aboutsummaryrefslogtreecommitdiffstats
path: root/litmus/sched_litmus.c
blob: 9a6fe487718ee5b8bc14b534accea73554f5ba6f (plain) (blame)
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
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
/* This file is included from kernel/sched.c */

#include <litmus/litmus.h>
#include <litmus/budget.h>
#include <litmus/sched_plugin.h>
#include <litmus/preempt.h>

static void update_time_litmus(struct rq *rq, struct task_struct *p)
{
	u64 delta = rq->clock - p->se.exec_start;
	if (unlikely((s64)delta < 0))
		delta = 0;
	/* per job counter */
	p->rt_param.job_params.exec_time += delta;
	/* task counter */
	p->se.sum_exec_runtime += delta;
	/* sched_clock() */
	p->se.exec_start = rq->clock;
	cpuacct_charge(p, delta);
}

static void double_rq_lock(struct rq *rq1, struct rq *rq2);
static void double_rq_unlock(struct rq *rq1, struct rq *rq2);

/*
 * litmus_tick gets called by scheduler_tick() with HZ freq
 * Interrupts are disabled
 */
static void litmus_tick(struct rq *rq, struct task_struct *p)
{
	TS_PLUGIN_TICK_START;

	if (is_realtime(p))
		update_time_litmus(rq, p);

	/* plugin tick */
	litmus->tick(p);

	TS_PLUGIN_TICK_END;

	return;
}

static struct task_struct *
litmus_schedule(struct rq *rq, struct task_struct *prev)
{
	struct rq* other_rq;
	struct task_struct *next;

	long was_running;
	lt_t _maybe_deadlock = 0;

	/* let the plugin schedule */
	next = litmus->schedule(prev);

	sched_state_plugin_check();

	/* check if a global plugin pulled a task from a different RQ */
	if (next && task_rq(next) != rq) {
		/* we need to migrate the task */
		other_rq = task_rq(next);
		TRACE_TASK(next, "migrate from %d\n", other_rq->cpu);

		/* while we drop the lock, the prev task could change its
		 * state
		 */
		was_running = is_running(prev);
		mb();
		raw_spin_unlock(&rq->lock);

		/* Don't race with a concurrent switch.  This could deadlock in
		 * the case of cross or circular migrations.  It's the job of
		 * the plugin to make sure that doesn't happen.
		 */
		TRACE_TASK(next, "stack_in_use=%d\n",
			   next->rt_param.stack_in_use);
		if (next->rt_param.stack_in_use != NO_CPU) {
			TRACE_TASK(next, "waiting to deschedule\n");
			_maybe_deadlock = litmus_clock();
		}
		while (next->rt_param.stack_in_use != NO_CPU) {
			cpu_relax();
			mb();
			if (next->rt_param.stack_in_use == NO_CPU)
				TRACE_TASK(next,"descheduled. Proceeding.\n");

			if (lt_before(_maybe_deadlock + 10000000,
				      litmus_clock())) {
				/* We've been spinning for 10ms.
				 * Something can't be right!
				 * Let's abandon the task and bail out; at least
				 * we will have debug info instead of a hard
				 * deadlock.
				 */
				TRACE_TASK(next,"stack too long in use. "
					   "Deadlock?\n");
				next = NULL;

				/* bail out */
				raw_spin_lock(&rq->lock);
				return next;
			}
		}
#ifdef  __ARCH_WANT_UNLOCKED_CTXSW
		if (next->oncpu)
		{
			TRACE_TASK(next, "waiting for !oncpu");
		}
		while (next->oncpu) {
			cpu_relax();
			mb();
		}
#endif
		double_rq_lock(rq, other_rq);
		mb();
		if (is_realtime(prev) && is_running(prev) != was_running) {
			TRACE_TASK(prev,
				   "state changed while we dropped"
				   " the lock: is_running=%d, was_running=%d\n",
				   is_running(prev), was_running);
			if (is_running(prev) && !was_running) {
				/* prev task became unblocked
				 * we need to simulate normal sequence of events
				 * to scheduler plugins.
				 */
				litmus->task_block(prev);
				litmus->task_wake_up(prev);
			}
		}

		set_task_cpu(next, smp_processor_id());

		/* DEBUG: now that we have the lock we need to make sure a
		 *  couple of things still hold:
		 *  - it is still a real-time task
		 *  - it is still runnable (could have been stopped)
		 * If either is violated, then the active plugin is
		 * doing something wrong.
		 */
		if (!is_realtime(next) || !is_running(next)) {
			/* BAD BAD BAD */
			TRACE_TASK(next,"BAD: migration invariant FAILED: "
				   "rt=%d running=%d\n",
				   is_realtime(next),
				   is_running(next));
			/* drop the task */
			next = NULL;
		}
		/* release the other CPU's runqueue, but keep ours */
		raw_spin_unlock(&other_rq->lock);
	}
	if (next) {
		next->rt_param.stack_in_use = rq->cpu;
		next->se.exec_start = rq->clock;
	}

