1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
|
/*
* pm_task NO CONTROL PAGE REAL USAGE
*/
/* common data structures and defines */
#include "pm_common.h"
#include "litmus.h"
#include "asm.h"
#include "cycles.h"
/* architectural dependend code for pm measurement */
#include "pm_arch.h"
struct fake_control_page {
/* preemption and migration overheads measurement*/
/* execution start time */
lt_t preempt_end;
/* increased if last_rt != current */
unsigned long sched_count;
unsigned int cpu;
/* job number for this task */
unsigned long job_count;
/* last rt task before current one */
unsigned long last_rt_task;
/* execution stop time */
lt_t preempt_start;
/* Is the task currently in a non-preemptive section? */
int np_flag;
/* Should the task call into the kernel when it leaves
* * its non-preemptive section? */
int delayed_preemption;
/* what do you want to share today ? */
};
int mem_block[NUMWS][INTS_PER_WSS] __attribute__ ((aligned(CACHEALIGNMENT)));
/* Setup flags, then enter loop to measure costs. */
int main(int argc, char **argv)
{
/* control_page to read data from kernel */
struct fake_control_page ctrl = {0};
unsigned long curr_job_count = 0;
unsigned long curr_sched_count = 0;
unsigned int curr_cpu = 0;
unsigned long curr_last_rt_task = 0;
unsigned long curr_ws = 0;
unsigned long long curr_preemption_length = 0;
unsigned long long start_time, end_time;
int *mem_ptr = NULL;
int *mem_ptr_end = NULL;
struct data_entry data_points[DATAPOINTS];
int data_count = 0;
int data_wrapped = 0;
int refcount;
int task_pid = gettid();
int task_period;
int read, *loc_ptr;
struct rt_task param;
char *filename;
#ifdef DEBUG
int i;
#endif
if (argc < 2) {
printf("pm_task: need a filename\n");
return -1;
}
filename = argv[1];
#ifdef DEBUG
fprintf(stderr, "Saving on %s\n",filename);
#endif
/* Initialize random library for read/write ratio enforcement. */
srandom(SEEDVAL);
/* this will lock all pages and will call init_kernel_iface */
// init_litmus();
/* Lock all pages for this task into memory. */
if (mlockall(MCL_CURRENT | MCL_FUTURE) < 0) {
perror("Cannot lock memory pages!\n");
printf("Are you root?\n");
return -1;
}
/* Ensure that the pages that we care about, either because they
* are shared with the kernel or they are performance-critical,
* are loaded and locked in memory before benchmarking begins.
*/
memset(¶m, 0, sizeof(struct rt_task));
memset(&mem_block, 0, sizeof(int) * NUMWS * INTS_PER_WSS);
memset(&mem_ptr, 0, sizeof(int*));
memset(&mem_ptr_end, 0, sizeof(int*));
/* Get task period. */
if (get_rt_task_param(task_pid, ¶m) < 0) {
perror("Cannot get task parameters\n");
return -1;
}
task_period = param.period / NS_PER_MS;
ctrl.job_count = 1;
/* Enter loop that measures preemption and migration costs. */
while (curr_job_count * task_period < SIMRUNTIME) {
if (curr_job_count != ctrl.job_count) {
/* ok, this is a new job. Get info from kernel */
curr_job_count = ctrl.job_count;
curr_sched_count = ctrl.sched_count;
curr_cpu = ctrl.cpu;
curr_last_rt_task = ctrl.last_rt_task;
barrier();
/* job's portion of the mem_block */
curr_ws = curr_job_count % NUMWS;
mem_ptr = &mem_block[curr_ws][0];
mem_ptr_end = mem_ptr + INTS_PER_WSS;
/* Access WS when cache cold, then immediately
* re-access to calculate "cache-hot" access time.
