The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_clock.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1991, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)kern_clock.c        8.5 (Berkeley) 1/21/94
   35  */
   36 
   37 #include <sys/cdefs.h>
   38 __FBSDID("$FreeBSD: releng/9.2/sys/kern/kern_clock.c 250600 2013-05-13 15:18:36Z fabient $");
   39 
   40 #include "opt_kdb.h"
   41 #include "opt_device_polling.h"
   42 #include "opt_hwpmc_hooks.h"
   43 #include "opt_kdtrace.h"
   44 #include "opt_ntp.h"
   45 #include "opt_watchdog.h"
   46 
   47 #include <sys/param.h>
   48 #include <sys/systm.h>
   49 #include <sys/callout.h>
   50 #include <sys/kdb.h>
   51 #include <sys/kernel.h>
   52 #include <sys/kthread.h>
   53 #include <sys/ktr.h>
   54 #include <sys/lock.h>
   55 #include <sys/mutex.h>
   56 #include <sys/proc.h>
   57 #include <sys/resource.h>
   58 #include <sys/resourcevar.h>
   59 #include <sys/sched.h>
   60 #include <sys/sdt.h>
   61 #include <sys/signalvar.h>
   62 #include <sys/sleepqueue.h>
   63 #include <sys/smp.h>
   64 #include <vm/vm.h>
   65 #include <vm/pmap.h>
   66 #include <vm/vm_map.h>
   67 #include <sys/sysctl.h>
   68 #include <sys/bus.h>
   69 #include <sys/interrupt.h>
   70 #include <sys/limits.h>
   71 #include <sys/timetc.h>
   72 
   73 #ifdef GPROF
   74 #include <sys/gmon.h>
   75 #endif
   76 
   77 #ifdef HWPMC_HOOKS
   78 #include <sys/pmckern.h>
   79 PMC_SOFT_DEFINE( , , clock, hard);
   80 PMC_SOFT_DEFINE( , , clock, stat);
   81 PMC_SOFT_DEFINE_EX( , , clock, prof, \
   82     cpu_startprofclock, cpu_stopprofclock);
   83 #endif
   84 
   85 #ifdef DEVICE_POLLING
   86 extern void hardclock_device_poll(void);
   87 #endif /* DEVICE_POLLING */
   88 
   89 static void initclocks(void *dummy);
   90 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
   91 
   92 /* Spin-lock protecting profiling statistics. */
   93 static struct mtx time_lock;
   94 
   95 SDT_PROVIDER_DECLARE(sched);
   96 SDT_PROBE_DEFINE2(sched, , , tick, tick, "struct thread *", "struct proc *");
   97 
   98 static int
   99 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
  100 {
  101         int error;
  102         long cp_time[CPUSTATES];
  103 #ifdef SCTL_MASK32
  104         int i;
  105         unsigned int cp_time32[CPUSTATES];
  106 #endif
  107 
  108         read_cpu_time(cp_time);
  109 #ifdef SCTL_MASK32
  110         if (req->flags & SCTL_MASK32) {
  111                 if (!req->oldptr)
  112                         return SYSCTL_OUT(req, 0, sizeof(cp_time32));
  113                 for (i = 0; i < CPUSTATES; i++)
  114                         cp_time32[i] = (unsigned int)cp_time[i];
  115                 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
  116         } else
  117 #endif
  118         {
  119                 if (!req->oldptr)
  120                         return SYSCTL_OUT(req, 0, sizeof(cp_time));
  121                 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
  122         }
  123         return error;
  124 }
  125 
  126 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
  127     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
  128 
  129 static long empty[CPUSTATES];
  130 
  131 static int
  132 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
  133 {
  134         struct pcpu *pcpu;
  135         int error;
  136         int c;
  137         long *cp_time;
  138 #ifdef SCTL_MASK32
  139         unsigned int cp_time32[CPUSTATES];
  140         int i;
  141 #endif
  142 
  143         if (!req->oldptr) {
  144 #ifdef SCTL_MASK32
  145                 if (req->flags & SCTL_MASK32)
  146                         return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
  147                 else
  148 #endif
  149                         return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
  150         }
  151         for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
  152                 if (!CPU_ABSENT(c)) {
  153                         pcpu = pcpu_find(c);
  154                         cp_time = pcpu->pc_cp_time;
  155                 } else {
  156                         cp_time = empty;
  157                 }
  158 #ifdef SCTL_MASK32
  159                 if (req->flags & SCTL_MASK32) {
  160                         for (i = 0; i < CPUSTATES; i++)
  161                                 cp_time32[i] = (unsigned int)cp_time[i];
  162                         error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
  163                 } else
  164 #endif
  165                         error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
  166         }
  167         return error;
  168 }
  169 
  170 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
  171     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
  172 
  173 #ifdef DEADLKRES
  174 static const char *blessed[] = {
  175         "getblk",
  176         "so_snd_sx",
  177         "so_rcv_sx",
  178         NULL
  179 };
  180 static int slptime_threshold = 1800;
  181 static int blktime_threshold = 900;
  182 static int sleepfreq = 3;
  183 
  184 static void
  185 deadlkres(void)
