bsd/kern/kern_synch.c

   1 /*
   2  * Copyright (c) 2000-2006 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  * Mach Operating System
  30  * Copyright (c) 1987 Carnegie-Mellon University
  31  * All rights reserved.  The CMU software License Agreement specifies
  32  * the terms and conditions for use and redistribution.
  33  */
  34
  35 #include <sys/param.h>
  36 #include <sys/systm.h>
  37 #include <sys/proc_internal.h>
  38 #include <sys/user.h>
  39 #include <sys/file_internal.h>
  40 #include <sys/vnode.h>
  41 #include <sys/kernel.h>
  42
  43 #include <machine/spl.h>
  44
  45 #include <kern/queue.h>
  46 #include <sys/lock.h>
  47 #include <kern/thread.h>
  48 #include <kern/sched_prim.h>
  49 #include <kern/ast.h>
  50
  51 #include <kern/cpu_number.h>
  52 #include <vm/vm_kern.h>
  53
  54 #include <kern/task.h>
  55 #include <mach/time_value.h>
  56 #include <kern/locks.h>
  57
  58 #include <sys/systm.h>                  /* for unix_syscall_return() */
  59 #include <libkern/OSAtomic.h>
  60
  61 extern void compute_averunnable(void *);        /* XXX */
  62
  63
  64
  65 static void
  66 _sleep_continue( __unused void *parameter, wait_result_t wresult)
  67 {
  68         struct proc *p = current_proc();
  69         thread_t self  = current_thread();
  70         struct uthread * ut;
  71         int sig, catch;
  72         int error = 0;
  73         int dropmutex, spinmutex;
  74
  75         ut = get_bsdthread_info(self);
  76         catch     = ut->uu_pri & PCATCH;
  77         dropmutex = ut->uu_pri & PDROP;
  78         spinmutex = ut->uu_pri & PSPIN;
  79
  80         switch (wresult) {
  81                 case THREAD_TIMED_OUT:
  82                         error = EWOULDBLOCK;
  83                         break;
  84                 case THREAD_AWAKENED:
  85                         /*
  86                          * Posix implies any signal should be delivered
  87                          * first, regardless of whether awakened due
  88                          * to receiving event.
  89                          */
  90                         if (!catch)
  91                                 break;
  92                         /* else fall through */
  93                 case THREAD_INTERRUPTED:
  94                         if (catch) {
  95                                 if (thread_should_abort(self)) {
  96                                         error = EINTR;
  97                                 } else if (SHOULDissignal(p,ut)) {
  98                                         if ((sig = CURSIG(p)) != 0) {
  99                                                 if (p->p_sigacts->ps_sigintr & sigmask(sig))
 100                                                         error = EINTR;
 101                                                 else
 102                                                         error = ERESTART;
 103                                         }
 104                                         if (thread_should_abort(self)) {
 105                                                 error = EINTR;
 106                                         }
 107                                 } else if( (ut->uu_flag & ( UT_CANCELDISABLE | UT_CANCEL | UT_CANCELED)) == UT_CANCEL) {
 108                                         /* due to thread cancel */
 109                                         error = EINTR;
 110                                 }
 111                         }  else
 112                                 error = EINTR;
 113                         break;
 114         }
 115
 116         if (error == EINTR || error == ERESTART)
 117                 act_set_astbsd(self);
 118
 119         if (ut->uu_mtx && !dropmutex) {
 120                 if (spinmutex)
 121                         lck_mtx_lock_spin(ut->uu_mtx);
 122                 else
 123                         lck_mtx_lock(ut->uu_mtx);
 124         }
 125         ut->uu_wchan = NULL;
 126         ut->uu_wmesg = NULL;
 127
 128         unix_syscall_return((*ut->uu_continuation)(error));
 129 }
 130
 131 /*
 132  * Give up the processor till a wakeup occurs
 133  * on chan, at which time the process
 134  * enters the scheduling queue at priority pri.
