1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ 22 /* All Rights Reserved */ 23 24 25 /* 26 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 27 * Use is subject to license terms. 28 */ 29 30 #ifndef _SYS_CALLO_H 31 #define _SYS_CALLO_H 32 33 #include <sys/t_lock.h> 34 #include <sys/taskq.h> 35 #include <sys/lgrp.h> 36 #include <sys/processor.h> 37 #include <sys/cyclic.h> 38 #include <sys/kstat.h> 39 #include <sys/systm.h> 40 41 #ifdef __cplusplus 42 extern "C" { 43 #endif 44 45 #ifdef _KERNEL 46 47 typedef struct callout_list callout_list_t; 48 49 /* 50 * The callout mechanism provides general-purpose event scheduling: 51 * an arbitrary function is called in a specified amount of time. 52 * The expiration time for a callout is kept in its callout list 53 * structure. 54 */ 55 typedef struct callout { 56 struct callout *c_idnext; /* next in ID hash, or on freelist */ 57 struct callout *c_idprev; /* prev in ID hash */ 58 struct callout *c_clnext; /* next in callout list */ 59 struct callout *c_clprev; /* prev in callout list */ 60 callout_id_t c_xid; /* extended callout ID; see below */ 61 callout_list_t *c_list; /* callout list */ 62 void (*c_func)(void *); /* function to call */ 63 void *c_arg; /* argument to function */ 64 kthread_t *c_executor; /* executing thread */ 65 kcondvar_t c_done; /* signal callout completion */ 66 ushort_t c_waiting; /* untimeout waiting flag */ 67 } callout_t; 68 69 /* 70 * The callout ID (callout_id_t) uniquely identifies a callout. The callout 71 * ID is always 64 bits internally. The lower 32 bits contain an ID value. 72 * The upper 32 bits contain a generation number and flags. When the ID value 73 * wraps the generation number is incremented during ID generation. This 74 * protects callers from ID collisions that can happen as a result of the wrap. 75 * 76 * The kernel internal interface, timeout_generic(), always returns a 77 * callout_id_t. But the legacy interfaces, timeout() and realtime_timeout() 78 * return a timeout_id_t. On a 64-bit system, timeout_id_t is also 64 bits. 79 * So, the full 64-bit ID (sans the flags) can be returned. However, on 32-bit 80 * systems, timeout_id_t is 32 bits. So, only the lower 32 bits can be 81 * returned. In such cases, a default generation number of 0 is assigned to 82 * the legacy IDs. 83 * 84 * The lower 32-bit ID space is partitioned into two spaces - one for 32-bit 85 * IDs and the other for 64-bit IDs. The 32-bit ID space is further divided 86 * into two spaces - one for short-term callouts and one for long-term. 87 * 88 * Here is the bit layout for the callout ID: 89 * 90 * 63 62 ... 32 31 30 29 .. X+1 X ... 1 0 91 * ---------------------------------------------------------------- 92 * | Exec | Generation | Long | Counter | ID bits | Table | Type | 93 * | | number | term | High | | number | | 94 * ---------------------------------------------------------------- 95 * 96 * Exec(uting): 97 * This is the executing bit which is only set in the extended callout 98 * ID. This bit indicates that the callout handler is currently being 99 * executed. 100 * 101 * Generation number: 102 * This is the generation part of the ID. 103 * 104 * Long term: 105 * This bit indicates whether this is a short-term or a long-term callout. 