  186 {
  187         struct proc *p;
  188         struct thread *td;
  189         void *wchan;
  190         int blkticks, i, slpticks, slptype, tryl, tticks;
  191 
  192         tryl = 0;
  193         for (;;) {
  194                 blkticks = blktime_threshold * hz;
  195                 slpticks = slptime_threshold * hz;
  196 
  197                 /*
  198                  * Avoid to sleep on the sx_lock in order to avoid a possible
  199                  * priority inversion problem leading to starvation.
  200                  * If the lock can't be held after 100 tries, panic.
  201                  */
  202                 if (!sx_try_slock(&allproc_lock)) {
  203                         if (tryl > 100)
  204                 panic("%s: possible deadlock detected on allproc_lock\n",
  205                                     __func__);
  206                         tryl++;
  207                         pause("allproc", sleepfreq * hz);
  208                         continue;
  209                 }
  210                 tryl = 0;
  211                 FOREACH_PROC_IN_SYSTEM(p) {
  212                         PROC_LOCK(p);
  213                         if (p->p_state == PRS_NEW) {
  214                                 PROC_UNLOCK(p);
  215                                 continue;
  216                         }
  217                         FOREACH_THREAD_IN_PROC(p, td) {
  218 
  219                                 /*
  220                                  * Once a thread is found in "interesting"
  221                                  * state a possible ticks wrap-up needs to be
  222                                  * checked.
  223                                  */
  224                                 thread_lock(td);
  225                                 if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
  226 
  227                                         /*
  228                                          * The thread should be blocked on a
  229                                          * turnstile, simply check if the
  230                                          * turnstile channel is in good state.
  231                                          */
  232                                         MPASS(td->td_blocked != NULL);
  233 
  234                                         tticks = ticks - td->td_blktick;
  235                                         thread_unlock(td);
  236                                         if (tticks > blkticks) {
  237 
  238                                                 /*
  239                                                  * Accordingly with provided
  240                                                  * thresholds, this thread is
  241                                                  * stuck for too long on a
  242                                                  * turnstile.
  243                                                  */
  244                                                 PROC_UNLOCK(p);
  245                                                 sx_sunlock(&allproc_lock);
  246         panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
  247                                                     __func__, td, tticks);
  248                                         }
  249                                 } else if (TD_IS_SLEEPING(td) &&
  250                                     TD_ON_SLEEPQ(td) &&
  251                                     ticks < td->td_blktick) {
  252 
  253                                         /*
  254                                          * Check if the thread is sleeping on a
  255                                          * lock, otherwise skip the check.
  256                                          * Drop the thread lock in order to
  257                                          * avoid a LOR with the sleepqueue
  258                                          * spinlock.
  259                                          */
  260                                         wchan = td->td_wchan;
  261                                         tticks = ticks - td->td_slptick;
  262                                         thread_unlock(td);
  263                                         slptype = sleepq_type(wchan);
  264                                         if ((slptype == SLEEPQ_SX ||
  265                                             slptype == SLEEPQ_LK) &&
  266                                             tticks > slpticks) {
  267 
  268                                                 /*
  269                                                  * Accordingly with provided
  270                                                  * thresholds, this thread is
  271                                                  * stuck for too long on a
  272                                                  * sleepqueue.
  273                                                  * However, being on a
  274                                                  * sleepqueue, we might still
  275                                                  * check for the blessed
  276                                                  * list.
