aboutsummaryrefslogtreecommitdiffstats
path: root/include/litmus/rt_param.h
blob: d63821d9467a87a87d90563d120c06d0b8ddb43c (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
/*
 * Definition of the scheduler plugin interface.
 *
 */
#ifndef _LINUX_RT_PARAM_H_
#define _LINUX_RT_PARAM_H_

/* Litmus time type. */
typedef unsigned long long lt_t;

static inline int lt_after(lt_t a, lt_t b)
{
	return ((long long) b) - ((long long) a) < 0;
}
#define lt_before(a, b) lt_after(b, a)

static inline int lt_after_eq(lt_t a, lt_t b)
{
	return ((long long) a) - ((long long) b) >= 0;
}
#define lt_before_eq(a, b) lt_after_eq(b, a)

/* different types of clients */
typedef enum {
	RT_CLASS_HARD,
	RT_CLASS_SOFT,
	RT_CLASS_BEST_EFFORT
} task_class_t;

typedef enum {
	NO_ENFORCEMENT,      /* job may overrun unhindered */
	QUANTUM_ENFORCEMENT, /* budgets are only checked on quantum boundaries */
	PRECISE_ENFORCEMENT  /* budgets are enforced with hrtimers */
} budget_policy_t;

/* We use the common priority interpretation "lower index == higher priority",
 * which is commonly used in fixed-priority schedulability analysis papers.
 * So, a numerically lower priority value implies higher scheduling priority,
 * with priority 1 being the highest priority. Priority 0 is reserved for
 * priority boosting. LITMUS_MAX_PRIORITY denotes the maximum priority value
 * range.
 */

#define LITMUS_MAX_PRIORITY     512
#define LITMUS_HIGHEST_PRIORITY   1
#define LITMUS_LOWEST_PRIORITY    (LITMUS_MAX_PRIORITY - 1)

/* Provide generic comparison macros for userspace,
 * in case that we change this later. */
#define litmus_higher_fixed_prio(a, b)	(a < b)
#define litmus_lower_fixed_prio(a, b)	(a > b)
#define litmus_is_valid_fixed_prio(p)		\
	((p) >= LITMUS_HIGHEST_PRIORITY &&	\
	 (p) <= LITMUS_LOWEST_PRIORITY)

struct rt_task {
	lt_t 		exec_cost;
	lt_t 		period;
	lt_t		relative_deadline;
	lt_t		phase;
	unsigned int	cpu;
	unsigned int	priority;
	task_class_t	cls;
	budget_policy_t budget_policy; /* ignored by pfair */
};

union np_flag {
	uint32_t raw;
	struct {
		/* Is the task currently in a non-preemptive section? */
		uint32_t flag:31;
		/* Should the task call into the scheduler? */
		uint32_t preempt:1;
	} np;
};

/* The definition of the data that is shared between the kernel and real-time
 * tasks via a shared page (see litmus/ctrldev.c).
 *
 * WARNING: User space can write to this, so don't trust
 * the correctness of the fields!
 *
 * This servees two purposes: to enable efficient signaling
 * of non-preemptive sections (user->kernel) and
 * delayed preemptions (kernel->user), and to export
 * some real-time relevant statistics such as preemption and
 * migration data to user space. We can't use a device to export
 * statistics because we want to avoid system call overhead when
 * determining preemption/migration overheads).
 */
struct control_page {
	volatile union np_flag sched;

	/* locking overhead tracing: time stamp prior to system call */
	uint64_t ts_syscall_start; /* Feather-Trace cycles */

	int colors_updated:8;
};

#ifndef __KERNEL__
/*
 * XXX This is a terrible hack so liblitmus can use the PAGE_SIZE macro.
 * We should fix liblitmus to do setup the page size at runtime.
 */
#define CACHE_LINE_SIZE 64
#if (ARCH == x86_64)
#define PAGE_SIZE 4096
#elif (ARCH == sparc64)
#define PAGE_SIZE 8192
#endif
#endif /* ifndef __KERNEL__ */

typedef uint8_t color_t;
#define COLORS_PER_CONTROL_PAGE (PAGE_SIZE / (2 * sizeof(color_t)))
struct color_ctrl_page {
	color_t colors[COLORS_PER_CONTROL_PAGE];
	/* must be same type to guarantee equal array sizes */
	color_t pages[COLORS_PER_CONTROL_PAGE];
};

/* don't export internal data structures to user space (liblitmus) */
#ifdef __KERNEL__

struct _rt_domain;
struct bheap_node;
struct release_heap;
struct domain;
struct rt_server;
struct dgl_group_req;

struct rt_job {
	/* Time instant the the job was or will be released.  */
	lt_t	release;
	/* What is the current deadline? */
	lt_t   	deadline;

	lt_t	real_release;
	lt_t	real_deadline;
	unsigned int fake_job_no;

