xref: /xnu-8796.121.2/osfmk/kern/sched_clutch.h (revision c54f35ca767986246321eb901baf8f5ff7923f6a)
1 /*
2  * Copyright (c) 2018 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 _KERN_SCHED_CLUTCH_H_
30 #define _KERN_SCHED_CLUTCH_H_
31 
32 #include <kern/sched.h>
33 #include <machine/atomic.h>
34 #include <kern/priority_queue.h>
35 #include <kern/thread_group.h>
36 #include <kern/bits.h>
37 
38 #if CONFIG_SCHED_CLUTCH
39 
40 /*
41  * For the current implementation, bound threads are not managed
42  * in the clutch hierarchy. This helper macro is used to indicate
43  * if the thread should be in the hierarchy.
44  */
45 #define SCHED_CLUTCH_THREAD_ELIGIBLE(thread)    ((thread->bound_processor) == PROCESSOR_NULL)
46 
47 #if CONFIG_SCHED_EDGE
48 #define SCHED_CLUTCH_THREAD_CLUSTER_BOUND(thread)       (thread->th_bound_cluster_id != THREAD_BOUND_CLUSTER_NONE)
49 #define SCHED_CLUTCH_THREAD_CLUSTER_BOUND_SOFT(thread)  ((thread->sched_flags & TH_SFLAG_BOUND_SOFT) != 0)
50 
51 #else /* CONFIG_SCHED_EDGE */
52 #define SCHED_CLUTCH_THREAD_CLUSTER_BOUND(thread)       (0)
53 #define SCHED_CLUTCH_THREAD_CLUSTER_BOUND_SOFT(thread)  (0)
54 #endif /* CONFIG_SCHED_EDGE */
55 
56 /*
57  * Clutch Bucket Runqueue Structure.
58  */
59 struct sched_clutch_bucket_runq {
60 	int                     scbrq_highq;
61 	int                     scbrq_count;
62 	bitmap_t                scbrq_bitmap[BITMAP_LEN(NRQS_MAX)];
63 	circle_queue_head_t     scbrq_queues[NRQS_MAX];
64 };
65 typedef struct sched_clutch_bucket_runq *sched_clutch_bucket_runq_t;
66 
67 /*
68  *
69  * Clutch hierarchy locking protocol
70  *
71  * The scheduler clutch hierarchy is protected by a combination of
72  * atomics and pset lock.
73  * - All fields protected by the pset lock are annotated with (P)
74  * - All fields updated using atomics are annotated with (A)
75  * - All fields that are unprotected and are not updated after
76  *   initialization are annotated with (I)
77  */
78 
79 /*
80  * struct sched_clutch_root_bucket
81  *
82  * A clutch_root_bucket represents all threads across all thread groups
83  * that are in the same scheduler bucket (FG/IN/...). The clutch_root_bucket
84  * is selected for execution by the root level bucket selection algorithm
85  * which bases the decision on the clutch_root_bucket's deadline (EDF). The
86  * deadline for a root bucket is calculated based on its runnable timestamp
87  * and the worst-case-execution-latency values specied in sched_clutch_root_bucket_wcel[]
88  */
89 struct sched_clutch_root_bucket {
90 	/* (I) sched bucket represented by this root bucket */
91 	uint8_t                         scrb_bucket;
92 	/* (I) Indicates the root bucket represents cluster bound threads */
93 	bool                            scrb_bound;
94 	/* (P) Indicates if the root bucket is in starvation avoidance mode */
95 	bool                            scrb_starvation_avoidance;
96 
97 	union {
98 		/* (P) priority queue for all unbound clutch buckets in this sched bucket */
99 		struct sched_clutch_bucket_runq scrb_clutch_buckets;
100 		/* (P) Runqueue for all bound threads part of this root bucket */
101 		struct run_queue                scrb_bound_thread_runq;
102 	};
103 	/* (P) priority queue entry to use for enqueueing root bucket into root prioq */
104 	struct priority_queue_entry_deadline scrb_pqlink;
105 	/* (P) warped deadline for root bucket */
106 	uint64_t                        scrb_warped_deadline;
107 	/* (P) warp remaining for root bucket */
108 	uint64_t                        scrb_warp_remaining;
109 	/* (P) timestamp for the start of the starvation avoidance window */
110 	uint64_t                        scrb_starvation_ts;
111 };
112 typedef struct sched_clutch_root_bucket *sched_clutch_root_bucket_t;
113 
114 /*
115  * struct sched_clutch_root