  277                                                  */
  278                                                 tryl = 0;
  279                                                 for (i = 0; blessed[i] != NULL;
  280                                                     i++) {
  281                                                         if (!strcmp(blessed[i],
  282                                                             td->td_wmesg)) {
  283                                                                 tryl = 1;
  284                                                                 break;
  285                                                         }
  286                                                 }
  287                                                 if (tryl != 0) {
  288                                                         tryl = 0;
  289                                                         continue;
  290                                                 }
  291                                                 PROC_UNLOCK(p);
  292                                                 sx_sunlock(&allproc_lock);
  293         panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
  294                                                     __func__, td, tticks);
  295                                         }
  296                                 } else
  297                                         thread_unlock(td);
  298                         }
  299                         PROC_UNLOCK(p);
  300                 }
  301                 sx_sunlock(&allproc_lock);
  302 
  303                 /* Sleep for sleepfreq seconds. */
  304                 pause("-", sleepfreq * hz);
  305         }
  306 }
  307 
  308 static struct kthread_desc deadlkres_kd = {
  309         "deadlkres",
  310         deadlkres,
  311         (struct thread **)NULL
  312 };
  313 
  314 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
  315 
  316 static SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0,
  317     "Deadlock resolver");
  318 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
  319     &slptime_threshold, 0,
  320     "Number of seconds within is valid to sleep on a sleepqueue");
  321 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
  322     &blktime_threshold, 0,
  323     "Number of seconds within is valid to block on a turnstile");
  324 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
  325     "Number of seconds between any deadlock resolver thread run");
  326 #endif  /* DEADLKRES */
  327 
  328 void
  329 read_cpu_time(long *cp_time)
  330 {
  331         struct pcpu *pc;
  332         int i, j;
  333 
  334         /* Sum up global cp_time[]. */
  335         bzero(cp_time, sizeof(long) * CPUSTATES);
  336         CPU_FOREACH(i) {
  337                 pc = pcpu_find(i);
  338                 for (j = 0; j < CPUSTATES; j++)
  339                         cp_time[j] += pc->pc_cp_time[j];
  340         }
  341 }
  342 
  343 #ifdef SW_WATCHDOG
  344 #include <sys/watchdog.h>
  345 
  346 static int watchdog_ticks;
  347 static int watchdog_enabled;
  348 static void watchdog_fire(void);
  349 static void watchdog_config(void *, u_int, int *);
  350 #endif /* SW_WATCHDOG */
  351 
  352 /*
  353  * Clock handling routines.
  354  *
  355  * This code is written to operate with two timers that run independently of
  356  * each other.
  357  *
  358  * The main timer, running hz times per second, is used to trigger interval
  359  * timers, timeouts and rescheduling as needed.
  360  *
  361  * The second timer handles kernel and user profiling,
  362  * and does resource use estimation.  If the second timer is programmable,
  363  * it is randomized to avoid aliasing between the two clocks.  For example,
  364  * the randomization prevents an adversary from always giving up the cpu
  365  * just before its quantum expires.  Otherwise, it would never accumulate
  366  * cpu ticks.  The mean frequency of the second timer is stathz.
  367  *
  368  * If no second timer exists, stathz will be zero; in this case we drive
  369  * profiling and statistics off the main clock.  This WILL NOT be accurate;
  370  * do not do it unless absolutely necessary.
  371  *
  372  * The statistics clock may (or may not) be run at a higher rate while
  373  * profiling.  This profile clock runs at profhz.  We require that profhz
  374  * be an integral multiple of stathz.
  375  *
  376  * If the statistics clock is running fast, it must be divided by the ratio
  377  * profhz/stathz for statistics.  (For profiling, every tick counts.)
  378  *
  379  * Time-of-day is maintained using a "timecounter", which may or may
  380  * not be related to the hardware generating the above mentioned
  381  * interrupts.
  382  */
  383 
  384 int     stathz;
  385 int     profhz;
  386 int     profprocs;
  387 volatile int    ticks;
  388 int     psratio;
  389 
  390 static DPCPU_DEFINE(int, pcputicks);    /* Per-CPU version of ticks. */
  391 static int global_hardclock_run = 0;
  392 
  393 /*
  394  * Initialize clock frequencies and start both clocks running.
