nreaders; i++) { 869 for (i = 0; i < nreaders; i++) {
1146 init_waitqueue_head(&reader_t 870 init_waitqueue_head(&reader_tasks[i].wq);
1147 firsterr = torture_create_kth 871 firsterr = torture_create_kthread(ref_scale_reader, (void *)i,
1148 872 reader_tasks[i].task);
1149 if (torture_init_error(firste 873 if (torture_init_error(firsterr))
1150 goto unwind; 874 goto unwind;
1151 } 875 }
1152 876
1153 // Main Task 877 // Main Task
1154 init_waitqueue_head(&main_wq); 878 init_waitqueue_head(&main_wq);
1155 firsterr = torture_create_kthread(mai 879 firsterr = torture_create_kthread(main_func, NULL, main_task);
1156 if (torture_init_error(firsterr)) 880 if (torture_init_error(firsterr))
1157 goto unwind; 881 goto unwind;
1158 882
1159 torture_init_end(); 883 torture_init_end();
1160 return 0; 884 return 0;
1161 885
1162 unwind: 886 unwind:
1163 torture_init_end(); 887 torture_init_end();
1164 ref_scale_cleanup(); 888 ref_scale_cleanup();
1165 if (shutdown) { 889 if (shutdown) {
1166 WARN_ON(!IS_MODULE(CONFIG_RCU 890 WARN_ON(!IS_MODULE(CONFIG_RCU_REF_SCALE_TEST));
1167 kernel_power_off(); 891 kernel_power_off();
1168 } 892 }
1169 return firsterr; 893 return firsterr;
1170 } 894 }
1171 895
1172 module_init(ref_scale_init); 896 module_init(ref_scale_init);
1173 module_exit(ref_scale_cleanup); 897 module_exit(ref_scale_cleanup);
1174 898
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.