 135  * The most important effect of pri is that when
 136  * pri<=PZERO a signal cannot disturb the sleep;
 137  * if pri>PZERO signals will be processed.
 138  * If pri&PCATCH is set, signals will cause sleep
 139  * to return 1, rather than longjmp.
 140  * Callers of this routine must be prepared for
 141  * premature return, and check that the reason for
 142  * sleeping has gone away.
 143  *
 144  * if msleep was the entry point, than we have a mutex to deal with
 145  *
 146  * The mutex is unlocked before the caller is blocked, and
 147  * relocked before msleep returns unless the priority includes the PDROP
 148  * flag... if PDROP is specified, _sleep returns with the mutex unlocked
 149  * regardless of whether it actually blocked or not.
 150  */
 151
 152 static int
 153 _sleep(
 154         caddr_t         chan,
 155         int             pri,
 156         const char      *wmsg,
 157         u_int64_t       abstime,
 158         int             (*continuation)(int),
 159         lck_mtx_t       *mtx)
 160 {
 161         struct proc *p;
 162         thread_t self = current_thread();
 163         struct uthread * ut;
 164         int sig, catch;
 165         int dropmutex  = pri & PDROP;
 166         int spinmutex  = pri & PSPIN;
 167         int wait_result;
 168         int error = 0;
 169
 170         ut = get_bsdthread_info(self);
 171
 172         p = current_proc();
 173         p->p_priority = pri & PRIMASK;
 174         /* It can still block in proc_exit() after the teardown. */
 175         if (p->p_stats != NULL)
 176                 OSIncrementAtomicLong(&p->p_stats->p_ru.ru_nvcsw);
 177
 178         if (pri & PCATCH)
 179                 catch = THREAD_ABORTSAFE;
 180         else
 181                 catch = THREAD_UNINT;
 182
 183         /* set wait message & channel */
 184         ut->uu_wchan = chan;
 185         ut->uu_wmesg = wmsg ? wmsg : "unknown";
 186
 187         if (mtx != NULL && chan != NULL && (thread_continue_t)continuation == THREAD_CONTINUE_NULL) {
 188                 int     flags;
 189
 190                 if (dropmutex)
 191                         flags = LCK_SLEEP_UNLOCK;
 192                 else
 193                         flags = LCK_SLEEP_DEFAULT;
 194
 195                 if (spinmutex)
 196                         flags |= LCK_SLEEP_SPIN;
 197
 198                 if (abstime)
 199                         wait_result = lck_mtx_sleep_deadline(mtx, flags, chan, catch, abstime);
 200                 else
 201                         wait_result = lck_mtx_sleep(mtx, flags, chan, catch);
 202         }
 203         else {
 204                 if (chan != NULL)
 205                         assert_wait_deadline(chan, catch, abstime);
 206                 if (mtx)
 207                         lck_mtx_unlock(mtx);
 208
 209                 if (catch == THREAD_ABORTSAFE) {
 210                         if (SHOULDissignal(p,ut)) {
 211                                 if ((sig = CURSIG(p)) != 0) {
 212                                         if (clear_wait(self, THREAD_INTERRUPTED) == KERN_FAILURE)
 213                                                 goto block;
 214                                         if (p->p_sigacts->ps_sigintr & sigmask(sig))
 215                                                 error = EINTR;
 216                                         else
 217                                                 error = ERESTART;
 218                                         if (mtx && !dropmutex) {
 219                                                 if (spinmutex)
 220                                                         lck_mtx_lock_spin(mtx);
 221                                                 else
 222                                                         lck_mtx_lock(mtx);
 223                                         }
 224                                         goto out;
 225                                 }
 226                         }
 227                         if (thread_should_abort(self)) {