  395  */
  396 /* ARGSUSED*/
  397 static void
  398 initclocks(dummy)
  399         void *dummy;
  400 {
  401         register int i;
  402 
  403         /*
  404          * Set divisors to 1 (normal case) and let the machine-specific
  405          * code do its bit.
  406          */
  407         mtx_init(&time_lock, "time lock", NULL, MTX_DEF);
  408         cpu_initclocks();
  409 
  410         /*
  411          * Compute profhz/stathz, and fix profhz if needed.
  412          */
  413         i = stathz ? stathz : hz;
  414         if (profhz == 0)
  415                 profhz = i;
  416         psratio = profhz / i;
  417 #ifdef SW_WATCHDOG
  418         EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
  419 #endif
  420 }
  421 
  422 /*
  423  * Each time the real-time timer fires, this function is called on all CPUs.
  424  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
  425  * the other CPUs in the system need to call this function.
  426  */
  427 void
  428 hardclock_cpu(int usermode)
  429 {
  430         struct pstats *pstats;
  431         struct thread *td = curthread;
  432         struct proc *p = td->td_proc;
  433         int flags;
  434 
  435         /*
  436          * Run current process's virtual and profile time, as needed.
  437          */
  438         pstats = p->p_stats;
  439         flags = 0;
  440         if (usermode &&
  441             timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
  442                 PROC_SLOCK(p);
  443                 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
  444                         flags |= TDF_ALRMPEND | TDF_ASTPENDING;
  445                 PROC_SUNLOCK(p);
  446         }
  447         if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
  448                 PROC_SLOCK(p);
  449                 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
  450                         flags |= TDF_PROFPEND | TDF_ASTPENDING;
  451                 PROC_SUNLOCK(p);
  452         }
  453         thread_lock(td);
  454         sched_tick(1);
  455         td->td_flags |= flags;
  456         thread_unlock(td);
  457 
  458 #ifdef HWPMC_HOOKS
  459         if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
  460                 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
  461         if (td->td_intr_frame != NULL)
  462                 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
  463 #endif
  464         callout_tick();
  465 }
  466 
  467 /*
  468  * The real-time timer, interrupting hz times per second.
  469  */
  470 void
  471 hardclock(int usermode, uintfptr_t pc)
  472 {
  473 
  474         atomic_add_int(&ticks, 1);
  475         hardclock_cpu(usermode);
  476         tc_ticktock(1);
  477         cpu_tick_calibration();
  478         /*
  479          * If no separate statistics clock is available, run it from here.
  480          *
  481          * XXX: this only works for UP
  482          */
  483         if (stathz == 0) {
  484                 profclock(usermode, pc);
  485                 statclock(usermode);
  486         }
  487 #ifdef DEVICE_POLLING
  488         hardclock_device_poll();        /* this is very short and quick */
  489 #endif /* DEVICE_POLLING */
  490 #ifdef SW_WATCHDOG
  491         if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
  492                 watchdog_fire();
  493 #endif /* SW_WATCHDOG */
  494 }
  495 
  496 void
  497 hardclock_cnt(int cnt, int usermode)
  498 {
  499         struct pstats *pstats;
  500         struct thread *td = curthread;
  501         struct proc *p = td->td_proc;
  502         int *t = DPCPU_PTR(pcputicks);
  503         int flags, global, newticks;
  504 #ifdef SW_WATCHDOG
  505         int i;
  506 #endif /* SW_WATCHDOG */
  507 
  508         /*
  509          * Update per-CPU and possibly global ticks values.
  510          */
  511         *t += cnt;
  512         do {
  513                 global = ticks;
  514                 newticks = *t - global;
  515                 if (newticks <= 0) {
  516                         if (newticks < -1)
  517                                 *t = global - 1;
  518                         newticks = 0;
  519                         break;
  520                 }
  521         } while (!atomic_cmpset_int(&ticks, global, *t));
  522 
  523         /*
  524          * Run current process's virtual and profile time, as needed.
