xref: /xnu-11215/bsd/sys/work_interval.h (revision 1031c584)
1 /*
2  * Copyright (c) 2015-2017 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 #ifndef _SYS_WORK_INTERVAL_H
30 #define _SYS_WORK_INTERVAL_H
31 
32 #include <stdint.h>
33 #include <sys/cdefs.h>
34 #include <sys/_types/_size_t.h>
35 
36 #include <mach/port.h>
37 
38 __BEGIN_DECLS
39 
40 /*
41  * A work interval is a repeatable unit of work characterized by a
42  * start, finish, and deadline.
43  *
44  * Trusted clients with deadline-sensitive work may report information
45  * about the execution of their work using the work interval facility.
46  * This is intended to be a higher-level semantic than realtime scheduling,
47  * which operates at the level of thread block/unblock. A high level
48  * operation may have many blocking points, including IPC to other tasks,
49  * and this this metric will capture the overall time to complete a unit of
50  * work.
51  *
52  * A work interval is defined by several timestamps, namely (S)tart,
53  * (F)inish, (D)eadline, and (N)ext start.
54  *
55  *   ... ----+==================+--------+--+==== ...
56  *           |                  |        |  |
57  *           S                  F        D  N
58  *
59  *           \__________________/
60  *                  Active
61  *           \___________________________/
62  *                   Work Interval
63  *
64  *                               \_________/
65  *                                    |
66  *   report information here ---------+
67  *
68  * Definitions:
69  *
70  *   Start: Absolute time when the current deadline-oriented work began. Due
71  *          to scheduling latency, preemption, and blocking points, the
72  *          thread controlling the work interval may actually begin
73  *          executing after this ideal time (which may be the previous work
74  *          interval's "next start")
75  *   Finish: Absolute time when the current deadline-oriented work finished.
76  *          This will typically be a timestamp taken before reporting using
77  *          the work interval interface.
78  *   Deadline: Absolute time by which the current work was expected to finish.
79  *          In cases where the amount of computation (or preemption, or time
80  *          spent blocked) causes the active period to take longer than
81  *          expected, F may be greater than D.
82  *   Next start: Absolute time when the next deadline-oriented work is
83  *          expected to begin. This is typically the same as Deadline.
84  *   Active: The fraction of the work interval spent completing the work. In
85  *          cases where the Finish time exceeded the Deadline, this fraction
86  *          will be >1.0.
87  *
88  * Basic Use:
89  *
90  *   Clients should report information for a work interval after finishing
91  *   work for the current interval but before the next work interval begins.
92  *
93  *   If Finish far exceeds the previously expected Deadline, the
94  *   caller may adjust Next Start to align to a multiple of the period
95  *   (and skip over several work intervals that could not be
96  *   executed).
97  *
98  * Caution (!):
99  *
100  *   Because the information supplied via this facility directly influences power
101  *   management decisions, clients should strive to be as accurate as possible.
102  *   Failure to do so will adversely impact system power and performance.
103  *
104  * Work Interval Auto Join Support:
105  *
106  * Work intervals support an optional flag WORK_INTERVAL_FLAG_ENABLE_AUTO_JOIN
107  * which allows RT threads from the same home thread group to join work
108  * intervals via wakeup relationship tracking. Based on the join policy,
109  * RT threads can temporarily join the work interval of other RT threads
110  * which make them runnable. The auto joined thread remains in the work
111  * interval until it blocks or terminates. The mechanism works through
112  * make runnable heuristic and it should be used with extreme caution.
113  * If a client specifies this flag, it gives up explicit control over its
114  * thread group membership and threads unrelated to the work interval
115  * could become part of the thread group. This could lead to serious power
116  * and performance issues. If the make runnable heuristic does not work
117  * for a client use case, it should adopt work_interval_join_port() or
118  * work_interval_join() to explicitly declare its intent.
