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/8.2/sys/kern/kern_clock.c 215044 2010-11-09 18:46:09Z jhb $");
   39 
   40 #include "opt_kdb.h"
   41 #include "opt_device_polling.h"
   42 #include "opt_hwpmc_hooks.h"
   43 #include "opt_ntp.h"
   44 #include "opt_watchdog.h"
   45 
   46 #include <sys/param.h>
   47 #include <sys/systm.h>
   48 #include <sys/callout.h>
   49 #include <sys/kdb.h>
   50 #include <sys/kernel.h>
   51 #include <sys/kthread.h>
   52 #include <sys/ktr.h>
   53 #include <sys/lock.h>
   54 #include <sys/mutex.h>
   55 #include <sys/proc.h>
   56 #include <sys/resource.h>
   57 #include <sys/resourcevar.h>
   58 #include <sys/sched.h>
   59 #include <sys/signalvar.h>
   60 #include <sys/sleepqueue.h>
   61 #include <sys/smp.h>
   62 #include <vm/vm.h>
   63 #include <vm/pmap.h>
   64 #include <vm/vm_map.h>
   65 #include <sys/sysctl.h>
   66 #include <sys/bus.h>
   67 #include <sys/interrupt.h>
   68 #include <sys/limits.h>
   69 #include <sys/timetc.h>
   70 
   71 #ifdef GPROF
   72 #include <sys/gmon.h>
   73 #endif
   74 
   75 #ifdef HWPMC_HOOKS
   76 #include <sys/pmckern.h>
   77 #endif
   78 
   79 #ifdef DEVICE_POLLING
   80 extern void hardclock_device_poll(void);
   81 #endif /* DEVICE_POLLING */
   82 
   83 static void initclocks(void *dummy);
   84 SYSINIT(clocks, SI_SUB_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);
   85 
   86 /* Spin-lock protecting profiling statistics. */
   87 static struct mtx time_lock;
   88 
   89 static int
   90 sysctl_kern_cp_time(SYSCTL_HANDLER_ARGS)
   91 {
   92         int error;
   93         long cp_time[CPUSTATES];
   94 #ifdef SCTL_MASK32
   95         int i;
   96         unsigned int cp_time32[CPUSTATES];
   97 #endif
   98 
   99         read_cpu_time(cp_time);
  100 #ifdef SCTL_MASK32
  101         if (req->flags & SCTL_MASK32) {
  102                 if (!req->oldptr)
  103                         return SYSCTL_OUT(req, 0, sizeof(cp_time32));
  104                 for (i = 0; i < CPUSTATES; i++)
  105                         cp_time32[i] = (unsigned int)cp_time[i];
  106                 error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
  107         } else
  108 #endif
  109         {
  110                 if (!req->oldptr)
  111                         return SYSCTL_OUT(req, 0, sizeof(cp_time));
  112                 error = SYSCTL_OUT(req, cp_time, sizeof(cp_time));
  113         }
  114         return error;
  115 }
  116 
  117 SYSCTL_PROC(_kern, OID_AUTO, cp_time, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
  118     0,0, sysctl_kern_cp_time, "LU", "CPU time statistics");
  119 
  120 static long empty[CPUSTATES];
  121 
  122 static int
  123 sysctl_kern_cp_times(SYSCTL_HANDLER_ARGS)
  124 {
  125         struct pcpu *pcpu;
  126         int error;
  127         int c;
  128         long *cp_time;
  129 #ifdef SCTL_MASK32
  130         unsigned int cp_time32[CPUSTATES];
  131         int i;
  132 #endif
  133 
  134         if (!req->oldptr) {
  135 #ifdef SCTL_MASK32
  136                 if (req->flags & SCTL_MASK32)
  137                         return SYSCTL_OUT(req, 0, sizeof(cp_time32) * (mp_maxid + 1));
  138                 else
  139 #endif
  140                         return SYSCTL_OUT(req, 0, sizeof(long) * CPUSTATES * (mp_maxid + 1));
  141         }
  142         for (error = 0, c = 0; error == 0 && c <= mp_maxid; c++) {
  143                 if (!CPU_ABSENT(c)) {
  144                         pcpu = pcpu_find(c);
  145                         cp_time = pcpu->pc_cp_time;
  146                 } else {
  147                         cp_time = empty;
  148                 }
  149 #ifdef SCTL_MASK32
  150                 if (req->flags & SCTL_MASK32) {
  151                         for (i = 0; i < CPUSTATES; i++)
  152                                 cp_time32[i] = (unsigned int)cp_time[i];
  153                         error = SYSCTL_OUT(req, cp_time32, sizeof(cp_time32));
  154                 } else
  155 #endif
  156                         error = SYSCTL_OUT(req, cp_time, sizeof(long) * CPUSTATES);
