usr/src/uts/common/sys/callo.h

   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 } callout_t;
  65
  66 /*
  67  * The callout ID (callout_id_t) uniquely identifies a callout. The callout
  68  * ID is always 64 bits internally. The lower 32 bits contain an ID value.
  69  * The upper 32 bits contain a generation number and flags. When the ID value
  70  * wraps the generation number is incremented during ID generation. This
  71  * protects callers from ID collisions that can happen as a result of the wrap.
  72  *
  73  * The kernel internal interface, timeout_generic(), always returns a
  74  * callout_id_t. But the legacy interfaces, timeout() and realtime_timeout()
  75  * return a timeout_id_t. On a 64-bit system, timeout_id_t is also 64 bits.
  76  * So, the full 64-bit ID (sans the flags) can be returned. However, on 32-bit
  77  * systems, timeout_id_t is 32 bits. So, only the lower 32 bits can be
  78  * returned. In such cases, a default generation number of 0 is assigned to
  79  * the legacy IDs.
  80  *
  81  * The lower 32-bit ID space is partitioned into two spaces - one for 32-bit
  82  * IDs and the other for 64-bit IDs. The 32-bit ID space is further divided
  83  * into two spaces - one for short-term callouts and one for long-term.
  84  *
  85  * Here is the bit layout for the callout ID:
  86  *
  87  *      63    62    61  ...  32   31       30    29 .. X+1  X ... 1   0
  88  *  -----------------------------------------------------------------------
  89  *  | Exec | Hres | Generation | Long | Counter | ID bits | Table  | Type |
  90  *  |      | time | number     | term | High    |         | number |      |
  91  *  -----------------------------------------------------------------------
  92  *
  93  * Exec(uting):
  94  *    This is the executing bit which is only set in the extended callout
  95  *    ID. This bit indicates that the callout handler is currently being
  96  *    executed.
  97  *
  98  * Hrestime:
  99  *    Kernel features like condition variables use hrestime (system date) in
 100  *    conjunction with callouts. Under normal circumstances, these callouts
 101  *    are handled in the usual manner. They go off at specified times. But
 102  *    when the system time is changed abruptly (e.g., via stime()), these
 103  *    callouts are required to be processed immediately so that they can
 104  *    wakeup their threads immediately. The Hrestime bit is used to mark
 105  *    such callouts. When the system time is changed, the callout subsystem
 106  *    is called to process all callouts with this bit set.
 107  *
 108  * Generation number:
 109  *    This is the generation part of the ID.
 110  *
 111  * Long term:
 112  *    This bit indicates whether this is a short-term or a long-term callout.
 113  *    The long-term bit exists to address the problem of callout ID collision
 114  *    on 32-bit systems. This is an issue because the system typically
 115  *    generates a large number of timeout() requests, which means that callout
 116  *    IDs eventually get recycled. Most timeouts are very short-lived, so that
 117  *    ID recycling isn't a problem; but there are a handful of timeouts which
 118  *    are sufficiently long-lived to see their own IDs reused. We use the
 119  *    long-term bit to partition the ID namespace into pieces; the short-term
 120  *    space gets all the heavy traffic and can wrap frequently (i.e., on the
 121  *    order of a day) with no ill effects; the long-term space gets very little
 122  *    traffic and thus never wraps. That said, we need to future proof callouts
 123  *    in case 32-bit systems grow in size and are able to consume callout IDs
 124  *    at faster rates. So, we should make all the kernel clients that use
 125  *    callouts to use the internal interface so that they can use IDs outside
 126  *    of the legacy space with a proper generation number.
 127  *
 128  * Counter High + ID counter bits:
 129  *    These bits represent the actual ID bits in the callout ID.
 130  *    The highest bit of the running counter is always set; this ensures that
 131  *    the callout ID is always non-zero, thus eliminating the need for an
 132  *    explicit wrap-around test during ID generation.
 133  *
 134  * Table number:
 135  *    These bits carry the table number for the callout table where the callout
 136  *    is queued. Each CPU has its own callout table. So, the callout tables are
 137  *    numbered from 0 - (max_ncpus - 1). Because max_ncpus is different on
 138  *    different systems, the actual number of table number bits will vary
 139  *    accordingly. And so will the ID counter bits.
 140  *
 141  * Type:
 142  *    This bit represents the callout (table) type. Each CPU has one realtime
 143  *    and one normal callout table.