	/* How much service has this job received so far? */
	lt_t	exec_time;

	/* By how much did the prior job miss its deadline by?
	 * Value differs from tardiness in that lateness may
	 * be negative (when job finishes before its deadline).
	 */
	long long	lateness;

	/* Which job is this. This is used to let user space
	 * specify which job to wait for, which is important if jobs
	 * overrun. If we just call sys_sleep_next_period() then we
	 * will unintentionally miss jobs after an overrun.
	 *
	 * Increase this sequence number when a job is released.
	 */
	unsigned int    job_no;
};

struct pfair_param;
#ifdef CONFIG_PLUGIN_MC
struct mc_data;
#endif

/*	RT task parameters for scheduling extensions
 *	These parameters are inherited during clone and therefore must
 *	be explicitly set up before the task set is launched.
 */
struct rt_param {
	/* is the task sleeping? */
	unsigned int 		flags:8;

	/* do we need to check for srp blocking? */
	unsigned int		srp_non_recurse:1;

	/* is the task present? (true if it can be scheduled) */
	unsigned int		present:1;

	unsigned int		is_server:1;

#ifdef CONFIG_LITMUS_LOCKING
	/* Is the task being priority-boosted by a locking protocol? */
	unsigned int		priority_boosted:1;
	/* If so, when did this start? */
	lt_t			boost_start_time;
#endif

#ifdef CONFIG_PLUGIN_MC
	/* mixed criticality specific data */
	struct mc_data *mc_data;
#endif
#ifdef CONFIG_MERGE_TIMERS
	struct rt_event *event;
#endif

	struct rt_server *server;


	/* user controlled parameters */
	struct rt_task 		task_params;

	/* kernelspace view of the job */
	struct rt_job		job_params;
	/* userspace view of the job */
	struct rt_job 		user_job;

	/* task representing the current "inherited" task
	 * priority, assigned by inherit_priority and
	 * return priority in the scheduler plugins.
	 * could point to self if PI does not result in
	 * an increased task priority.
	 */
	 struct task_struct*	inh_task;

#ifdef CONFIG_NP_SECTION
	/* For the FMLP under PSN-EDF, it is required to make the task
	 * non-preemptive from kernel space. In order not to interfere with
	 * user space, this counter indicates the kernel space np setting.
	 * kernel_np > 0 => task is non-preemptive
	 */
	unsigned int	kernel_np;
#endif

	/* This field can be used by plugins to store where the task
	 * is currently scheduled. It is the responsibility of the
	 * plugin to avoid race conditions.
	 *
	 * This used by GSN-EDF and PFAIR.
	 */
	volatile int		scheduled_on;

	/* Is the stack of the task currently in use? This is updated by
	 * the LITMUS core.
	 *
	 * Be careful to avoid deadlocks!
	 */
	volatile int		stack_in_use;

	/* This field can be used by plugins to store where the task
	 * is currently linked. It is the responsibility of the plugin
	 * to avoid race conditions.
	 *
	 * Used by GSN-EDF.
	 */
	volatile int		linked_on;

	/* PFAIR/PD^2 state. Allocated on demand. */
	struct pfair_param*	pfair;

	/* Fields saved before BE->RT transition.
	 */
	int old_policy;
	int old_prio;

	/* TODO: rename */
	struct domain *_domain;

	/* ready queue for this task */
	struct _rt_domain* domain;

	/* heap element for this task
	 *
	 * Warning: Don't statically allocate this node. The heap
	 *          implementation swaps these between tasks, thus after
	 *          dequeuing from a heap you may end up with a different node
	 *          then the one you had when enqueuing the task.  For the same
	 *          reason, don't obtain and store references to this node
	 *          other than this pointer (which is updated by the heap
	 *          implementation).
	 */
	struct bheap_node*	heap_node;
	struct release_heap*	rel_heap;

	/* Used by rt_domain to queue task in release list.
	 */
	struct list_head list;

	/* Pointer to the page shared between userspace and kernel. */
	struct control_page * ctrl_page;

	lt_t		total_tardy;
	lt_t		max_tardy;
	unsigned int    missed;

	lt_t max_exec_time;
	lt_t tot_exec_time;
	lt_t last_exec_time;
	struct color_ctrl_page *color_ctrl_page;
	struct dgl_group_req *req;
};

/*	Possible RT flags	*/
#define RT_F_RUNNING		0x00000000
#define RT_F_SLEEP		0x00000001
#define RT_F_EXIT_SEM		0x00000008

#endif

#endif