106 * The long-term bit exists to address the problem of callout ID collision 107 * on 32-bit systems. This is an issue because the system typically 108 * generates a large number of timeout() requests, which means that callout 109 * IDs eventually get recycled. Most timeouts are very short-lived, so that 110 * ID recycling isn't a problem; but there are a handful of timeouts which 111 * are sufficiently long-lived to see their own IDs reused. We use the 112 * long-term bit to partition the ID namespace into pieces; the short-term 113 * space gets all the heavy traffic and can wrap frequently (i.e., on the 114 * order of a day) with no ill effects; the long-term space gets very little 115 * traffic and thus never wraps. That said, we need to future proof callouts 116 * in case 32-bit systems grow in size and are able to consume callout IDs 117 * at faster rates. So, we should make all the kernel clients that use 118 * callouts to use the internal interface so that they can use IDs outside 119 * of the legacy space with a proper generation number. 120 * 121 * Counter High + ID counter bits: 122 * These bits represent the actual ID bits in the callout ID. 123 * The highest bit of the running counter is always set; this ensures that 124 * the callout ID is always non-zero, thus eliminating the need for an 125 * explicit wrap-around test during ID generation. 126 * 127 * Table number: 128 * These bits carry the table number for the callout table where the callout 129 * is queued. Each CPU has its own callout table. So, the callout tables are 130 * numbered from 0 - (max_ncpus - 1). Because max_ncpus is different on 131 * different systems, the actual number of table number bits will vary 132 * accordingly. And so will the ID counter bits. 133 * 134 * Type: 135 * This bit represents the callout (table) type. Each CPU has one realtime 136 * and one normal callout table. 137 */ 138 #define CALLOUT_EXECUTING 0x8000000000000000ULL 139 #define CALLOUT_ID_MASK ~(CALLOUT_EXECUTING) 140 #define CALLOUT_GENERATION_LOW 0x100000000ULL 141 #define CALLOUT_LONGTERM 0x80000000 142 #define CALLOUT_COUNTER_HIGH 0x40000000 143 #define CALLOUT_TYPE_BITS 1 144 #define CALLOUT_NTYPES (1 << CALLOUT_TYPE_BITS) 145 #define CALLOUT_TYPE_MASK (CALLOUT_NTYPES - 1) 146 #define CALLOUT_COUNTER_SHIFT callout_table_bits 147 #define CALLOUT_TABLE(t, f) (((f) << CALLOUT_TYPE_BITS) | (t)) 148 #define CALLOUT_TABLE_NUM(ct) ((ct) - callout_table) 149 #define CALLOUT_TABLE_TYPE(ct) (CALLOUT_TABLE_NUM(ct) & CALLOUT_TYPE_MASK) 150 #define CALLOUT_TABLE_SEQID(ct) (CALLOUT_TABLE_NUM(ct) >> CALLOUT_TYPE_BITS) 151 152 /* 153 * We assume that during any period of CALLOUT_LONGTERM_TICKS ticks, at most 154 * (CALLOUT_COUNTER_HIGH / callout_counter_low) callouts will be generated. 155 */ 156 #define CALLOUT_LONGTERM_TICKS 0x4000UL 157 #define CALLOUT_BUCKET_SHIFT 9 158 #define CALLOUT_BUCKETS (1 << CALLOUT_BUCKET_SHIFT) 159 #define CALLOUT_BUCKET_MASK (CALLOUT_BUCKETS - 1) 160 #define CALLOUT_HASH(x) ((x) & CALLOUT_BUCKET_MASK) 161 #define CALLOUT_IDHASH(x) CALLOUT_HASH((x) >> CALLOUT_COUNTER_SHIFT) 162 /* 163 * The multiply by 0 and 1 below are cosmetic. Just to align things better 164 * and make it more readable. The multiplications will be done at compile 165 * time. 166 */ 167 #define CALLOUT_CLHASH(x) \ 168 CALLOUT_HASH( \ 169 ((x)>>(CALLOUT_BUCKET_SHIFT*0)) ^ \ 170 ((x)>>(CALLOUT_BUCKET_SHIFT*1)) ^ \ 171 ((x)>>(CALLOUT_BUCKET_SHIFT*2)) ^ \ 172 ((x)>>(CALLOUT_BUCKET_SHIFT*3))) 173 174 #define CALLOUT_ID_TO_TABLE(id) ((id) & callout_table_mask) 175 176 #define CALLOUT_SHORT_ID(table) \ 177 ((callout_id_t)(table) | CALLOUT_COUNTER_HIGH) 178 #define CALLOUT_LONG_ID(table) \ 179 (CALLOUT_SHORT_ID(table) | CALLOUT_LONGTERM) 180 181 #define CALLOUT_THREADS 2 /* keep it simple for now */ 182 183 #define CALLOUT_REALTIME 0 /* realtime callout type */ 184 #define CALLOUT_NORMAL 1 /* normal callout type */ 185 186 /* 187 * callout_t's are cache-aligned structures allocated from kmem caches. One kmem 188 * cache is created per lgrp and is shared by all CPUs in that lgrp. Benefits: 189 * - cache pages are mapped only in the TLBs of the CPUs of the lgrp 190 * - data in cache pages is present only in those CPU caches 191 * - memory access performance improves with locality-awareness in kmem 192 * 193 * The following structure is used to manage per-lgroup kmem caches. 