 144  */
 145 #define CALLOUT_EXECUTING       0x8000000000000000ULL
 146 #define CALLOUT_HRESTIME        0x4000000000000000ULL
 147 #define CALLOUT_ID_MASK         ~(CALLOUT_EXECUTING | CALLOUT_HRESTIME)
 148 #define CALLOUT_GENERATION_LOW  0x100000000ULL
 149 #define CALLOUT_LONGTERM        0x80000000
 150 #define CALLOUT_COUNTER_HIGH    0x40000000
 151 #define CALLOUT_TYPE_BITS       1
 152 #define CALLOUT_NTYPES          (1 << CALLOUT_TYPE_BITS)
 153 #define CALLOUT_TYPE_MASK       (CALLOUT_NTYPES - 1)
 154 #define CALLOUT_COUNTER_SHIFT   callout_table_bits
 155 #define CALLOUT_TABLE(t, f)     (((f) << CALLOUT_TYPE_BITS) | (t))
 156 #define CALLOUT_TABLE_NUM(ct)   ((ct) - callout_table)
 157 #define CALLOUT_TABLE_TYPE(ct)  (CALLOUT_TABLE_NUM(ct) & CALLOUT_TYPE_MASK)
 158 #define CALLOUT_TABLE_SEQID(ct) (CALLOUT_TABLE_NUM(ct) >> CALLOUT_TYPE_BITS)
 159
 160 /*
 161  * We assume that during any period of CALLOUT_LONGTERM_TICKS ticks, at most
 162  * (CALLOUT_COUNTER_HIGH / callout_counter_low) callouts will be generated.
 163  */
 164 #define CALLOUT_LONGTERM_TICKS  0x4000UL
 165 #define CALLOUT_BUCKET_SHIFT    9
 166 #define CALLOUT_BUCKETS         (1 << CALLOUT_BUCKET_SHIFT)
 167 #define CALLOUT_BUCKET_MASK     (CALLOUT_BUCKETS - 1)
 168 #define CALLOUT_HASH(x)         ((x) & CALLOUT_BUCKET_MASK)
 169 #define CALLOUT_IDHASH(x)       CALLOUT_HASH((x) >> CALLOUT_COUNTER_SHIFT)
 170 /*
 171  * The multiply by 0 and 1 below are cosmetic. Just to align things better
 172  * and make it more readable. The multiplications will be done at compile
 173  * time.
 174  */
 175 #define CALLOUT_CLHASH(x)                       \
 176         CALLOUT_HASH(                           \
 177             ((x)>>(CALLOUT_BUCKET_SHIFT*0)) ^   \
 178             ((x)>>(CALLOUT_BUCKET_SHIFT*1)) ^   \
 179             ((x)>>(CALLOUT_BUCKET_SHIFT*2)) ^   \
 180             ((x)>>(CALLOUT_BUCKET_SHIFT*3)))
 181
 182 #define CALLOUT_ID_TO_TABLE(id)         ((id) & callout_table_mask)
 183
 184 #define CALLOUT_SHORT_ID(table)         \
 185                 ((callout_id_t)(table) | CALLOUT_COUNTER_HIGH)
 186 #define CALLOUT_LONG_ID(table)          \
 187                 (CALLOUT_SHORT_ID(table) | CALLOUT_LONGTERM)
 188
 189 #define CALLOUT_THREADS         2               /* keep it simple for now */
 190
 191 #define CALLOUT_REALTIME        0               /* realtime callout type */
 192 #define CALLOUT_NORMAL          1               /* normal callout type */
 193
 194 /*
 195  * callout_t's are cache-aligned structures allocated from kmem caches. One kmem
 196  * cache is created per lgrp and is shared by all CPUs in that lgrp. Benefits:
 197  *      - cache pages are mapped only in the TLBs of the CPUs of the lgrp
 198  *      - data in cache pages is present only in those CPU caches
 199  *      - memory access performance improves with locality-awareness in kmem
 200  *
 201  * The following structure is used to manage per-lgroup kmem caches.
 202  *
 203  * NOTE: Free callout_t's go to a callout table's freelist. CPUs map to callout
 204  * tables via their sequence IDs, not CPU IDs. DR operations can cause a
 205  * free list to have callouts from multiple lgrp caches. This takes away some
 206  * performance, but is no worse than if we did not use lgrp caches at all.
 207  */
 208 typedef struct callout_cache {
 209         struct callout_cache    *cc_next;       /* link in the global list */
 210         lgrp_handle_t           cc_hand;        /* lgroup handle */
 211         kmem_cache_t            *cc_cache;      /* kmem cache pointer */
 212         kmem_cache_t            *cc_lcache;     /* kmem cache pointer */
 213 } callout_cache_t;
 214
 215 /*
 216  * The callout hash structure is used for queueing both callouts and
 217  * callout lists. That is why the fields are declared as void *.