  157         }
  158         return error;
  159 }
  160 
  161 SYSCTL_PROC(_kern, OID_AUTO, cp_times, CTLTYPE_LONG|CTLFLAG_RD|CTLFLAG_MPSAFE,
  162     0,0, sysctl_kern_cp_times, "LU", "per-CPU time statistics");
  163 
  164 #ifdef DEADLKRES
  165 static const char *blessed[] = {
  166         "getblk",
  167         "so_snd_sx",
  168         "so_rcv_sx",
  169         NULL
  170 };
  171 static int slptime_threshold = 1800;
  172 static int blktime_threshold = 900;
  173 static int sleepfreq = 3;
  174 
  175 static void
  176 deadlkres(void)
  177 {
  178         struct proc *p;
  179         struct thread *td;
  180         void *wchan;
  181         int blkticks, i, slpticks, slptype, tryl, tticks;
  182 
  183         tryl = 0;
  184         for (;;) {
  185                 blkticks = blktime_threshold * hz;
  186                 slpticks = slptime_threshold * hz;
  187 
  188                 /*
  189                  * Avoid to sleep on the sx_lock in order to avoid a possible
  190                  * priority inversion problem leading to starvation.
  191                  * If the lock can't be held after 100 tries, panic.
  192                  */
  193                 if (!sx_try_slock(&allproc_lock)) {
  194                         if (tryl > 100)
  195                 panic("%s: possible deadlock detected on allproc_lock\n",
  196                                     __func__);
  197                         tryl++;
  198                         pause("allproc", sleepfreq * hz);
  199                         continue;
  200                 }
  201                 tryl = 0;
  202                 FOREACH_PROC_IN_SYSTEM(p) {
  203                         PROC_LOCK(p);
  204                         FOREACH_THREAD_IN_PROC(p, td) {
  205 
  206                                 /*
  207                                  * Once a thread is found in "interesting"
  208                                  * state a possible ticks wrap-up needs to be
  209                                  * checked.
  210                                  */
  211                                 thread_lock(td);
  212                                 if (TD_ON_LOCK(td) && ticks < td->td_blktick) {
  213 
  214                                         /*
  215                                          * The thread should be blocked on a
  216                                          * turnstile, simply check if the
  217                                          * turnstile channel is in good state.
  218                                          */
  219                                         MPASS(td->td_blocked != NULL);
  220 
  221                                         tticks = ticks - td->td_blktick;
  222                                         thread_unlock(td);
  223                                         if (tticks > blkticks) {
  224 
  225                                                 /*
  226                                                  * Accordingly with provided
  227                                                  * thresholds, this thread is
  228                                                  * stuck for too long on a
  229                                                  * turnstile.
  230                                                  */
  231                                                 PROC_UNLOCK(p);
  232                                                 sx_sunlock(&allproc_lock);
  233         panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
  234                                                     __func__, td, tticks);
  235                                         }
  236                                 } else if (TD_IS_SLEEPING(td) &&
  237                                     TD_ON_SLEEPQ(td) &&
  238                                     ticks < td->td_blktick) {
  239 
  240                                         /*
  241                                          * Check if the thread is sleeping on a
  242                                          * lock, otherwise skip the check.