 228                                 if (clear_wait(self, THREAD_INTERRUPTED) == KERN_FAILURE)
 229                                         goto block;
 230                                 error = EINTR;
 231
 232                                 if (mtx && !dropmutex) {
 233                                         if (spinmutex)
 234                                                 lck_mtx_lock_spin(mtx);
 235                                         else
 236                                                 lck_mtx_lock(mtx);
 237                                 }
 238                                 goto out;
 239                         }
 240                 }
 241
 242
 243 block:
 244                 if ((thread_continue_t)continuation != THREAD_CONTINUE_NULL) {
 245                         ut->uu_continuation = continuation;
 246                         ut->uu_pri  = pri;
 247                         ut->uu_timo = abstime? 1: 0;
 248                         ut->uu_mtx  = mtx;
 249                         (void) thread_block(_sleep_continue);
 250                         /* NOTREACHED */
 251                 }
 252
 253                 wait_result = thread_block(THREAD_CONTINUE_NULL);
 254
 255                 if (mtx && !dropmutex) {
 256                         if (spinmutex)
 257                                 lck_mtx_lock_spin(mtx);
 258                         else
 259                                 lck_mtx_lock(mtx);
 260                 }
 261         }
 262
 263         switch (wait_result) {
 264                 case THREAD_TIMED_OUT:
 265                         error = EWOULDBLOCK;
 266                         break;
 267                 case THREAD_AWAKENED:
 268                         /*
 269                          * Posix implies any signal should be delivered
 270                          * first, regardless of whether awakened due
 271                          * to receiving event.
 272                          */
 273                         if (catch != THREAD_ABORTSAFE)
 274                                 break;
 275                         /* else fall through */
 276                 case THREAD_INTERRUPTED:
 277                         if (catch == THREAD_ABORTSAFE) {
 278                                 if (thread_should_abort(self)) {
 279                                         error = EINTR;
 280                                 } else if (SHOULDissignal(p, ut)) {
 281                                         if ((sig = CURSIG(p)) != 0) {
 282                                                 if (p->p_sigacts->ps_sigintr & sigmask(sig))
 283                                                         error = EINTR;
 284                                                 else
 285                                                         error = ERESTART;
 286                                         }
 287                                         if (thread_should_abort(self)) {
 288                                                 error = EINTR;
 289                                         }
 290                                 } else if( (ut->uu_flag & ( UT_CANCELDISABLE | UT_CANCEL | UT_CANCELED)) == UT_CANCEL) {
 291                                         /* due to thread cancel */
 292                                         error = EINTR;
 293                                 }
 294                         }  else
 295                                 error = EINTR;
 296                         break;
 297         }
 298 out:
 299         if (error == EINTR || error == ERESTART)
 300                 act_set_astbsd(self);
 301         ut->uu_wchan = NULL;
 302         ut->uu_wmesg = NULL;
 303
 304         return (error);
 305 }
 306
 307 int
 308 sleep(
 309         void    *chan,
 310         int             pri)
 311 {
 312         return _sleep((caddr_t)chan, pri, (char *)NULL, 0, (int (*)(int))0, (lck_mtx_t *)0);
 313 }
 314
 315 int
 316 msleep0(
 317         void            *chan,
 318         lck_mtx_t       *mtx,
 319         int             pri,
 320         const char      *wmsg,
 321         int             timo,