 218  */
 219 typedef struct callout_hash {
 220         void    *ch_head;
 221         void    *ch_tail;
 222 } callout_hash_t;
 223
 224 struct callout_list {
 225         callout_list_t  *cl_next;       /* next in clhash */
 226         callout_list_t  *cl_prev;       /* prev in clhash */
 227         hrtime_t        cl_expiration;  /* expiration for callouts in list */
 228         callout_hash_t  cl_callouts;    /* list of callouts */
 229         kcondvar_t      cl_done;        /* signal callout completion */
 230         ushort_t        cl_waiting;     /* count of waiting untimeouts */
 231         kthread_id_t    cl_executor;    /* thread executing callout */
 232         ulong_t         cl_pad;         /* cache alignment */
 233 };
 234
 235 /*
 236  * Per-callout table kstats.
 237  *
 238  * CALLOUT_TIMEOUTS
 239  *      Callouts created since boot.
 240  * CALLOUT_TIMEOUTS_PENDING
 241  *      Number of outstanding callouts.
 242  * CALLOUT_UNTIMEOUTS_UNEXPIRED
 243  *      Number of cancelled callouts that have not expired.
 244  * CALLOUT_UNTIMEOUTS_EXECUTING
 245  *      Number of cancelled callouts that were executing at the time of
 246  *      cancellation.
 247  * CALLOUT_UNTIMEOUTS_EXPIRED
 248  *      Number of cancelled callouts that had already expired at the time
 249  *      of cancellations.
 250  * CALLOUT_EXPIRATIONS
 251  *      Number of callouts that expired.
 252  * CALLOUT_ALLOCATIONS
 253  *      Number of callout structures allocated.
 254  */
 255 typedef enum callout_stat_type {
 256         CALLOUT_TIMEOUTS,
 257         CALLOUT_TIMEOUTS_PENDING,
 258         CALLOUT_UNTIMEOUTS_UNEXPIRED,
 259         CALLOUT_UNTIMEOUTS_EXECUTING,
 260         CALLOUT_UNTIMEOUTS_EXPIRED,
 261         CALLOUT_EXPIRATIONS,
 262         CALLOUT_ALLOCATIONS,
 263         CALLOUT_NUM_STATS
 264 } callout_stat_type_t;
 265
 266 /*
 267  * Callout flags:
 268  *
 269  * CALLOUT_FLAG_ROUNDUP
 270  *      Roundup the expiration time to the nearest resolution boundary.
 271  *      If this flag is not specified, the expiration time is rounded down.
 272  * CALLOUT_FLAG_ABSOLUTE
 273  *      Normally, the expiration passed to the timeout API functions is an
 274  *      expiration interval. If this flag is specified, then it is
 275  *      interpreted as the expiration time itself.
 276  * CALLOUT_FLAG_HRESTIME
 277  *      Normally, callouts are not affected by changes to system time
 278  *      (hrestime). This flag is used to create a callout that is affected
 279  *      by system time. If system time changes, these timers must expire
 280  *      at once. These are used by condition variables and LWP timers that
 281  *      need this behavior.
 282  * CALLOUT_FLAG_32BIT
 283  *      Legacy interfaces timeout() and realtime_timeout() pass this flag
 284  *      to timeout_generic() to indicate that a 32-bit ID should be allocated.
 285  */
 286 #define CALLOUT_FLAG_ROUNDUP            0x1
 287 #define CALLOUT_FLAG_ABSOLUTE           0x2
 288 #define CALLOUT_FLAG_HRESTIME           0x4
 289 #define CALLOUT_FLAG_32BIT              0x8
 290
 291 /*
 292  * On 32-bit systems, the legacy interfaces, timeout() and realtime_timeout(),
 293  * must pass CALLOUT_FLAG_32BIT to timeout_generic() so that a 32-bit ID
 294  * can be generated.
 295  */
 296 #ifdef _LP64
 297 #define CALLOUT_LEGACY          0
 298 #else
 299 #define CALLOUT_LEGACY          CALLOUT_FLAG_32BIT
 300 #endif
 301
 302 /*
 303  * All of the state information associated with a callout table.
 304  * The fields are ordered with cache performance in mind.