  243                                          * Drop the thread lock in order to
  244                                          * avoid a LOR with the sleepqueue
  245                                          * spinlock.
  246                                          */
  247                                         wchan = td->td_wchan;
  248                                         tticks = ticks - td->td_slptick;
  249                                         thread_unlock(td);
  250                                         slptype = sleepq_type(wchan);
  251                                         if ((slptype == SLEEPQ_SX ||
  252                                             slptype == SLEEPQ_LK) &&
  253                                             tticks > slpticks) {
  254 
  255                                                 /*
  256                                                  * Accordingly with provided
  257                                                  * thresholds, this thread is
  258                                                  * stuck for too long on a
  259                                                  * sleepqueue.
  260                                                  * However, being on a
  261                                                  * sleepqueue, we might still
  262                                                  * check for the blessed
  263                                                  * list.
  264                                                  */
  265                                                 tryl = 0;
  266                                                 for (i = 0; blessed[i] != NULL;
  267                                                     i++) {
  268                                                         if (!strcmp(blessed[i],
  269                                                             td->td_wmesg)) {
  270                                                                 tryl = 1;
  271                                                                 break;
  272                                                         }
  273                                                 }
  274                                                 if (tryl != 0) {
  275                                                         tryl = 0;
  276                                                         continue;
  277                                                 }
  278                                                 PROC_UNLOCK(p);
  279                                                 sx_sunlock(&allproc_lock);
  280         panic("%s: possible deadlock detected for %p, blocked for %d ticks\n",
  281                                                     __func__, td, tticks);
  282                                         }
  283                                 } else
  284                                         thread_unlock(td);
  285                         }
  286                         PROC_UNLOCK(p);
  287                 }
  288                 sx_sunlock(&allproc_lock);
  289 
  290                 /* Sleep for sleepfreq seconds. */
  291                 pause("-", sleepfreq * hz);
  292         }
  293 }
  294 
  295 static struct kthread_desc deadlkres_kd = {
  296         "deadlkres",
  297         deadlkres,
  298         (struct thread **)NULL
  299 };
  300 
  301 SYSINIT(deadlkres, SI_SUB_CLOCKS, SI_ORDER_ANY, kthread_start, &deadlkres_kd);
  302 
  303 SYSCTL_NODE(_debug, OID_AUTO, deadlkres, CTLFLAG_RW, 0, "Deadlock resolver");
  304 SYSCTL_INT(_debug_deadlkres, OID_AUTO, slptime_threshold, CTLFLAG_RW,
  305     &slptime_threshold, 0,
  306     "Number of seconds within is valid to sleep on a sleepqueue");
  307 SYSCTL_INT(_debug_deadlkres, OID_AUTO, blktime_threshold, CTLFLAG_RW,
  308     &blktime_threshold, 0,
  309     "Number of seconds within is valid to block on a turnstile");
  310 SYSCTL_INT(_debug_deadlkres, OID_AUTO, sleepfreq, CTLFLAG_RW, &sleepfreq, 0,
  311     "Number of seconds between any deadlock resolver thread run");
  312 #endif  /* DEADLKRES */
  313 
  314 void
  315 read_cpu_time(long *cp_time)
  316 {
  317         struct pcpu *pc;
  318         int i, j;
  319 
  320         /* Sum up global cp_time[]. */
  321         bzero(cp_time, sizeof(long) * CPUSTATES);
  322         for (i = 0; i <= mp_maxid; i++) {
  323                 if (CPU_ABSENT(i))
  324                         continue;
  325                 pc = pcpu_find(i);
  326                 for (j = 0; j < CPUSTATES; j++)
  327                         cp_time[j] += pc->pc_cp_time[j];
  328         }
  329 }
  330 
  331 #ifdef SW_WATCHDOG
  332 #include <sys/watchdog.h>
  333 
  334 static int watchdog_ticks;
  335 static int watchdog_enabled;
  336 static void watchdog_fire(void);
  337 static void watchdog_config(void *, u_int, int *);
  338 #endif /* SW_WATCHDOG */
  339 
  340 /*
  341  * Clock handling routines.