  525          */
  526         pstats = p->p_stats;
  527         flags = 0;
  528         if (usermode &&
  529             timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
  530                 PROC_SLOCK(p);
  531                 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL],
  532                     tick * cnt) == 0)
  533                         flags |= TDF_ALRMPEND | TDF_ASTPENDING;
  534                 PROC_SUNLOCK(p);
  535         }
  536         if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
  537                 PROC_SLOCK(p);
  538                 if (itimerdecr(&pstats->p_timer[ITIMER_PROF],
  539                     tick * cnt) == 0)
  540                         flags |= TDF_PROFPEND | TDF_ASTPENDING;
  541                 PROC_SUNLOCK(p);
  542         }
  543         thread_lock(td);
  544         sched_tick(cnt);
  545         td->td_flags |= flags;
  546         thread_unlock(td);
  547 
  548 #ifdef  HWPMC_HOOKS
  549         if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
  550                 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
  551         if (td->td_intr_frame != NULL)
  552                 PMC_SOFT_CALL_TF( , , clock, hard, td->td_intr_frame);
  553 #endif
  554         callout_tick();
  555         /* We are in charge to handle this tick duty. */
  556         if (newticks > 0) {
  557                 /* Dangerous and no need to call these things concurrently. */
  558                 if (atomic_cmpset_acq_int(&global_hardclock_run, 0, 1)) {
  559                         tc_ticktock(newticks);
  560 #ifdef DEVICE_POLLING
  561                         /* This is very short and quick. */
  562                         hardclock_device_poll();
  563 #endif /* DEVICE_POLLING */
  564                         atomic_store_rel_int(&global_hardclock_run, 0);
  565                 }
  566 #ifdef SW_WATCHDOG
  567                 if (watchdog_enabled > 0) {
  568                         i = atomic_fetchadd_int(&watchdog_ticks, -newticks);
  569                         if (i > 0 && i <= newticks)
  570                                 watchdog_fire();
  571                 }
  572 #endif /* SW_WATCHDOG */
  573         }
  574         if (curcpu == CPU_FIRST())
  575                 cpu_tick_calibration();
  576 }
  577 
  578 void
  579 hardclock_sync(int cpu)
  580 {
  581         int     *t = DPCPU_ID_PTR(cpu, pcputicks);
  582 
  583         *t = ticks;
  584 }
  585 
  586 /*
  587  * Compute number of ticks in the specified amount of time.
  588  */
  589 int
  590 tvtohz(tv)
  591         struct timeval *tv;
  592 {
  593         register unsigned long ticks;
  594         register long sec, usec;
  595 
  596         /*
  597          * If the number of usecs in the whole seconds part of the time
  598          * difference fits in a long, then the total number of usecs will
  599          * fit in an unsigned long.  Compute the total and convert it to
  600          * ticks, rounding up and adding 1 to allow for the current tick
  601          * to expire.  Rounding also depends on unsigned long arithmetic
  602          * to avoid overflow.
  603          *
  604          * Otherwise, if the number of ticks in the whole seconds part of
  605          * the time difference fits in a long, then convert the parts to
  606          * ticks separately and add, using similar rounding methods and
  607          * overflow avoidance.  This method would work in the previous
  608          * case but it is slightly slower and assumes that hz is integral.
  609          *
  610          * Otherwise, round the time difference down to the maximum
  611          * representable value.
  612          *
  613          * If ints have 32 bits, then the maximum value for any timeout in
  614          * 10ms ticks is 248 days.
  615          */
  616         sec = tv->tv_sec;
  617         usec = tv->tv_usec;
  618         if (usec < 0) {
  619                 sec--;
  620                 usec += 1000000;
  621         }
  622         if (sec < 0) {
  623 #ifdef DIAGNOSTIC
  624                 if (usec > 0) {
  625                         sec++;
  626                         usec -= 1000000;
  627                 }
  628                 printf("tvotohz: negative time difference %ld sec %ld usec\n",
  629                        sec, usec);
  630 #endif
  631                 ticks = 1;
  632         } else if (sec <= LONG_MAX / 1000000)
  633                 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
  634                         / tick + 1;
  635         else if (sec <= LONG_MAX / hz)
  636                 ticks = sec * hz
  637                         + ((unsigned long)usec + (tick - 1)) / tick + 1;
  638         else
  639                 ticks = LONG_MAX;
  640         if (ticks > INT_MAX)
  641                 ticks = INT_MAX;
  642         return ((int)ticks);
  643 }
  644 
  645 /*
  646  * Start profiling on a process.
  647  *
  648  * Kernel profiling passes proc0 which never exits and hence
  649  * keeps the profile clock running constantly.