 305  */
 306 typedef struct callout_table {
 307         kmutex_t        ct_mutex;       /* protects all callout state */
 308         callout_t       *ct_free;       /* free callout structures */
 309         callout_list_t  *ct_lfree;      /* free callout list structures */
 310         callout_id_t    ct_short_id;    /* most recently issued short-term ID */
 311         callout_id_t    ct_long_id;     /* most recently issued long-term ID */
 312         callout_hash_t  *ct_idhash;     /* ID hash chains */
 313         callout_hash_t  *ct_clhash;     /* callout list hash */
 314         kstat_named_t   *ct_kstat_data; /* callout kstat data */
 315
 316         uint_t          ct_type;        /* callout table type */
 317         uint_t          ct_suspend;     /* suspend count */
 318         cyclic_id_t     ct_cyclic;      /* cyclic for this table */
 319         hrtime_t        *ct_heap;       /* callout expiration heap */
 320         ulong_t         ct_heap_num;    /* occupied slots in the heap */
 321         ulong_t         ct_heap_max;    /* end of the heap */
 322         kmem_cache_t    *ct_cache;      /* callout kmem cache */
 323         kmem_cache_t    *ct_lcache;     /* callout list kmem cache */
 324         callout_id_t    ct_gen_id;      /* generation based ID */
 325
 326         callout_hash_t  ct_expired;     /* list of expired callout lists */
 327         taskq_t         *ct_taskq;      /* taskq to execute normal callouts */
 328         kstat_t         *ct_kstats;     /* callout kstats */
 329 #ifdef _LP64
 330         ulong_t         ct_pad[4];      /* cache alignment */
 331 #else
 332         ulong_t         ct_pad[7];      /* cache alignment */
 333 #endif
 334 } callout_table_t;
 335
 336 /*
 337  * Short hand definitions for the callout kstats.
 338  */
 339 #define ct_timeouts                                                     \
 340                 ct_kstat_data[CALLOUT_TIMEOUTS].value.ui64
 341 #define ct_timeouts_pending                                             \
 342                 ct_kstat_data[CALLOUT_TIMEOUTS_PENDING].value.ui64
 343 #define ct_untimeouts_unexpired                                         \
 344                 ct_kstat_data[CALLOUT_UNTIMEOUTS_UNEXPIRED].value.ui64
 345 #define ct_untimeouts_executing                                         \
 346                 ct_kstat_data[CALLOUT_UNTIMEOUTS_EXECUTING].value.ui64
 347 #define ct_untimeouts_expired                                           \
 348                 ct_kstat_data[CALLOUT_UNTIMEOUTS_EXPIRED].value.ui64
 349 #define ct_expirations                                                  \
 350                 ct_kstat_data[CALLOUT_EXPIRATIONS].value.ui64
 351 #define ct_allocations                                                  \
 352                 ct_kstat_data[CALLOUT_ALLOCATIONS].value.ui64
 353
 354 #define CALLOUT_CHUNK   128
 355
 356 #define CALLOUT_HEAP_PARENT(index)      (((index) - 1) >> 1)
 357 #define CALLOUT_HEAP_RIGHT(index)       (((index) + 1) << 1)
 358 #define CALLOUT_HEAP_LEFT(index)        ((((index) + 1) << 1) - 1)
 359
 360 #define CALLOUT_CYCLIC_HANDLER(t)                                       \
 361         ((t == CALLOUT_REALTIME) ? callout_realtime : callout_normal)
 362
 363 /*
 364  * We define a blanket minimum resolution for callouts of 1 millisecond.
 365  * 1 millisecond is a safe value as it is already supported when the clock
 366  * resolution is set to high.
 367  */
 368 #define CALLOUT_MIN_RESOLUTION          1000000ULL
 369 #define CALLOUT_TCP_RESOLUTION          10000000ULL
 370
 371 #define CALLOUT_ALIGN   64      /* cache line size */
 372
 373 #ifdef _LP64
 374 #define CALLOUT_MAX_TICKS       NSEC_TO_TICK(CY_INFINITY);
 375 #else
 376 #define CALLOUT_MAX_TICKS       LONG_MAX
 377 #endif
 378
 379 extern void             callout_init(void);
 380 extern void             membar_sync(void);
 381 extern void             callout_cpu_online(cpu_t *);
 382 extern void             callout_cpu_offline(cpu_t *);
 383 extern void             callout_hrestime(void);
 384
 385 #endif
 386
 387 #ifdef  __cplusplus
 388 }
 389 #endif
 390
 391 #endif  /* _SYS_CALLO_H */