  342  *
  343  * This code is written to operate with two timers that run independently of
  344  * each other.
  345  *
  346  * The main timer, running hz times per second, is used to trigger interval
  347  * timers, timeouts and rescheduling as needed.
  348  *
  349  * The second timer handles kernel and user profiling,
  350  * and does resource use estimation.  If the second timer is programmable,
  351  * it is randomized to avoid aliasing between the two clocks.  For example,
  352  * the randomization prevents an adversary from always giving up the cpu
  353  * just before its quantum expires.  Otherwise, it would never accumulate
  354  * cpu ticks.  The mean frequency of the second timer is stathz.
  355  *
  356  * If no second timer exists, stathz will be zero; in this case we drive
  357  * profiling and statistics off the main clock.  This WILL NOT be accurate;
  358  * do not do it unless absolutely necessary.
  359  *
  360  * The statistics clock may (or may not) be run at a higher rate while
  361  * profiling.  This profile clock runs at profhz.  We require that profhz
  362  * be an integral multiple of stathz.
  363  *
  364  * If the statistics clock is running fast, it must be divided by the ratio
  365  * profhz/stathz for statistics.  (For profiling, every tick counts.)
  366  *
  367  * Time-of-day is maintained using a "timecounter", which may or may
  368  * not be related to the hardware generating the above mentioned
  369  * interrupts.
  370  */
  371 
  372 int     stathz;
  373 int     profhz;
  374 int     profprocs;
  375 int     ticks;
  376 int     psratio;
  377 
  378 /*
  379  * Initialize clock frequencies and start both clocks running.
  380  */
  381 /* ARGSUSED*/
  382 static void
  383 initclocks(dummy)
  384         void *dummy;
  385 {
  386         register int i;
  387 
  388         /*
  389          * Set divisors to 1 (normal case) and let the machine-specific
  390          * code do its bit.
  391          */
  392         mtx_init(&time_lock, "time lock", NULL, MTX_SPIN);
  393         cpu_initclocks();
  394 
  395         /*
  396          * Compute profhz/stathz, and fix profhz if needed.
  397          */
  398         i = stathz ? stathz : hz;
  399         if (profhz == 0)
  400                 profhz = i;
  401         psratio = profhz / i;
  402 #ifdef SW_WATCHDOG
  403         EVENTHANDLER_REGISTER(watchdog_list, watchdog_config, NULL, 0);
  404 #endif
  405 }
  406 
  407 /*
  408  * Each time the real-time timer fires, this function is called on all CPUs.
  409  * Note that hardclock() calls hardclock_cpu() for the boot CPU, so only
  410  * the other CPUs in the system need to call this function.
  411  */
  412 void
  413 hardclock_cpu(int usermode)
  414 {
  415         struct pstats *pstats;
  416         struct thread *td = curthread;
  417         struct proc *p = td->td_proc;
  418         int flags;
  419 
  420         /*
  421          * Run current process's virtual and profile time, as needed.
  422          */
  423         pstats = p->p_stats;
  424         flags = 0;
  425         if (usermode &&
  426             timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
  427                 PROC_SLOCK(p);
  428                 if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
  429                         flags |= TDF_ALRMPEND | TDF_ASTPENDING;
  430                 PROC_SUNLOCK(p);
  431         }
  432         if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
  433                 PROC_SLOCK(p);
  434                 if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
  435                         flags |= TDF_PROFPEND | TDF_ASTPENDING;
  436                 PROC_SUNLOCK(p);
  437         }
  438         thread_lock(td);
  439         sched_tick();
  440         td->td_flags |= flags;
  441         thread_unlock(td);
  442 
  443 #ifdef  HWPMC_HOOKS
  444         if (PMC_CPU_HAS_SAMPLES(PCPU_GET(cpuid)))
  445                 PMC_CALL_HOOK_UNLOCKED(curthread, PMC_FN_DO_SAMPLES, NULL);
  446 #endif
  447         callout_tick();
  448 }
  449 
  450 /*
  451  * The real-time timer, interrupting hz times per second.