  650  */
  651 void
  652 startprofclock(p)
  653         register struct proc *p;
  654 {
  655 
  656         PROC_LOCK_ASSERT(p, MA_OWNED);
  657         if (p->p_flag & P_STOPPROF)
  658                 return;
  659         if ((p->p_flag & P_PROFIL) == 0) {
  660                 p->p_flag |= P_PROFIL;
  661                 mtx_lock(&time_lock);
  662                 if (++profprocs == 1)
  663                         cpu_startprofclock();
  664                 mtx_unlock(&time_lock);
  665         }
  666 }
  667 
  668 /*
  669  * Stop profiling on a process.
  670  */
  671 void
  672 stopprofclock(p)
  673         register struct proc *p;
  674 {
  675 
  676         PROC_LOCK_ASSERT(p, MA_OWNED);
  677         if (p->p_flag & P_PROFIL) {
  678                 if (p->p_profthreads != 0) {
  679                         p->p_flag |= P_STOPPROF;
  680                         while (p->p_profthreads != 0)
  681                                 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
  682                                     "stopprof", 0);
  683                         p->p_flag &= ~P_STOPPROF;
  684                 }
  685                 if ((p->p_flag & P_PROFIL) == 0)
  686                         return;
  687                 p->p_flag &= ~P_PROFIL;
  688                 mtx_lock(&time_lock);
  689                 if (--profprocs == 0)
  690                         cpu_stopprofclock();
  691                 mtx_unlock(&time_lock);
  692         }
  693 }
  694 
  695 /*
  696  * Statistics clock.  Updates rusage information and calls the scheduler
  697  * to adjust priorities of the active thread.
  698  *
  699  * This should be called by all active processors.
  700  */
  701 void
  702 statclock(int usermode)
  703 {
  704 
  705         statclock_cnt(1, usermode);
  706 }
  707 
  708 void
  709 statclock_cnt(int cnt, int usermode)
  710 {
  711         struct rusage *ru;
  712         struct vmspace *vm;
  713         struct thread *td;
  714         struct proc *p;
  715         long rss;
  716         long *cp_time;
  717 
  718         td = curthread;
  719         p = td->td_proc;
  720 
  721         cp_time = (long *)PCPU_PTR(cp_time);
  722         if (usermode) {
  723                 /*
  724                  * Charge the time as appropriate.
  725                  */
  726                 td->td_uticks += cnt;
  727                 if (p->p_nice > NZERO)
  728                         cp_time[CP_NICE] += cnt;
  729                 else
  730                         cp_time[CP_USER] += cnt;
  731         } else {
  732                 /*
  733                  * Came from kernel mode, so we were:
  734                  * - handling an interrupt,
  735                  * - doing syscall or trap work on behalf of the current
  736                  *   user process, or
  737                  * - spinning in the idle loop.
  738                  * Whichever it is, charge the time as appropriate.
  739                  * Note that we charge interrupts to the current process,
  740                  * regardless of whether they are ``for'' that process,
  741                  * so that we know how much of its real time was spent
  742                  * in ``non-process'' (i.e., interrupt) work.
  743                  */
  744                 if ((td->td_pflags & TDP_ITHREAD) ||
  745                     td->td_intr_nesting_level >= 2) {
  746                         td->td_iticks += cnt;
  747                         cp_time[CP_INTR] += cnt;
  748                 } else {
  749                         td->td_pticks += cnt;
  750                         td->td_sticks += cnt;
  751                         if (!TD_IS_IDLETHREAD(td))
  752                                 cp_time[CP_SYS] += cnt;
  753                         else
  754                                 cp_time[CP_IDLE] += cnt;
  755                 }
  756         }
  757 
  758         /* Update resource usage integrals and maximums. */
  759         MPASS(p->p_vmspace != NULL);
  760         vm = p->p_vmspace;
  761         ru = &td->td_ru;
  762         ru->ru_ixrss += pgtok(vm->vm_tsize) * cnt;
  763         ru->ru_idrss += pgtok(vm->vm_dsize) * cnt;
  764         ru->ru_isrss += pgtok(vm->vm_ssize) * cnt;
  765         rss = pgtok(vmspace_resident_count(vm));
  766         if (ru->ru_maxrss < rss)
  767                 ru->ru_maxrss = rss;
  768         KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
  769             "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
  770         SDT_PROBE2(sched, , , tick, td, td->td_proc);
  771         thread_lock_flags(td, MTX_QUIET);
  772         for ( ; cnt > 0; cnt--)
  773                 sched_clock(td);
  774         thread_unlock(td);
  775 #ifdef HWPMC_HOOKS
  776         if (td->td_intr_frame != NULL)
  777                 PMC_SOFT_CALL_TF( , , clock, stat, td->td_intr_frame);
  778 #endif
  779 }
  780 
  781 void
  782 profclock(int usermode, uintfptr_t pc)
  783 {
  784 
  785         profclock_cnt(1, usermode, pc);
  786 }
  787 
  788 void
  789 profclock_cnt(int cnt, int usermode, uintfptr_t pc)
  790 {
  791         struct thread *td;
  792 #ifdef GPROF
  793         struct gmonparam *g;
  794         uintfptr_t i;
  795 #endif
  796 
  797         td = curthread;
  798         if (usermode) {
  799                 /*
  800                  * Came from user mode; CPU was in user state.
