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