194 * 195 * NOTE: Free callout_t's go to a callout table's freelist. CPUs map to callout 196 * tables via their sequence IDs, not CPU IDs. DR operations can cause a 197 * free list to have callouts from multiple lgrp caches. This takes away some 198 * performance, but is no worse than if we did not use lgrp caches at all. 199 */ 200 typedef struct callout_cache { 201 struct callout_cache *cc_next; /* link in the global list */ 202 lgrp_handle_t cc_hand; /* lgroup handle */ 203 kmem_cache_t *cc_cache; /* kmem cache pointer */ 204 kmem_cache_t *cc_lcache; /* kmem cache pointer */ 205 } callout_cache_t; 206 207 /* 208 * The callout hash structure is used for queueing both callouts and 209 * callout lists. That is why the fields are declared as void *. 210 */ 211 typedef struct callout_hash { 212 void *ch_head; 213 void *ch_tail; 214 } callout_hash_t; 215 216 struct callout_list { 217 callout_list_t *cl_next; /* next in clhash */ 218 callout_list_t *cl_prev; /* prev in clhash */ 219 hrtime_t cl_expiration; /* expiration for callouts in list */ 220 callout_hash_t cl_callouts; /* list of callouts */ 221 int cl_flags; /* callout flags */ 222 }; 223 224 /* 225 * Per-callout table kstats. 226 * 227 * CALLOUT_TIMEOUTS 228 * Callouts created since boot. 229 * CALLOUT_TIMEOUTS_PENDING 230 * Number of outstanding callouts. 231 * CALLOUT_UNTIMEOUTS_UNEXPIRED 232 * Number of cancelled callouts that have not expired. 233 * CALLOUT_UNTIMEOUTS_EXECUTING 234 * Number of cancelled callouts that were executing at the time of 235 * cancellation. 236 * CALLOUT_UNTIMEOUTS_EXPIRED 237 * Number of cancelled callouts that had already expired at the time 238 * of cancellations. 239 * CALLOUT_EXPIRATIONS 240 * Number of callouts that expired. 241 * CALLOUT_ALLOCATIONS 242 * Number of callout structures allocated. 243 */ 244 typedef enum callout_stat_type { 245 CALLOUT_TIMEOUTS, 246 CALLOUT_TIMEOUTS_PENDING, 247 CALLOUT_UNTIMEOUTS_UNEXPIRED, 248 CALLOUT_UNTIMEOUTS_EXECUTING, 249 CALLOUT_UNTIMEOUTS_EXPIRED, 250 CALLOUT_EXPIRATIONS, 251 CALLOUT_ALLOCATIONS, 252 CALLOUT_NUM_STATS 253 } callout_stat_type_t; 254 255 /* 256 * Callout flags: 257 * 258 * CALLOUT_FLAG_ROUNDUP 259 * Roundup the expiration time to the next resolution boundary. 260 * If this flag is not specified, the expiration time is rounded down. 261 * CALLOUT_FLAG_ABSOLUTE 262 * Normally, the expiration passed to the timeout API functions is an 263 * expiration interval. If this flag is specified, then it is 264 * interpreted as the expiration time itself. 265 * CALLOUT_FLAG_HRESTIME 266 * Normally, callouts are not affected by changes to system time 267 * (hrestime). This flag is used to create a callout that is affected 268 * by system time. If system time changes, these timers must be 269 * handled in a special way (see callout.c). These are used by condition 270 * variables and LWP timers that need this behavior. 271 * CALLOUT_FLAG_32BIT 272 * Legacy interfaces timeout() and realtime_timeout() pass this flag 273 * to timeout_generic() to indicate that a 32-bit ID should be allocated. 