  801                  * If this process is being profiled, record the tick.
  802                  * if there is no related user location yet, don't
  803                  * bother trying to count it.
  804                  */
  805                 if (td->td_proc->p_flag & P_PROFIL)
  806                         addupc_intr(td, pc, cnt);
  807         }
  808 #ifdef GPROF
  809         else {
  810                 /*
  811                  * Kernel statistics are just like addupc_intr, only easier.
  812                  */
  813                 g = &_gmonparam;
  814                 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
  815                         i = PC_TO_I(g, pc);
  816                         if (i < g->textsize) {
  817                                 KCOUNT(g, i) += cnt;
  818                         }
  819                 }
  820         }
  821 #endif
  822 #ifdef HWPMC_HOOKS
  823         if (td->td_intr_frame != NULL)
  824                 PMC_SOFT_CALL_TF( , , clock, prof, td->td_intr_frame);
  825 #endif
  826 }
  827 
  828 /*
  829  * Return information about system clocks.
  830  */
  831 static int
  832 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
  833 {
  834         struct clockinfo clkinfo;
  835         /*
  836          * Construct clockinfo structure.
  837          */
  838         bzero(&clkinfo, sizeof(clkinfo));
  839         clkinfo.hz = hz;
  840         clkinfo.tick = tick;
  841         clkinfo.profhz = profhz;
  842         clkinfo.stathz = stathz ? stathz : hz;
  843         return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
  844 }
  845 
  846 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
  847         CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
  848         0, 0, sysctl_kern_clockrate, "S,clockinfo",
  849         "Rate and period of various kernel clocks");
  850 
  851 #ifdef SW_WATCHDOG
  852 
  853 static void
  854 watchdog_config(void *unused __unused, u_int cmd, int *error)
  855 {
  856         u_int u;
  857 
  858         u = cmd & WD_INTERVAL;
  859         if (u >= WD_TO_1SEC) {
  860                 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
  861                 watchdog_enabled = 1;
  862                 *error = 0;
  863         } else {
  864                 watchdog_enabled = 0;
  865         }
  866 }
  867 
  868 /*
  869  * Handle a watchdog timeout by dumping interrupt information and
  870  * then either dropping to DDB or panicking.
  871  */
  872 static void
  873 watchdog_fire(void)
  874 {
  875         int nintr;
  876         uint64_t inttotal;
  877         u_long *curintr;
  878         char *curname;
  879 
  880         curintr = intrcnt;
  881         curname = intrnames;
  882         inttotal = 0;
  883         nintr = sintrcnt / sizeof(u_long);
  884 
  885         printf("interrupt                   total\n");
  886         while (--nintr >= 0) {
  887                 if (*curintr)
  888                         printf("%-12s %20lu\n", curname, *curintr);
  889                 curname += strlen(curname) + 1;
  890                 inttotal += *curintr++;
  891         }
  892         printf("Total        %20ju\n", (uintmax_t)inttotal);
  893 
  894 #if defined(KDB) && !defined(KDB_UNATTENDED)
  895         kdb_backtrace();
  896         kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
  897 #else
  898         panic("watchdog timeout");
  899 #endif
  900 }
  901 
  902 #endif /* SW_WATCHDOG */

Cache object: 286e31d24452f9f2cf8c8e2b7ccab063


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.