  452  */
  453 void
  454 hardclock(int usermode, uintfptr_t pc)
  455 {
  456 
  457         atomic_add_int((volatile int *)&ticks, 1);
  458         hardclock_cpu(usermode);
  459         tc_ticktock();
  460         /*
  461          * If no separate statistics clock is available, run it from here.
  462          *
  463          * XXX: this only works for UP
  464          */
  465         if (stathz == 0) {
  466                 profclock(usermode, pc);
  467                 statclock(usermode);
  468         }
  469 #ifdef DEVICE_POLLING
  470         hardclock_device_poll();        /* this is very short and quick */
  471 #endif /* DEVICE_POLLING */
  472 #ifdef SW_WATCHDOG
  473         if (watchdog_enabled > 0 && --watchdog_ticks <= 0)
  474                 watchdog_fire();
  475 #endif /* SW_WATCHDOG */
  476 }
  477 
  478 /*
  479  * Compute number of ticks in the specified amount of time.
  480  */
  481 int
  482 tvtohz(tv)
  483         struct timeval *tv;
  484 {
  485         register unsigned long ticks;
  486         register long sec, usec;
  487 
  488         /*
  489          * If the number of usecs in the whole seconds part of the time
  490          * difference fits in a long, then the total number of usecs will
  491          * fit in an unsigned long.  Compute the total and convert it to
  492          * ticks, rounding up and adding 1 to allow for the current tick
  493          * to expire.  Rounding also depends on unsigned long arithmetic
  494          * to avoid overflow.
  495          *
  496          * Otherwise, if the number of ticks in the whole seconds part of
  497          * the time difference fits in a long, then convert the parts to
  498          * ticks separately and add, using similar rounding methods and
  499          * overflow avoidance.  This method would work in the previous
  500          * case but it is slightly slower and assumes that hz is integral.
  501          *
  502          * Otherwise, round the time difference down to the maximum
  503          * representable value.
  504          *
  505          * If ints have 32 bits, then the maximum value for any timeout in
  506          * 10ms ticks is 248 days.
  507          */
  508         sec = tv->tv_sec;
  509         usec = tv->tv_usec;
  510         if (usec < 0) {
  511                 sec--;
  512                 usec += 1000000;
  513         }
  514         if (sec < 0) {
  515 #ifdef DIAGNOSTIC
  516                 if (usec > 0) {
  517                         sec++;
  518                         usec -= 1000000;
  519                 }
  520                 printf("tvotohz: negative time difference %ld sec %ld usec\n",
  521                        sec, usec);
  522 #endif
  523                 ticks = 1;
  524         } else if (sec <= LONG_MAX / 1000000)
  525                 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1))
  526                         / tick + 1;
  527         else if (sec <= LONG_MAX / hz)
  528                 ticks = sec * hz
  529                         + ((unsigned long)usec + (tick - 1)) / tick + 1;
  530         else
  531                 ticks = LONG_MAX;
  532         if (ticks > INT_MAX)
  533                 ticks = INT_MAX;
  534         return ((int)ticks);
  535 }
  536 
  537 /*
  538  * Start profiling on a process.
  539  *
  540  * Kernel profiling passes proc0 which never exits and hence
  541  * keeps the profile clock running constantly.
  542  */
  543 void
  544 startprofclock(p)
  545         register struct proc *p;
  546 {
  547 
  548         PROC_LOCK_ASSERT(p, MA_OWNED);
  549         if (p->p_flag & P_STOPPROF)
  550                 return;
  551         if ((p->p_flag & P_PROFIL) == 0) {
  552                 p->p_flag |= P_PROFIL;
  553                 mtx_lock_spin(&time_lock);
  554                 if (++profprocs == 1)
  555                         cpu_startprofclock();
  556                 mtx_unlock_spin(&time_lock);
  557         }
  558 }
  559 
  560 /*
  561  * Stop profiling on a process.