119  *
120  * Work Interval Deferred Finish Support:
121  *
122  * Another advanced feature for work intervals is the ability to defer the finish
123  * calls for the work interval until all auto-joined threads for the work interval
124  * have blocked or terminated. This feature is enabled via an optional flag
125  * WORK_INTERVAL_FLAG_ENABLE_DEFERRED_FINISH and is valid only if the work interval
126  * is configured with the WORK_INTERVAL_FLAG_ENABLE_AUTO_JOIN flag as well. The
127  * deferred finish mechanism allows the work interval to defer the finish call
128  * for the work interval until all auto-join threads have blocked/terminated
129  * (and have therefore un-joined the work interval) or one of the work interval
130  * threads calls start for the next frame. The deferred finish works only for
131  * workloads that have no temporal overlap across frames i.e. previous frame has to
132  * finish before next frame can start. This feature should be used with caution
133  * since auto-joined threads would delay finish calls to the performance controller
134  * which could lead to poor performance and battery life.
135  */
136 
137 /* Flags to be passed with work_interval_create() */
138 
139 /* If interval is joinable, create no longer implicitly joins, you must use work_interval_join */
140 #define WORK_INTERVAL_FLAG_JOINABLE                     (0x1)
141 /* Only threads that join the group are measured together, otherwise the group is the creator's home group */
142 #define WORK_INTERVAL_FLAG_GROUP                        (0x2)
143 /* Specifies that the work interval is being created by a client who doesn't
144  * necessarily have the PRIV_WORK_INTERVAL entitlement. Skip privilege checks.
145  * This can only be masked in for work intervals of types COREAUDIO, CA_CLIENT
146  * and DEFAULT */
147 #define WORK_INTERVAL_FLAG_UNRESTRICTED                 (0x4)
148 
149 /* [Advanced Flag] Read section on "Work Interval Auto Join Support" above for details */
150 #define WORK_INTERVAL_FLAG_ENABLE_AUTO_JOIN             (0x8)
151 /* [Advanced Flag] Read section on "Work Interval Deferred Finish Support" above for details */
152 #define WORK_INTERVAL_FLAG_ENABLE_DEFERRED_FINISH       (0x10)
153 
154 /* Kernel-supplied flag: Work interval has been ignored by the kernel */
155 #define WORK_INTERVAL_FLAG_IGNORED                      (0x20)
156 
157 /* Specifies that the work interval requests the system to provide just enough performance
158  * to be able to finish at the provided deadline and no sooner. */
159 #define WORK_INTERVAL_FLAG_FINISH_AT_DEADLINE           (0x40)
160 
161 /* Internal-only flag: workinterval will have the SET_WORKLOAD_ID operation called on it immediately
162  * after creation. */
163 #define WORK_INTERVAL_FLAG_HAS_WORKLOAD_ID              (0x80)
164 
165 /* Internal-only flag: Telemetry data will be aggregated from threads while they are joined to
166  * the work interval. */
167 #define WORK_INTERVAL_FLAG_ENABLE_TELEMETRY_DATA        (0x100)
168 
169 /* Flags to describe the interval flavor to the performance controller */
170 #define WORK_INTERVAL_TYPE_MASK                 (0xF0000000)
171 #define WORK_INTERVAL_TYPE_DEFAULT              (0x0 << 28)
172 #define WORK_INTERVAL_TYPE_COREAUDIO            (0x1 << 28)
173 #define WORK_INTERVAL_TYPE_COREANIMATION        (0x2 << 28)
174 #define WORK_INTERVAL_TYPE_CA_RENDER_SERVER     (0x2 << 28)
175 #define WORK_INTERVAL_TYPE_CA_CLIENT            (0x3 << 28)
176 #define WORK_INTERVAL_TYPE_HID_DELIVERY         (0x4 << 28)
177 #define WORK_INTERVAL_TYPE_COREMEDIA            (0x5 << 28)
178 #define WORK_INTERVAL_TYPE_ARKIT                (0x6 << 28)
179 #define WORK_INTERVAL_TYPE_FRAME_COMPOSITOR     (0x7 << 28)
180 #define WORK_INTERVAL_TYPE_LAST                 (0xF << 28)
181 
182 #ifndef KERNEL
183 
184 typedef struct work_interval *work_interval_t;
185 typedef struct work_interval_instance *work_interval_instance_t;
186 typedef struct work_interval_data *work_interval_data_t;
187 
188 /*
189  * Create a new work interval handle.
