osfmk/kern/sched_average.c

   1 /*
   2  * Copyright (c) 2000-2007 Apple Computer, 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  * @OSF_COPYRIGHT@
  30  */
  31 /*
  32  * Mach Operating System
  33  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
  34  * All Rights Reserved.
  35  *
  36  * Permission to use, copy, modify and distribute this software and its
  37  * documentation is hereby granted, provided that both the copyright
  38  * notice and this permission notice appear in all copies of the
  39  * software, derivative works or modified versions, and any portions
  40  * thereof, and that both notices appear in supporting documentation.
  41  *
  42  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  43  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  44  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  45  *
  46  * Carnegie Mellon requests users of this software to return to
  47  *
  48  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  49  *  School of Computer Science
  50  *  Carnegie Mellon University
  51  *  Pittsburgh PA 15213-3890
  52  *
  53  * any improvements or extensions that they make and grant Carnegie Mellon
  54  * the rights to redistribute these changes.
  55  */
  56 /*
  57  */
  58 /*
  59  *      Author: Avadis Tevanian, Jr.
  60  *      Date:   1986
  61  *
  62  *      Compute various averages.
  63  */
  64
  65 #include <mach/mach_types.h>
  66
  67 #include <kern/sched.h>
  68 #include <kern/assert.h>
  69 #include <kern/processor.h>
  70 #include <kern/thread.h>
  71 #if CONFIG_TELEMETRY
  72 #include <kern/telemetry.h>
  73 #endif
  74
  75 uint32_t        avenrun[3] = {0, 0, 0};
  76 uint32_t        mach_factor[3] = {0, 0, 0};
  77
  78 #if defined(CONFIG_SCHED_TRADITIONAL)
  79 /*
  80  * Values are scaled by LOAD_SCALE, defined in processor_info.h
  81  */
  82 #define base(n)         ((n) << SCHED_TICK_SHIFT)
  83 #define frac(n)         (((base(n) - 1) * LOAD_SCALE) / base(n))
  84
  85 static uint32_t         fract[3] = {
  86         frac(5),                /* 5 second average */
  87         frac(30),               /* 30 second average */
  88         frac(60),               /* 1 minute average */
  89 };
  90
  91 #undef base
  92 #undef frac
  93
  94 #endif /* CONFIG_SCHED_TRADITIONAL */
  95
  96 static unsigned int             sched_nrun;
  97
  98 typedef void    (*sched_avg_comp_t)(
  99                                         void                    *param);
 100
 101 static struct sched_average {
 102         sched_avg_comp_t        comp;
 103         void                    *param;
 104         int                     period; /* in seconds */
 105         uint64_t                deadline;
 106 } sched_average[] = {
 107         { compute_averunnable, &sched_nrun, 5, 0 },
 108         { compute_stack_target, NULL, 5, 1 },
 109         { compute_memory_pressure, NULL, 1, 0 },
 110         { compute_zone_gc_throttle, NULL, 60, 0 },
 111         { compute_pageout_gc_throttle, NULL, 1, 0 },
 112         { compute_pmap_gc_throttle, NULL, 60, 0 },
 113 #if CONFIG_TELEMETRY
 114         { compute_telemetry, NULL, 1, 0 },
 115 #endif
 116         { NULL, NULL, 0, 0 }
 117 };
 118
 119 typedef struct sched_average    *sched_average_t;
 120
 121 /* The "stdelta" parameter represents the number of scheduler maintenance
 122  * "ticks" that have elapsed since the last invocation, subject to
 123  * integer division imprecision.
 124  */
 125
 126 void
 127 compute_averages(uint64_t stdelta)
 128 {
 129         int                     ncpus, nthreads, nshared, nbackground, nshared_non_bg;
 130         uint32_t                factor_now, average_now, load_now = 0, background_load_now = 0, combined_fgbg_load_now = 0;
 131         sched_average_t         avg;
 132         uint64_t                abstime, index;
 133
 134         /*
 135          *      Retrieve counts, ignoring
 136          *      the current thread.
 137          */
 138         ncpus = processor_avail_count;
 139         nthreads = sched_run_count - 1;
 140         nshared = sched_share_count;
 141         nbackground = sched_background_count;
 142
 143         /*
 144          *      Load average and mach factor calculations for
 145          *      those which ask about these things.