  562  */
  563 void
  564 stopprofclock(p)
  565         register struct proc *p;
  566 {
  567 
  568         PROC_LOCK_ASSERT(p, MA_OWNED);
  569         if (p->p_flag & P_PROFIL) {
  570                 if (p->p_profthreads != 0) {
  571                         p->p_flag |= P_STOPPROF;
  572                         while (p->p_profthreads != 0)
  573                                 msleep(&p->p_profthreads, &p->p_mtx, PPAUSE,
  574                                     "stopprof", 0);
  575                         p->p_flag &= ~P_STOPPROF;
  576                 }
  577                 if ((p->p_flag & P_PROFIL) == 0)
  578                         return;
  579                 p->p_flag &= ~P_PROFIL;
  580                 mtx_lock_spin(&time_lock);
  581                 if (--profprocs == 0)
  582                         cpu_stopprofclock();
  583                 mtx_unlock_spin(&time_lock);
  584         }
  585 }
  586 
  587 /*
  588  * Statistics clock.  Updates rusage information and calls the scheduler
  589  * to adjust priorities of the active thread.
  590  *
  591  * This should be called by all active processors.
  592  */
  593 void
  594 statclock(int usermode)
  595 {
  596         struct rusage *ru;
  597         struct vmspace *vm;
  598         struct thread *td;
  599         struct proc *p;
  600         long rss;
  601         long *cp_time;
  602 
  603         td = curthread;
  604         p = td->td_proc;
  605 
  606         cp_time = (long *)PCPU_PTR(cp_time);
  607         if (usermode) {
  608                 /*
  609                  * Charge the time as appropriate.
  610                  */
  611                 td->td_uticks++;
  612                 if (p->p_nice > NZERO)
  613                         cp_time[CP_NICE]++;
  614                 else
  615                         cp_time[CP_USER]++;
  616         } else {
  617                 /*
  618                  * Came from kernel mode, so we were:
  619                  * - handling an interrupt,
  620                  * - doing syscall or trap work on behalf of the current
  621                  *   user process, or
  622                  * - spinning in the idle loop.
  623                  * Whichever it is, charge the time as appropriate.
  624                  * Note that we charge interrupts to the current process,
  625                  * regardless of whether they are ``for'' that process,
  626                  * so that we know how much of its real time was spent
  627                  * in ``non-process'' (i.e., interrupt) work.
  628                  */
  629                 if ((td->td_pflags & TDP_ITHREAD) ||
  630                     td->td_intr_nesting_level >= 2) {
  631                         td->td_iticks++;
  632                         cp_time[CP_INTR]++;
  633                 } else {
  634                         td->td_pticks++;
  635                         td->td_sticks++;
  636                         if (!TD_IS_IDLETHREAD(td))
  637                                 cp_time[CP_SYS]++;
  638                         else
  639                                 cp_time[CP_IDLE]++;
  640                 }
  641         }
  642 
  643         /* Update resource usage integrals and maximums. */
  644         MPASS(p->p_vmspace != NULL);
  645         vm = p->p_vmspace;
  646         ru = &td->td_ru;
  647         ru->ru_ixrss += pgtok(vm->vm_tsize);
  648         ru->ru_idrss += pgtok(vm->vm_dsize);
  649         ru->ru_isrss += pgtok(vm->vm_ssize);
  650         rss = pgtok(vmspace_resident_count(vm));
  651         if (ru->ru_maxrss < rss)
  652                 ru->ru_maxrss = rss;
  653         KTR_POINT2(KTR_SCHED, "thread", sched_tdname(td), "statclock",
  654             "prio:%d", td->td_priority, "stathz:%d", (stathz)?stathz:hz);
  655         thread_lock_flags(td, MTX_QUIET);
  656         sched_clock(td);
  657         thread_unlock(td);
  658 }
  659 
  660 void
  661 profclock(int usermode, uintfptr_t pc)
  662 {
  663         struct thread *td;
  664 #ifdef GPROF
  665         struct gmonparam *g;
  666         uintfptr_t i;
  667 #endif
  668 
  669         td = curthread;
  670         if (usermode) {
  671                 /*
  672                  * Came from user mode; CPU was in user state.