190  *
191  * May fail with EALREADY if the current group already has a work interval.
192  *
193  * With no flags:
194  *      Auto-joins the work interval to the creating thread
195  *      May only use interval_handle from creating thread
196  *      Data provided affects native thread group
197  *
198  * With the JOINABLE flag
199  *      interval_handle is usable by the process
200  *      creating thread does not auto-join
201  *      notifying thread must have joined when notifying
202  *
203  * With the GROUP flag
204  *      creates a new thread group to isolate the joined threads from
205  *      the rest of the process for performance controller analysis
206  *      Threads which join the work interval become members of this new group
207  *
208  * TODO: Add a name parameter so that clients can name the work interval
209  * Can also take the thread name from the notifying thread
210  *
211  * Requires the 'com.apple.private.kernel.work-interval' entitlement (PRIV_WORK_INTERVAL)
212  *
213  * Note that joining a work interval supersedes automatic thread group management via vouchers
214  */
215 int     work_interval_create(work_interval_t *interval_handle, uint32_t flags);
216 
217 /* Returns the flags used for the work interval when it was created.
218  *
219  * May fail with EINVAL if the port isn't from a prior call to
220  * work_interval_copy_port.
221  */
222 int work_interval_get_flags_from_port(mach_port_t port, uint32_t *flags);
223 
224 
225 /*
226  * Notify the power management subsystem that the work for a current interval has completed
227  *
228  * Only the process which created the work interval may notify
229  */
230 int     work_interval_notify(work_interval_t    interval_handle,
231     uint64_t start, uint64_t finish,
232     uint64_t deadline, uint64_t next_start,
233     uint32_t flags);
234 
235 /*
236  * Notify, with "finish" implicitly set to the current time
237  *
238  * Only the process which created the work interval may notify
239  */
240 int     work_interval_notify_simple(work_interval_t interval_handle,
241     uint64_t start, uint64_t deadline,
242     uint64_t next_start);
243 
244 /*
245  * Deallocate work interval handle
246  * For non-JOINABLE, also removes thread from work interval
247  * For JOINABLE, does not remove thread (needs a leave as well)
248  */
249 int     work_interval_destroy(work_interval_t interval_handle);
250 
251 /*
252  * Join work interval via work interval handle
253  * Only allowed if interval is using the joinable and group flags
254  *
255  * Supersedes automatic thread group management via vouchers
256  */
257 int     work_interval_join(work_interval_t interval_handle);
258 
259 /*
260  * extract Mach send right representing work interval thread group
261  * Returns a +1 send right ref, which must be deallocated via mach_port_deallocate
262  * Only allowed if interval is joinable, otherwise returns ENOTSUP
263  *
264  * Supersedes automatic thread group management via vouchers
265  */
266 int     work_interval_copy_port(work_interval_t interval_handle, mach_port_t *port);
267 
268 /*
269  * Join work interval via Mach send right
270  *
271  * Does NOT consume Mach send right, must deallocate with mach_port_deallocate after using
272  * It's safe to deallocate the right after joining, the thread will stay joined
273  *
274  * Can be sent to clients via xpc_dictionary_copy_mach_send, and similar
275  *