 322         int             (*continuation)(int))
 323 {
 324         u_int64_t       abstime = 0;
 325
 326         if (timo)
 327                 clock_interval_to_deadline(timo, NSEC_PER_SEC / hz, &abstime);
 328
 329         return _sleep((caddr_t)chan, pri, wmsg, abstime, continuation, mtx);
 330 }
 331
 332 int
 333 msleep(
 334         void            *chan,
 335         lck_mtx_t       *mtx,
 336         int             pri,
 337         const char      *wmsg,
 338         struct timespec         *ts)
 339 {
 340         u_int64_t       abstime = 0;
 341
 342         if (ts && (ts->tv_sec || ts->tv_nsec)) {
 343                 nanoseconds_to_absolutetime((uint64_t)ts->tv_sec * NSEC_PER_SEC + ts->tv_nsec,  &abstime );
 344                 clock_absolutetime_interval_to_deadline( abstime, &abstime );
 345         }
 346
 347         return _sleep((caddr_t)chan, pri, wmsg, abstime, (int (*)(int))0, mtx);
 348 }
 349
 350 int
 351 msleep1(
 352         void            *chan,
 353         lck_mtx_t       *mtx,
 354         int             pri,
 355         const char      *wmsg,
 356         u_int64_t       abstime)
 357 {
 358         return _sleep((caddr_t)chan, pri, wmsg, abstime, (int (*)(int))0, mtx);
 359 }
 360
 361 int
 362 tsleep(
 363         void            *chan,
 364         int             pri,
 365         const char      *wmsg,
 366         int             timo)
 367 {
 368         u_int64_t       abstime = 0;
 369
 370         if (timo)
 371                 clock_interval_to_deadline(timo, NSEC_PER_SEC / hz, &abstime);
 372         return _sleep((caddr_t)chan, pri, wmsg, abstime, (int (*)(int))0, (lck_mtx_t *)0);
 373 }
 374
 375 int
 376 tsleep0(
 377         void            *chan,
 378         int             pri,
 379         const char      *wmsg,
 380         int             timo,
 381         int             (*continuation)(int))
 382 {
 383         u_int64_t       abstime = 0;
 384
 385         if (timo)
 386                 clock_interval_to_deadline(timo, NSEC_PER_SEC / hz, &abstime);
 387         return _sleep((caddr_t)chan, pri, wmsg, abstime, continuation, (lck_mtx_t *)0);
 388 }
 389
 390 int
 391 tsleep1(
 392         void            *chan,
 393         int             pri,
 394         const char      *wmsg,
 395         u_int64_t       abstime,
 396         int             (*continuation)(int))
 397 {
 398         return _sleep((caddr_t)chan, pri, wmsg, abstime, continuation, (lck_mtx_t *)0);
 399 }
 400
 401 /*
 402  * Wake up all processes sleeping on chan.
 403  */
 404 void
 405 wakeup(void *chan)
 406 {
 407         thread_wakeup((caddr_t)chan);
 408 }
 409
 410 /*
 411  * Wake up the first process sleeping on chan.
 412  *
 413  * Be very sure that the first process is really
 414  * the right one to wakeup.
 415  */
 416 void
 417 wakeup_one(caddr_t chan)
 418 {
 419         thread_wakeup_one((caddr_t)chan);
 420 }
 421
 422 /*
 423  * Compute the priority of a process when running in user mode.
 424  * Arrange to reschedule if the resulting priority is better
 425  * than that of the current process.
 426  */
 427 void
 428 resetpriority(struct proc *p)
 429 {
 430         (void)task_importance(p->task, -p->p_nice);
 431 }
 432
 433 struct loadavg averunnable =
 434         { {0, 0, 0}, FSCALE };          /* load average, of runnable procs */
 435 /*
 436  * Constants for averages over 1, 5, and 15 minutes
 437  * when sampling at 5 second intervals.
 438  */
 439 static fixpt_t cexp[3] = {
 440     (fixpt_t)(0.9200444146293232 * FSCALE),    /* exp(-1/12) */
 441     (fixpt_t)(0.9834714538216174 * FSCALE),    /* exp(-1/60) */
 442     (fixpt_t)(0.9944598480048967 * FSCALE),    /* exp(-1/180) */
 443 };
 444
 445 void
 446 compute_averunnable(void *arg)
 447 {
 448         unsigned int            nrun = *(unsigned int *)arg;
 449         struct loadavg          *avg = &averunnable;
 450         int             i;
 451
 452     for (i = 0; i < 3; i++)
 453         avg->ldavg[i] = (cexp[i] * avg->ldavg[i] +
 454             nrun * FSCALE * (FSCALE - cexp[i])) >> FSHIFT;
 455 }