	update_enforcement_timer(next);
	return next;
}

static void enqueue_task_litmus(struct rq *rq, struct task_struct *p,
				int flags)
{
	if (flags & ENQUEUE_WAKEUP) {
		sched_trace_task_resume(p);
		tsk_rt(p)->present = 1;
		/* LITMUS^RT plugins need to update the state
		 * _before_ making it available in global structures.
		 * Linux gets away with being lazy about the task state
		 * update. We can't do that, hence we update the task
		 * state already here.
		 *
		 * WARNING: this needs to be re-evaluated when porting
		 *          to newer kernel versions.
		 */
		p->state = TASK_RUNNING;
		litmus->task_wake_up(p);

		rq->litmus.nr_running++;
	} else
		TRACE_TASK(p, "ignoring an enqueue, not a wake up.\n");
}

static void dequeue_task_litmus(struct rq *rq, struct task_struct *p,
				int flags)
{
	if (flags & DEQUEUE_SLEEP) {
		litmus->task_block(p);
		tsk_rt(p)->present = 0;
		sched_trace_task_block(p);

		rq->litmus.nr_running--;
	} else
		TRACE_TASK(p, "ignoring a dequeue, not going to sleep.\n");
}

static void yield_task_litmus(struct rq *rq)
{
	BUG_ON(rq->curr != current);
	/* sched_yield() is called to trigger delayed preemptions.
	 * Thus, mark the current task as needing to be rescheduled.
	 * This will cause the scheduler plugin to be invoked, which can
	 * then determine if a preemption is still required.
	 */
	clear_exit_np(current);
	litmus_reschedule_local();
}

/* Plugins are responsible for this.
 */
static void check_preempt_curr_litmus(struct rq *rq, struct task_struct *p, int flags)
{
}

static void put_prev_task_litmus(struct rq *rq, struct task_struct *p)
{
}

static void pre_schedule_litmus(struct rq *rq, struct task_struct *prev)
{
	update_time_litmus(rq, prev);
	if (!is_running(prev))
		tsk_rt(prev)->present = 0;
}

/* pick_next_task_litmus() - litmus_schedule() function
 *
 * return the next task to be scheduled
 */
static struct task_struct *pick_next_task_litmus(struct rq *rq)
{
	/* get the to-be-switched-out task (prev) */
	struct task_struct *prev = rq->litmus.prev;
	struct task_struct *next;

	/* if not called from schedule() but from somewhere
	 * else (e.g., migration), return now!
	 */
	if(!rq->litmus.prev)
		return NULL;

	rq->litmus.prev = NULL;

	TS_PLUGIN_SCHED_START;
	next = litmus_schedule(rq, prev);
	TS_PLUGIN_SCHED_END;

	return next;
}

static void task_tick_litmus(struct rq *rq, struct task_struct *p, int queued)
{
	/* nothing to do; tick related tasks are done by litmus_tick() */
	return;
}

static void switched_to_litmus(struct rq *rq, struct task_struct *p)
{
}

static void prio_changed_litmus(struct rq *rq, struct task_struct *p,
				int oldprio)
{
}

unsigned int get_rr_interval_litmus(struct rq *rq, struct task_struct *p)
{
	/* return infinity */
	return 0;
}

/* This is called when a task became a real-time task, either due to a SCHED_*
 * class transition or due to PI mutex inheritance. We don't handle Linux PI
 * mutex inheritance yet (and probably never will). Use LITMUS provided
 * synchronization primitives instead.
 */
static void set_curr_task_litmus(struct rq *rq)
{
	rq->curr->se.exec_start = rq->clock;
}


#ifdef CONFIG_SMP
/* execve tries to rebalance task in this scheduling domain.
 * We don't care about the scheduling domain; can gets called from
 * exec, fork, wakeup.
 */
static int
select_task_rq_litmus(struct task_struct *p, int sd_flag, int flags)
{
	/* preemption is already disabled.
	 * We don't want to change cpu here
	 */
	return task_cpu(p);
}
#endif

static const struct sched_class litmus_sched_class = {
	/* From 34f971f6 the stop/migrate worker threads have a class on
	 * their own, which is the highest prio class. We don't support
	 * cpu-hotplug or cpu throttling. Allows Litmus to use up to 1.0
	 * CPU capacity.
	 */
	.next			= &stop_sched_class,
	.enqueue_task		= enqueue_task_litmus,
	.dequeue_task		= dequeue_task_litmus,
	.yield_task		= yield_task_litmus,

	.check_preempt_curr	= check_preempt_curr_litmus,

	.pick_next_task		= pick_next_task_litmus,
	.put_prev_task		= put_prev_task_litmus,

#ifdef CONFIG_SMP
	.select_task_rq		= select_task_rq_litmus,

	.pre_schedule		= pre_schedule_litmus,
#endif

	.set_curr_task          = set_curr_task_litmus,
	.task_tick		= task_tick_litmus,

	.get_rr_interval	= get_rr_interval_litmus,

	.prio_changed		= prio_changed_litmus,
	.switched_to		= switched_to_litmus,
};