274 */ 275 #define CALLOUT_FLAG_ROUNDUP 0x1 276 #define CALLOUT_FLAG_ABSOLUTE 0x2 277 #define CALLOUT_FLAG_HRESTIME 0x4 278 #define CALLOUT_FLAG_32BIT 0x8 279 280 #define CALLOUT_LIST_FLAGS (CALLOUT_FLAG_ABSOLUTE | CALLOUT_FLAG_HRESTIME) 281 /* 282 * On 32-bit systems, the legacy interfaces, timeout() and realtime_timeout(), 283 * must pass CALLOUT_FLAG_32BIT to timeout_generic() so that a 32-bit ID 284 * can be generated. 285 */ 286 #ifdef _LP64 287 #define CALLOUT_LEGACY 0 288 #else 289 #define CALLOUT_LEGACY CALLOUT_FLAG_32BIT 290 #endif 291 292 /* 293 * All of the state information associated with a callout table. 294 * The fields are ordered with cache performance in mind. 295 */ 296 typedef struct callout_table { 297 kmutex_t ct_mutex; /* protects all callout state */ 298 callout_t *ct_free; /* free callout structures */ 299 callout_list_t *ct_lfree; /* free callout list structures */ 300 callout_id_t ct_short_id; /* most recently issued short-term ID */ 301 callout_id_t ct_long_id; /* most recently issued long-term ID */ 302 callout_hash_t *ct_idhash; /* ID hash chains */ 303 callout_hash_t *ct_clhash; /* callout list hash */ 304 kstat_named_t *ct_kstat_data; /* callout kstat data */ 305 306 uint_t ct_type; /* callout table type */ 307 uint_t ct_suspend; /* suspend count */ 308 cyclic_id_t ct_cyclic; /* cyclic for this table */ 309 hrtime_t *ct_heap; /* callout expiration heap */ 310 ulong_t ct_heap_num; /* occupied slots in the heap */ 311 ulong_t ct_heap_max; /* end of the heap */ 312 kmem_cache_t *ct_cache; /* callout kmem cache */ 313 kmem_cache_t *ct_lcache; /* callout list kmem cache */ 314 callout_id_t ct_gen_id; /* generation based ID */ 315 316 callout_hash_t ct_expired; /* list of expired callout lists */ 317 taskq_t *ct_taskq; /* taskq to execute normal callouts */ 318 kstat_t *ct_kstats; /* callout kstats */ 319 #ifdef _LP64 320 ulong_t ct_pad[4]; /* cache alignment */ 321 #else 322 ulong_t ct_pad[7]; /* cache alignment */ 323 #endif 324 } callout_table_t; 325 326 /* 327 * Short hand definitions for the callout kstats. 328 */ 329 #define ct_timeouts \ 330 ct_kstat_data[CALLOUT_TIMEOUTS].value.ui64 331 #define ct_timeouts_pending \ 332 ct_kstat_data[CALLOUT_TIMEOUTS_PENDING].value.ui64 333 #define ct_untimeouts_unexpired \ 334 ct_kstat_data[CALLOUT_UNTIMEOUTS_UNEXPIRED].value.ui64 335 #define ct_untimeouts_executing \ 336 ct_kstat_data[CALLOUT_UNTIMEOUTS_EXECUTING].value.ui64 337 #define ct_untimeouts_expired \ 338 ct_kstat_data[CALLOUT_UNTIMEOUTS_EXPIRED].value.ui64 339 #define ct_expirations \ 340 ct_kstat_data[CALLOUT_EXPIRATIONS].value.ui64 341 #define ct_allocations \ 342 ct_kstat_data[CALLOUT_ALLOCATIONS].value.ui64 343 344 #define CALLOUT_CHUNK 128 345 346 #define CALLOUT_HEAP_PARENT(index) (((index) - 1) >> 1) 347 #define CALLOUT_HEAP_RIGHT(index) (((index) + 1) << 1) 348 #define CALLOUT_HEAP_LEFT(index) ((((index) + 1) << 1) - 1) 349 350 #define CALLOUT_CYCLIC_HANDLER(t) \ 351 ((t == CALLOUT_REALTIME) ? callout_realtime : callout_normal) 352 353 /* 354 * We define a blanket minimum resolution for callouts of 1 millisecond. 355 * 1 millisecond is a safe value as it is already supported when the clock 356 * resolution is set to high. 357 */ 358 #define CALLOUT_MIN_RESOLUTION 1000000ULL 359 #define CALLOUT_TCP_RESOLUTION 10000000ULL 360 361 #define CALLOUT_ALIGN 64 /* cache line size */ 362 363 #ifdef _LP64 364 #define CALLOUT_MAX_TICKS NSEC_TO_TICK(CY_INFINITY); 365 #else 366 #define CALLOUT_MAX_TICKS LONG_MAX 367 #endif 368 369 extern void callout_init(void); 370 extern void membar_sync(void); 371 extern void callout_cpu_online(cpu_t *); 372 extern void callout_cpu_offline(cpu_t *); 373 extern void callout_hrestime(void); 374 375 #endif 376 377 #ifdef __cplusplus 378 } 379 #endif 380 381 #endif /* _SYS_CALLO_H */ 382