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 #include <sys/kdebug.h>
  76
  77 uint32_t        avenrun[3] = {0, 0, 0};
  78 uint32_t        mach_factor[3] = {0, 0, 0};
  79
  80 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
  81 /*
  82  * Values are scaled by LOAD_SCALE, defined in processor_info.h
  83  */
  84 #define base(n)         ((n) << SCHED_TICK_SHIFT)
  85 #define frac(n)         (((base(n) - 1) * LOAD_SCALE) / base(n))
  86
  87 static uint32_t         fract[3] = {
  88         frac(5),                /* 5 second average */
  89         frac(30),               /* 30 second average */
  90         frac(60),               /* 1 minute average */
  91 };
  92
  93 #undef base
  94 #undef frac
  95
  96 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
  97
  98 static unsigned int             sched_nrun;
  99
 100 typedef void    (*sched_avg_comp_t)(
 101                                         void                    *param);
 102
 103 static struct sched_average {
 104         sched_avg_comp_t        comp;
 105         void                    *param;
 106         int                     period; /* in seconds */
 107         uint64_t                deadline;
 108 } sched_average[] = {
 109         { compute_averunnable, &sched_nrun, 5, 0 },
 110         { compute_stack_target, NULL, 5, 1 },
 111         { compute_memory_pressure, NULL, 1, 0 },
 112         { compute_zone_gc_throttle, NULL, 60, 0 },
 113         { compute_pageout_gc_throttle, NULL, 1, 0 },
 114         { compute_pmap_gc_throttle, NULL, 60, 0 },
 115 #if CONFIG_TELEMETRY
 116         { compute_telemetry, NULL, 1, 0 },
 117 #endif
 118         { NULL, NULL, 0, 0 }
 119 };
 120
 121 typedef struct sched_average    *sched_average_t;
 122
 123 /* The "stdelta" parameter represents the number of scheduler maintenance
 124  * "ticks" that have elapsed since the last invocation, subject to
 125  * integer division imprecision.
 126  */
 127
 128 void
 129 compute_averages(uint64_t stdelta)
 130 {
 131         int                     ncpus, nthreads, nshared, nbackground, nshared_non_bg;
 132         uint32_t                factor_now, average_now, load_now = 0, background_load_now = 0, combined_fgbg_load_now = 0;
 133         sched_average_t         avg;
 134         uint64_t                abstime, index;
 135
 136         /*
 137          *      Retrieve counts, ignoring
 138          *      the current thread.
 139          */
 140         ncpus = processor_avail_count;
 141         nthreads = sched_run_count - 1;
 142         nshared = sched_share_count;
 143         nbackground = sched_background_count;
 144
 145         /*
 146          *      Load average and mach factor calculations for
 147          *      those which ask about these things.
 148          */
 149         average_now = nthreads * LOAD_SCALE;
 150
 151         if (nthreads > ncpus)
 152                 factor_now = (ncpus * LOAD_SCALE) / (nthreads + 1);
 153         else
 154                 factor_now = (ncpus - nthreads) * LOAD_SCALE;
 155
 156         /* For those statistics that formerly relied on being recomputed
 157          * on timer ticks, advance by the approximate number of corresponding
 158          * elapsed intervals, thus compensating for potential idle intervals.
 159          */
 160         for (index = 0; index < stdelta; index++) {
 161                 sched_mach_factor = ((sched_mach_factor << 2) + factor_now) / 5;
 162                 sched_load_average = ((sched_load_average << 2) + average_now) / 5;
 163         }
 164         /*
 165          * Compute the timeshare priority conversion factor based on loading.
 166          * Because our counters may be incremented and accessed
 167          * concurrently with respect to each other, we may have
 168          * windows where the invariant nthreads >= nshared >= nbackground
 169          * is broken, so truncate values in these cases.
 170          */
 171
 172         if (nshared > nthreads)
 173                 nshared = nthreads;
 174
 175         if (nbackground > nshared)
 176                 nbackground = nshared;
 177
 178         nshared_non_bg = nshared - nbackground;
 179
 180         if (nshared_non_bg > ncpus) {
 181                 if (ncpus > 1)
 182                         load_now = nshared_non_bg / ncpus;
 183                 else
 184                         load_now = nshared_non_bg;
 185
 186                 if (load_now > NRQS - 1)
 187                         load_now = NRQS - 1;
 188         }
 189
 190         if (nbackground > ncpus) {
 191                 if (ncpus > 1)
 192                         background_load_now = nbackground / ncpus;
 193                 else
 194                         background_load_now = nbackground;
 195
 196                 if (background_load_now > NRQS - 1)
 197                         background_load_now = NRQS - 1;
 198         }
 199
 200         if (nshared > ncpus) {
 201                 if (ncpus > 1)
 202                         combined_fgbg_load_now = nshared / ncpus;
 203                 else
 204                         combined_fgbg_load_now = nshared;
 205
 206                 if (combined_fgbg_load_now > NRQS - 1)
 207                         combined_fgbg_load_now = NRQS - 1;
 208         }
 209
 210         KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
 211                 MACHDBG_CODE(DBG_MACH_SCHED, MACH_SCHED_LOAD) | DBG_FUNC_NONE,
 212                 (nthreads - nshared), (nshared - nbackground), nbackground, 0, 0);
 213
 214         /*
 215          *      Sample total running threads.
 216          */
 217         sched_nrun = nthreads;
 218
 219 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
 220
 221         /*
 222          *      The conversion factor consists of
 223          *      two components: a fixed value based
 224          *      on the absolute time unit, and a
 225          *      dynamic portion based on loading.
 226          *
 227          *      Zero loading results in a out of range
 228          *      shift count.  Accumulated usage is ignored
 229          *      during conversion and new usage deltas
 230          *      are discarded.
 231          */
 232         sched_pri_shift = sched_fixed_shift - sched_load_shifts[load_now];
 233         sched_background_pri_shift = sched_fixed_shift - sched_load_shifts[background_load_now];
 234         sched_combined_fgbg_pri_shift = sched_fixed_shift - sched_load_shifts[combined_fgbg_load_now];
 235
 236         /*
 237          * Compute old-style Mach load averages.
 238          */
 239
 240         for (index = 0; index < stdelta; index++) {
 241                 register int            i;
 242
 243                 for (i = 0; i < 3; i++) {
 244                         mach_factor[i] = ((mach_factor[i] * fract[i]) +
 245                                                 (factor_now * (LOAD_SCALE - fract[i]))) / LOAD_SCALE;
 246
 247                         avenrun[i] = ((avenrun[i] * fract[i]) +
 248                                                 (average_now * (LOAD_SCALE - fract[i]))) / LOAD_SCALE;
 249                 }
 250         }
 251 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
 252
 253         /*
 254          *      Compute averages in other components.
 255          */
 256         abstime = mach_absolute_time();
 257         for (avg = sched_average; avg->comp != NULL; ++avg) {
 258                 if (abstime >= avg->deadline) {
 259                         uint64_t period_abs = (avg->period * sched_one_second_interval);
 260                         uint64_t ninvokes = 1;
 261
 262                         ninvokes += (abstime - avg->deadline) / period_abs;
 263                         ninvokes = MIN(ninvokes, SCHED_TICK_MAX_DELTA);
 264
 265                         for (index = 0; index < ninvokes; index++) {
 266                                 (*avg->comp)(avg->param);
 267                         }
 268                         avg->deadline = abstime + period_abs;
 269                 }
 270         }
 271 }