  673                  * If this process is being profiled, record the tick.
  674                  * if there is no related user location yet, don't
  675                  * bother trying to count it.
  676                  */
  677                 if (td->td_proc->p_flag & P_PROFIL)
  678                         addupc_intr(td, pc, 1);
  679         }
  680 #ifdef GPROF
  681         else {
  682                 /*
  683                  * Kernel statistics are just like addupc_intr, only easier.
  684                  */
  685                 g = &_gmonparam;
  686                 if (g->state == GMON_PROF_ON && pc >= g->lowpc) {
  687                         i = PC_TO_I(g, pc);
  688                         if (i < g->textsize) {
  689                                 KCOUNT(g, i)++;
  690                         }
  691                 }
  692         }
  693 #endif
  694 }
  695 
  696 /*
  697  * Return information about system clocks.
  698  */
  699 static int
  700 sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
  701 {
  702         struct clockinfo clkinfo;
  703         /*
  704          * Construct clockinfo structure.
  705          */
  706         bzero(&clkinfo, sizeof(clkinfo));
  707         clkinfo.hz = hz;
  708         clkinfo.tick = tick;
  709         clkinfo.profhz = profhz;
  710         clkinfo.stathz = stathz ? stathz : hz;
  711         return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
  712 }
  713 
  714 SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate,
  715         CTLTYPE_STRUCT|CTLFLAG_RD|CTLFLAG_MPSAFE,
  716         0, 0, sysctl_kern_clockrate, "S,clockinfo",
  717         "Rate and period of various kernel clocks");
  718 
  719 #ifdef SW_WATCHDOG
  720 
  721 static void
  722 watchdog_config(void *unused __unused, u_int cmd, int *error)
  723 {
  724         u_int u;
  725 
  726         u = cmd & WD_INTERVAL;
  727         if (u >= WD_TO_1SEC) {
  728                 watchdog_ticks = (1 << (u - WD_TO_1SEC)) * hz;
  729                 watchdog_enabled = 1;
  730                 *error = 0;
  731         } else {
  732                 watchdog_enabled = 0;
  733         }
  734 }
  735 
  736 /*
  737  * Handle a watchdog timeout by dumping interrupt information and
  738  * then either dropping to DDB or panicking.
  739  */
  740 static void
  741 watchdog_fire(void)
  742 {
  743         int nintr;
  744         u_int64_t inttotal;
  745         u_long *curintr;
  746         char *curname;
  747 
  748         curintr = intrcnt;
  749         curname = intrnames;
  750         inttotal = 0;
  751         nintr = eintrcnt - intrcnt;
  752 
  753         printf("interrupt                   total\n");
  754         while (--nintr >= 0) {
  755                 if (*curintr)
  756                         printf("%-12s %20lu\n", curname, *curintr);
  757                 curname += strlen(curname) + 1;
  758                 inttotal += *curintr++;
  759         }
  760         printf("Total        %20ju\n", (uintmax_t)inttotal);
  761 
  762 #if defined(KDB) && !defined(KDB_UNATTENDED)
  763         kdb_backtrace();
  764         kdb_enter(KDB_WHY_WATCHDOG, "watchdog timeout");
  765 #else
  766         panic("watchdog timeout");
  767 #endif
  768 }
  769 
  770 #endif /* SW_WATCHDOG */

Cache object: 03d2ec03bce2bd5a82c2f7d0f327a6bc


[ 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.