276  * Supersedes automatic thread group management via vouchers
277  *
278  * If the underlying work interval object is terminated then this may return ENOENT
279  * <rdar://problem/31819320>
280  */
281 int     work_interval_join_port(mach_port_t port);
282 
283 /*
284  * Leave the current thread's work interval
285  */
286 int     work_interval_leave(void);
287 
288 #endif /* !KERNEL */
289 
290 
291 #if PRIVATE
292 
293 /* Private interface between Libsyscall and xnu */
294 #define WORK_INTERVAL_OPERATION_CREATE  0x00000001      /* deprecated */
295 #define WORK_INTERVAL_OPERATION_DESTROY 0x00000002      /* arg is NULL */
296 #define WORK_INTERVAL_OPERATION_NOTIFY  0x00000003      /* arg is a work_interval_notification_t */
297 #define WORK_INTERVAL_OPERATION_CREATE2 0x00000004      /* arg is a work_interval_create_params */
298 #define WORK_INTERVAL_OPERATION_JOIN    0x00000005      /* arg is a port_name */
299 #define WORK_INTERVAL_OPERATION_GET_FLAGS 0x00000009    /* arg is a port name */
300 #define WORK_INTERVAL_OPERATION_SET_NAME  0x0000000a    /* arg is name string (char[WORK_INTERVAL_NAME_MAX])*/
301 #define WORK_INTERVAL_OPERATION_SET_WORKLOAD_ID  0x0000000b    /* arg is a work_interval_workload_id_params */
302 #define WORK_INTERVAL_NAME_MAX  32
303 #define WORK_INTERVAL_WORKLOAD_ID_NAME_MAX  64
304 
305 /* Flags passed in work_interval_workload_id_params wlidp_flags field */
306 #define WORK_INTERVAL_WORKLOAD_ID_HAS_ID                (1u << 0)
307 #define WORK_INTERVAL_WORKLOAD_ID_RT_ALLOWED            (1u << 1)
308 #define WORK_INTERVAL_WORKLOAD_ID_RT_CRITICAL           (1u << 2)
309 
310 /* Flags allowed to be passed in from userspace as part of kern_work_interval_set_workload_id() */
311 #define WORK_INTERVAL_SET_WORKLOAD_ID_FLAGS_MASK        (WORK_INTERVAL_WORKLOAD_ID_RT_CRITICAL | WORK_INTERVAL_WORKLOAD_ID_RT_ALLOWED)
312 
313 #ifdef XNU_KERNEL_PRIVATE
314 
315 /* Marker that workinterval was joined before workload ID was set */
316 #define WORK_INTERVAL_WORKLOAD_ID_ALREADY_JOINED        (1u << 31)
317 /* Work interval is allowed to provide complexity values per frame as part of {start, update, finish} calls */
318 #define WORK_INTERVAL_WORKLOAD_ID_COMPLEXITY_ALLOWED    (1u << 30)
319 
320 #endif /* XNU_KERNEL_PRIVATE */
321 
322 struct work_interval_notification {
323 	uint64_t        start;
324 	uint64_t        finish;
325 	uint64_t        deadline;
326 	uint64_t        next_start;
327 	uint32_t        notify_flags;
328 	uint32_t        create_flags;
329 };
330 typedef struct work_interval_notification *work_interval_notification_t;
331 
332 struct work_interval_create_params {
333 	uint64_t        wicp_id;          /* in/out param */
334 	mach_port_name_t wicp_port;       /* in/out param */
335 	uint32_t        wicp_create_flags;
336 };
337 
338 struct work_interval_workload_id_params {
339 	uint32_t        wlidp_flags;            /* in/out param */
340 	uint32_t        wlidp_wicreate_flags;   /* in/out param */
341 	uint64_t        wlidp_name;             /* in param (pointer to char[WORK_INTERVAL_WORKLOAD_ID_NAME_MAX])*/
342 	uint64_t        wlidp_syscall_mask[2];  /* out param (needs to fit MACH_TRAP_TABLE_COUNT + nsysent bits) */
343 };
344 
345 
346 int     __work_interval_ctl(uint32_t operation, uint64_t work_interval_id, void *arg, size_t len);
347 
348 #endif /* PRIVATE */
349 
350 __END_DECLS
351 
352 #endif /* _SYS_WORK_INTERVAL_H */
353