116  *
117  * A clutch_root represents the root of the hierarchy. It maintains a
118  * priority queue of all runnable root buckets. The clutch_root also
119  * maintains the information about the last clutch_root_bucket scheduled
120  * in order to implement bucket level quantum. The bucket level quantums
121  * allow low priority buckets to get a "fair" chance of using the CPU even
122  * if they contain a bunch of short executing threads. The bucket quantums
123  * are configured using sched_clutch_root_bucket_quantum[]
124  */
125 struct sched_clutch_root {
126 	/* (P) root level priority; represents the highest runnable thread in the hierarchy */
127 	int16_t                         scr_priority;
128 	/* (P) total number of runnable threads in the hierarchy */
129 	uint16_t                        scr_thr_count;
130 	/* (P) root level urgency; represents the urgency of the whole hierarchy for pre-emption purposes */
131 	int16_t                         scr_urgency;
132 	/* (P) runnable shared resource load enqueued in this cluster/root hierarchy */
133 	uint16_t                        scr_shared_rsrc_load_runnable[CLUSTER_SHARED_RSRC_TYPE_COUNT];
134 
135 	uint32_t                        scr_cluster_id;
136 	/* (I) processor set this hierarchy belongs to */
137 	processor_set_t                 scr_pset;
138 	/*
139 	 * (P) list of all runnable clutch buckets across the system;
140 	 * allows easy iteration in the sched tick based timesharing code
141 	 */
142 	queue_head_t                    scr_clutch_buckets;
143 
144 	/*
145 	 * (P) priority queue of all runnable foreign buckets in this hierarchy;
146 	 * used for tracking thread groups which need to be migrated when
147 	 * psets are available or rebalancing threads on CPU idle.
148 	 */
149 	struct priority_queue_sched_max scr_foreign_buckets;
150 
151 	/* Root level bucket management */
152 
153 	/* (P) bitmap of all runnable unbounded root buckets */
154 	bitmap_t                        scr_unbound_runnable_bitmap[BITMAP_LEN(TH_BUCKET_SCHED_MAX)];
155 	/* (P) bitmap of all runnable unbounded root buckets which have warps remaining */
156 	bitmap_t                        scr_unbound_warp_available[BITMAP_LEN(TH_BUCKET_SCHED_MAX)];
157 	/* (P) bitmap of all runnable bounded root buckets */
158 	bitmap_t                        scr_bound_runnable_bitmap[BITMAP_LEN(TH_BUCKET_SCHED_MAX)];
159 	/* (P) bitmap of all runnable bounded root buckets which have warps remaining */
160 	bitmap_t                        scr_bound_warp_available[BITMAP_LEN(TH_BUCKET_SCHED_MAX)];
161 
162 	/* (P) priority queue of all runnable unbounded root buckets in deadline order */
163 	struct priority_queue_deadline_min scr_unbound_root_buckets;
164 	/* (P) priority queue of all bounded root buckets in deadline order */
165 	struct priority_queue_deadline_min scr_bound_root_buckets;
166 
167 	/* (P) cumulative run counts at each bucket for load average calculation */
168 	uint16_t _Atomic                scr_cumulative_run_count[TH_BUCKET_SCHED_MAX];
169 
170 	/* (P) storage for all unbound clutch_root_buckets */
171 	struct sched_clutch_root_bucket scr_unbound_buckets[TH_BUCKET_SCHED_MAX];
172 	/* (P) storage for all bound clutch_root_buckets */
173 	struct sched_clutch_root_bucket scr_bound_buckets[TH_BUCKET_SCHED_MAX];
174 };
175 typedef struct sched_clutch_root *sched_clutch_root_t;
176 
177 /* forward declaration for sched_clutch */
178 struct sched_clutch;
179 
180 /*
181  * sched_clutch_bucket_cpu_data_t
182  *
183  * Used for maintaining clutch bucket used and blocked time. The
184  * values are used for calculating the interactivity score for the
185  * clutch bucket.
186  *
187  * Since the CPU used/blocked calculation uses wide atomics, the data
188  * types used are different based on the platform.