 146          */
 147         average_now = nthreads * LOAD_SCALE;
 148
 149         if (nthreads > ncpus)
 150                 factor_now = (ncpus * LOAD_SCALE) / (nthreads + 1);
 151         else
 152                 factor_now = (ncpus - nthreads) * LOAD_SCALE;
 153
 154         /* For those statistics that formerly relied on being recomputed
 155          * on timer ticks, advance by the approximate number of corresponding
 156          * elapsed intervals, thus compensating for potential idle intervals.
 157          */
 158         for (index = 0; index < stdelta; index++) {
 159                 sched_mach_factor = ((sched_mach_factor << 2) + factor_now) / 5;
 160                 sched_load_average = ((sched_load_average << 2) + average_now) / 5;
 161         }
 162         /*
 163          * Compute the timeshare priority conversion factor based on loading.
 164          * Because our counters may be incremented and accessed
 165          * concurrently with respect to each other, we may have
 166          * windows where the invariant nthreads >= nshared >= nbackground
 167          * is broken, so truncate values in these cases.
 168          */
 169
 170         if (nshared > nthreads)
 171                 nshared = nthreads;
 172
 173         if (nbackground > nshared)
 174                 nbackground = nshared;
 175
 176         nshared_non_bg = nshared - nbackground;
 177
 178         if (nshared_non_bg > ncpus) {
 179                 if (ncpus > 1)
 180                         load_now = nshared_non_bg / ncpus;
 181                 else
 182                         load_now = nshared_non_bg;
 183
 184                 if (load_now > NRQS - 1)
 185                         load_now = NRQS - 1;
 186         }
 187
 188         if (nbackground > ncpus) {
 189                 if (ncpus > 1)
 190                         background_load_now = nbackground / ncpus;
 191                 else
 192                         background_load_now = nbackground;
 193
 194                 if (background_load_now > NRQS - 1)
 195                         background_load_now = NRQS - 1;
 196         }
 197
 198         if (nshared > ncpus) {
 199                 if (ncpus > 1)
 200                         combined_fgbg_load_now = nshared / ncpus;
 201                 else
 202                         combined_fgbg_load_now = nshared;
 203
 204                 if (combined_fgbg_load_now > NRQS - 1)
 205                         combined_fgbg_load_now = NRQS - 1;
 206         }
 207
 208         /*
 209          *      Sample total running threads.
 210          */
 211         sched_nrun = nthreads;
 212
 213 #if defined(CONFIG_SCHED_TRADITIONAL)
 214
 215         /*
 216          *      The conversion factor consists of
 217          *      two components: a fixed value based
 218          *      on the absolute time unit, and a
 219          *      dynamic portion based on loading.
 220          *
 221          *      Zero loading results in a out of range
 222          *      shift count.  Accumulated usage is ignored
 223          *      during conversion and new usage deltas
 224          *      are discarded.
 225          */
 226         sched_pri_shift = sched_fixed_shift - sched_load_shifts[load_now];
 227         sched_background_pri_shift = sched_fixed_shift - sched_load_shifts[background_load_now];
 228         sched_combined_fgbg_pri_shift = sched_fixed_shift - sched_load_shifts[combined_fgbg_load_now];
 229
 230         /*
 231          * Compute old-style Mach load averages.
 232          */
 233
 234         for (index = 0; index < stdelta; index++) {
 235                 register int            i;
 236
 237                 for (i = 0; i < 3; i++) {
 238                         mach_factor[i] = ((mach_factor[i] * fract[i]) +
 239                                                 (factor_now * (LOAD_SCALE - fract[i]))) / LOAD_SCALE;
 240
 241                         avenrun[i] = ((avenrun[i] * fract[i]) +
 242                                                 (average_now * (LOAD_SCALE - fract[i]))) / LOAD_SCALE;
 243                 }
 244         }
 245 #endif /* CONFIG_SCHED_TRADITIONAL */
 246
 247         /*
 248          *      Compute averages in other components.
 249          */
 250         abstime = mach_absolute_time();
 251         for (avg = sched_average; avg->comp != NULL; ++avg) {
 252                 if (abstime >= avg->deadline) {
 253                         uint64_t period_abs = (avg->period * sched_one_second_interval);
 254                         uint64_t ninvokes = 1;
 255
 256                         ninvokes += (abstime - avg->deadline) / period_abs;
 257                         ninvokes = MIN(ninvokes, SCHED_TICK_MAX_DELTA);
 258
 259                         for (index = 0; index < ninvokes; index++) {
 260                                 (*avg->comp)(avg->param);
 261                         }
 262                         avg->deadline = abstime + period_abs;
 263                 }
 264         }
 265 }