*/
/* Cache-cold accesses. */
start_time = get_cycles();
for (; mem_ptr < mem_ptr_end; mem_ptr += 1024)
readwrite_one_thousand_ints(mem_ptr);
end_time = get_cycles();
data_points[data_count].timestamp = end_time;
/* Am I the same I was before? */
if (curr_job_count != ctrl.job_count ||
curr_sched_count != ctrl.sched_count ||
curr_cpu != ctrl.cpu)
/* fishiness */
data_points[data_count].access_type = 'c';
else
/* okay */
data_points[data_count].access_type = 'C';
data_points[data_count].access_time =
end_time - start_time;
data_points[data_count].cpu = curr_cpu;
data_points[data_count].job_count = curr_job_count;
data_points[data_count].sched_count = curr_sched_count;
data_points[data_count].last_rt_task = curr_last_rt_task;
data_points[data_count].preemption_length = 0;
data_wrapped = ((data_count+1) / DATAPOINTS > 0);
data_count = (data_count+1) % DATAPOINTS;
barrier();
/* "Best case". Read multiple times. */
for (refcount = 0; refcount < REFTOTAL; refcount++) {
mem_ptr = &mem_block[curr_ws][0];
start_time = get_cycles();
for (; mem_ptr < mem_ptr_end; mem_ptr += 1024)
readwrite_one_thousand_ints(mem_ptr);
end_time = get_cycles();
data_points[data_count].timestamp = end_time;
if (curr_job_count != ctrl.job_count ||
curr_sched_count != ctrl.sched_count ||
curr_cpu != ctrl.cpu)
/* fishiness */
data_points[data_count].
access_type = 'h';
else
/* okay */
data_points[data_count].
access_type = 'H';
data_points[data_count].access_time =
end_time - start_time;
data_points[data_count].cpu = curr_cpu;
data_points[data_count].job_count =
curr_job_count;
data_points[data_count].sched_count =
curr_sched_count;
data_points[data_count].last_rt_task =
curr_last_rt_task;
data_points[data_count].preemption_length = 0;
data_wrapped =
((data_count+1) / DATAPOINTS > 0);
data_count = (data_count+1) % DATAPOINTS;
}
} else if (mem_ptr && mem_ptr_end &&
(curr_sched_count != ctrl.sched_count ||
curr_cpu != ctrl.cpu)) {
/* we have done at least one go in the "best case".
* job is the same => preempted / migrated
*/
curr_preemption_length =
ctrl.preempt_end - ctrl.preempt_start;
curr_job_count = ctrl.job_count;
curr_sched_count = ctrl.sched_count;
curr_cpu = ctrl.cpu;
curr_last_rt_task = ctrl.last_rt_task;
barrier();
/* Measure preemption or migration cost. */
mem_ptr = &mem_block[curr_ws][0];
start_time = get_cycles();
for (; mem_ptr < mem_ptr_end; mem_ptr += 1024)
readwrite_one_thousand_ints(mem_ptr);
end_time = get_cycles();
data_points[data_count].timestamp = end_time;
/* just record pP, we tell the difference later */
if (curr_job_count != ctrl.job_count ||
curr_sched_count != ctrl.sched_count ||
curr_cpu != ctrl.cpu)
/* fishiness */
data_points[data_count].access_type = 'p';
else
/* okay */
data_points[data_count].access_type = 'P';
data_points[data_count].access_time =
end_time - start_time;
data_points[data_count].cpu = curr_cpu;
data_points[data_count].job_count = curr_job_count;
data_points[data_count].sched_count = curr_sched_count;
data_points[data_count].last_rt_task =
curr_last_rt_task;
data_points[data_count].preemption_length =
curr_preemption_length;
data_wrapped = ((data_count+1) / DATAPOINTS > 0);
data_count = (data_count+1) % DATAPOINTS;
} else if (mem_ptr && mem_ptr_end) {
/*
* Ok, we run:
* Read or write some random location in the WS
* to keep the task "cache warm". We only do
* this if the pointers are valid, because we
* do not want to skew the "cold" read of the WS
* on the first job.
*/
read = (random() % 100) < READRATIO;
loc_ptr = &mem_block[curr_ws][0];
loc_ptr += (random() % INTS_PER_WSS);
barrier();
if (read) {
read_mem(loc_ptr);
ctrl.job_count++;
}
else
write_mem(loc_ptr);
}
}
#ifdef DEBUG
/* Print (most recent) results. */
for (i = 0; i < (data_wrapped ? DATAPOINTS : data_count) ; i++)
fprintf(stderr, "(%c) - ACC %llu, CPU %u, PLEN %llu\n",
data_points[i].access_type,
data_points[i].access_time, data_points[i].cpu,
data_points[i].preemption_length);
#endif
serialize_data_entry(filename, data_points,
(data_wrapped ? DATAPOINTS : data_count));
return 0;
}
|