189  */
190 
191 #if __LP64__
192 
193 #define CLUTCH_CPU_DATA_MAX             (UINT64_MAX)
194 typedef uint64_t                        clutch_cpu_data_t;
195 typedef unsigned __int128               clutch_cpu_data_wide_t;
196 
197 #else /* __LP64__ */
198 
199 #define CLUTCH_CPU_DATA_MAX             (UINT32_MAX)
200 typedef uint32_t                        clutch_cpu_data_t;
201 typedef uint64_t                        clutch_cpu_data_wide_t;
202 
203 #endif /* __LP64__ */
204 
205 typedef union sched_clutch_bucket_cpu_data {
206 	struct {
207 		/* Clutch bucket CPU used across all threads */
208 		clutch_cpu_data_t       scbcd_cpu_used;
209 		/* Clutch bucket voluntary blocked time */
210 		clutch_cpu_data_t       scbcd_cpu_blocked;
211 	} cpu_data;
212 	clutch_cpu_data_wide_t          scbcd_cpu_data_packed;
213 } sched_clutch_bucket_cpu_data_t;
214 
215 /*
216  * struct sched_clutch_bucket
217  *
218  * A sched_clutch_bucket represents the set of threads for a thread
219  * group at a particular scheduling bucket in a specific cluster.
220  * It maintains information about the CPU usage & blocking behavior
221  * of all threads part of the clutch_bucket. It inherits the timeshare
222  * values from the clutch_bucket_group for decay and timesharing among
223  * threads in the clutch.
224  *
225  * Since the clutch bucket is a per thread group per-QoS entity it is
226  * important to keep its size small and the structure well aligned.
227  */
228 struct sched_clutch_bucket {
229 #if CONFIG_SCHED_EDGE
230 	/* (P) flag to indicate if the bucket is a foreign bucket */
231 	bool                            scb_foreign;
232 #endif /* CONFIG_SCHED_EDGE */
233 	/* (I) bucket for the clutch_bucket */
234 	uint8_t                         scb_bucket;
235 	/* (P) priority of the clutch bucket */
236 	uint8_t                         scb_priority;
237 	/* (P) number of threads in this clutch_bucket; should match runq.count */
238 	uint16_t                        scb_thr_count;
239 
240 	/* Pointer to the clutch bucket group this clutch bucket belongs to */
241 	struct sched_clutch_bucket_group *scb_group;
242 	/* (A) pointer to the root of the hierarchy this bucket is in */
243 	struct sched_clutch_root        *scb_root;
244 	/* (P) priority queue of threads based on their promoted/base priority */
245 	struct priority_queue_sched_max scb_clutchpri_prioq;
246 	/* (P) runq of threads in clutch_bucket */
247 	struct priority_queue_sched_stable_max scb_thread_runq;
248 
249 	/* (P) linkage for all clutch_buckets in a root bucket; used for tick operations */
250 	queue_chain_t                   scb_listlink;
251 	/* (P) linkage for clutch_bucket in root_bucket runqueue */
252 	queue_chain_t                   scb_runqlink;
253 	/* (P) queue of threads for timesharing purposes */
254 	queue_head_t                    scb_thread_timeshare_queue;
255 #if CONFIG_SCHED_EDGE
256 	/* (P) linkage for all "foreign" clutch buckets in the root clutch */
257 	struct priority_queue_entry_sched     scb_foreignlink;
258 #endif /* CONFIG_SCHED_EDGE */
259 };
260 typedef struct sched_clutch_bucket *sched_clutch_bucket_t;
261 
262 /*
263  * sched_clutch_counter_time_t
264  *
265  * Holds thread counts and a timestamp (typically for a clutch bucket group).
266  * Used to allow atomic updates to these fields.
267  */
268 typedef union sched_clutch_counter_time {
269 	struct {
270 		uint64_t                scct_count;
271 		uint64_t                scct_timestamp;
272 	};
273 #if __LP64__
274 	unsigned __int128               scct_packed;
275 #endif /* __LP64__ */
276 } __attribute__((aligned(16))) sched_clutch_counter_time_t;
277 
278 /*
279  * struct sched_clutch_bucket_group
280  *
281  * It represents all the threads for a thread group at a particular
282  * QoS/Scheduling bucket. This structure also maintains the timesharing
283  * properties that are used for decay calculation for all threads in the
284  * thread group at the specific scheduling bucket.
285  */
286 struct sched_clutch_bucket_group {
287 	/* (I) bucket for the clutch_bucket_group */
288 	uint8_t                         scbg_bucket;
289 	/* (A) sched tick when the clutch bucket group load/shifts were updated */
290 	uint32_t _Atomic                scbg_timeshare_tick;
291 	/* (A) priority shifts for threads in the clutch_bucket_group */
292 	uint32_t _Atomic                scbg_pri_shift;
293 	/* (A) preferred cluster ID for clutch bucket */
294 	uint32_t _Atomic                scbg_preferred_cluster;
295 	/* (A) cluster ID for AMP rebalancing */
296 	uint32_t                        scbg_amp_rebalance_last_chosen;
297 	/* (I) clutch to which this clutch bucket_group belongs */
298 	struct sched_clutch             *scbg_clutch;
299 #if !__LP64__
300 	/* Lock for synchronizing updates to blocked data (only on platforms without 128-atomics) */
301 	lck_spin_t                      scbg_stats_lock;
302 #endif /* !__LP64__ */
303 	/* (A/L depending on arch) holds blcked timestamp and runnable/running count */
304 	sched_clutch_counter_time_t     scbg_blocked_data;
305 	/* (P/A depending on scheduler) holds pending timestamp and thread count */
306 	sched_clutch_counter_time_t     scbg_pending_data;
307 	/* (P/A depending on scheduler) holds interactivity timestamp and score */
308 	sched_clutch_counter_time_t     scbg_interactivity_data;
309 	/* (A) CPU usage information for the clutch bucket group */
310 	sched_clutch_bucket_cpu_data_t  scbg_cpu_data;
311 	/* Storage for all clutch buckets for a thread group at scbg_bucket */
312 	struct sched_clutch_bucket      *scbg_clutch_buckets;
313 };
314 typedef struct sched_clutch_bucket_group *sched_clutch_bucket_group_t;
315 
316 
317 /*
318  * struct sched_clutch
319  *
320  * A sched_clutch is a 1:1 mapping to a thread group. It maintains the
321  * storage for all clutch buckets for this thread group and some properties
322  * of the thread group (such as flags etc.)
323  */
324 struct sched_clutch {
325 	/*
326 	 * (A) number of runnable threads in sched_clutch; needs to be atomic
327 	 * to support cross cluster sched_clutch migrations.
328 	 */
329 	uint16_t _Atomic                sc_thr_count;
330 	/*
331 	 * Grouping specific parameters. Currently the implementation only
332 	 * supports thread_group based grouping.
333 	 */
334 	union {
335 		/* (I) Pointer to thread group */
336 		struct thread_group     *sc_tg;
337 	};
338 	/* (I) storage for all clutch_buckets for this clutch */
339 	struct sched_clutch_bucket_group sc_clutch_groups[TH_BUCKET_SCHED_MAX];
340 };
341 typedef struct sched_clutch *sched_clutch_t;
342 
343 
344 /* Clutch lifecycle management */
345 void sched_clutch_init_with_thread_group(sched_clutch_t, struct thread_group *);
346 void sched_clutch_destroy(sched_clutch_t);
347 
348 /* Clutch thread membership management */
349 void sched_clutch_thread_clutch_update(thread_t, sched_clutch_t, sched_clutch_t);
350 uint32_t sched_edge_thread_preferred_cluster(thread_t);
351 
352 /* Clutch timesharing stats management */
353 uint32_t sched_clutch_thread_run_bucket_incr(thread_t, sched_bucket_t);
354 uint32_t sched_clutch_thread_run_bucket_decr(thread_t, sched_bucket_t);
355 void sched_clutch_cpu_usage_update(thread_t, uint64_t);
356 uint32_t sched_clutch_thread_pri_shift(thread_t, sched_bucket_t);
357 
358 /* Clutch properties accessors */
359 uint32_t sched_clutch_root_count(sched_clutch_root_t);
360 
361 /* Grouping specific external routines */
362 extern sched_clutch_t sched_clutch_for_thread(thread_t);
363 extern sched_clutch_t sched_clutch_for_thread_group(struct thread_group *);
364 
365 #if CONFIG_SCHED_EDGE
366 
367 /*
368  * Getter and Setter for Edge configuration. Used by CLPC to affect thread migration behavior.
369  */
370 void sched_edge_matrix_get(sched_clutch_edge *edge_matrix, bool *edge_request_bitmap, uint64_t flags, uint64_t matrix_order);
371 void sched_edge_matrix_set(sched_clutch_edge *edge_matrix, bool *edge_changes_bitmap, uint64_t flags, uint64_t matrix_order);
372 void sched_edge_tg_preferred_cluster_change(struct thread_group *tg, uint32_t *tg_bucket_preferred_cluster, sched_perfcontrol_preferred_cluster_options_t options);
373 
374 uint16_t sched_edge_cluster_cumulative_count(sched_clutch_root_t root_clutch, sched_bucket_t bucket);
375 uint16_t sched_edge_shared_rsrc_runnable_load(sched_clutch_root_t root_clutch, cluster_shared_rsrc_type_t load_type);
376 
377 #endif /* CONFIG_SCHED_EDGE */
378 
379 #endif /* CONFIG_SCHED_CLUTCH */
380 
381 #endif /* _KERN_SCHED_CLUTCH_H_ */
382