kern_synch.c

早期freebsd实现

/*-
 * Copyright (c) 1982, 1986, 1990, 1991, 1993
 *	The Regents of the University of California.  All rights reserved.
 * (c) UNIX System Laboratories, Inc.
 * All or some portions of this file are derived from material licensed
 * to the University of California by American Telephone and Telegraph
 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
 * the permission of UNIX System Laboratories, Inc.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	@(#)kern_synch.c	8.6 (Berkeley) 1/21/94
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/kernel.h>
#include <sys/buf.h>
#include <sys/signalvar.h>
#include <sys/resourcevar.h>
#include <sys/vmmeter.h>
#ifdef KTRACE
#include <sys/ktrace.h>
#endif

#include <machine/cpu.h>

u_char	curpriority;		/* usrpri of curproc */
int	lbolt;			/* once a second sleep address */

/*
 * Force switch among equal priority processes every 100ms.
 */
/* ARGSUSED */
void
roundrobin(arg)
	void *arg;
{

	need_resched();
	timeout(roundrobin, NULL, hz / 10);
}

/*
 * Constants for digital decay and forget:
 *	90% of (p_estcpu) usage in 5 * loadav time
 *	95% of (p_pctcpu) usage in 60 seconds (load insensitive)
 *          Note that, as ps(1) mentions, this can let percentages
 *          total over 100% (I've seen 137.9% for 3 processes).
 *
 * Note that hardclock updates p_estcpu and p_cpticks independently.
 *
 * We wish to decay away 90% of p_estcpu in (5 * loadavg) seconds.
 * That is, the system wants to compute a value of decay such
 * that the following for loop:
 * 	for (i = 0; i < (5 * loadavg); i++)
 * 		p_estcpu *= decay;
 * will compute
 * 	p_estcpu *= 0.1;
 * for all values of loadavg:
 *
 * Mathematically this loop can be expressed by saying:
 * 	decay ** (5 * loadavg) ~= .1
 *
 * The system computes decay as:
 * 	decay = (2 * loadavg) / (2 * loadavg + 1)
 *
 * We wish to prove that the system's computation of decay
 * will always fulfill the equation:
 * 	decay ** (5 * loadavg) ~= .1
 *
 * If we compute b as:
 * 	b = 2 * loadavg
 * then
 * 	decay = b / (b + 1)
 *
 * We now need to prove two things:
 *	1) Given factor ** (5 * loadavg) ~= .1, prove factor == b/(b+1)
 *	2) Given b/(b+1) ** power ~= .1, prove power == (5 * loadavg)
 *	
 * Facts:
 *         For x close to zero, exp(x) =~ 1 + x, since
 *              exp(x) = 0! + x**1/1! + x**2/2! + ... .
 *              therefore exp(-1/b) =~ 1 - (1/b) = (b-1)/b.
 *         For x close to zero, ln(1+x) =~ x, since
 *              ln(1+x) = x - x**2/2 + x**3/3 - ...     -1 < x < 1
 *              therefore ln(b/(b+1)) = ln(1 - 1/(b+1)) =~ -1/(b+1).
 *         ln(.1) =~ -2.30
 *
 * Proof of (1):
 *    Solve (factor)**(power) =~ .1 given power (5*loadav):
 *	solving for factor,
 *      ln(factor) =~ (-2.30/5*loadav), or
 *      factor =~ exp(-1/((5/2.30)*loadav)) =~ exp(-1/(2*loadav)) =
 *          exp(-1/b) =~ (b-1)/b =~ b/(b+1).                    QED
 *
 * Proof of (2):
 *    Solve (factor)**(power) =~ .1 given factor == (b/(b+1)):
 *	solving for power,
 *      power*ln(b/(b+1)) =~ -2.30, or
 *      power =~ 2.3 * (b + 1) = 4.6*loadav + 2.3 =~ 5*loadav.  QED
 *
 * Actual power values for the implemented algorithm are as follows:
 *      loadav: 1       2       3       4
 *      power:  5.68    10.32   14.94   19.55
 */

/* calculations for digital decay to forget 90% of usage in 5*loadav sec */
#define	loadfactor(loadav)	(2 * (loadav))
#define	decay_cpu(loadfac, cpu)	(((loadfac) * (cpu)) / ((loadfac) + FSCALE))

/* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */
fixpt_t	ccpu = 0.95122942450071400909 * FSCALE;		/* exp(-1/20) */

/*
 * If `ccpu' is not equal to `exp(-1/20)' and you still want to use the
 * faster/more-accurate formula, you'll have to estimate CCPU_SHIFT below
 * and possibly adjust FSHIFT in "param.h" so that (FSHIFT >= CCPU_SHIFT).
 *
 * To estimate CCPU_SHIFT for exp(-1/20), the following formula was used:
 *	1 - exp(-1/20) ~= 0.0487 ~= 0.0488 == 1 (fixed pt, *11* bits).
 *
 * If you dont want to bother with the faster/more-accurate formula, you
 * can set CCPU_SHIFT to (FSHIFT + 1) which will use a slower/less-accurate
 * (more general) method of calculating the %age of CPU used by a process.
 */
#define	CCPU_SHIFT	11

/*
 * Recompute process priorities, every hz ticks.
 */
/* ARGSUSED */
void
schedcpu(arg)
	void *arg;
{
	register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);
	register struct proc *p;
	register int s;
	register unsigned int newcpu;

	wakeup((caddr_t)&lbolt);
	for (p = (struct proc *)allproc; p != NULL; p = p->p_next) {
		/*
		 * Increment time in/out of memory and sleep time
		 * (if sleeping).  We ignore overflow; with 16-bit int's
		 * (remember them?) overflow takes 45 days.
		 */
		p->p_swtime++;
		if (p->p_stat == SSLEEP || p->p_stat == SSTOP)
			p->p_slptime++;
		p->p_pctcpu = (p->p_pctcpu * ccpu) >> FSHIFT;
		/*
		 * If the process has slept the entire second,
		 * stop recalculating its priority until it wakes up.
		 */
		if (p->p_slptime > 1)
			continue;
		s = splstatclock();	/* prevent state changes */
		/*
		 * p_pctcpu is only for ps.
		 */
#if	(FSHIFT >= CCPU_SHIFT)
		p->p_pctcpu += (hz == 100)?
			((fixpt_t) p->p_cpticks) << (FSHIFT - CCPU_SHIFT):
                	100 * (((fixpt_t) p->p_cpticks)
				<< (FSHIFT - CCPU_SHIFT)) / hz;
#else
		p->p_pctcpu += ((FSCALE - ccpu) *
			(p->p_cpticks * FSCALE / hz)) >> FSHIFT;
#endif
		p->p_cpticks = 0;
		newcpu = (u_int) decay_cpu(loadfac, p->p_estcpu) + p->p_nice;
		p->p_estcpu = min(newcpu, UCHAR_MAX);
		resetpriority(p);
		if (p->p_priority >= PUSER) {
#define	PPQ	(128 / NQS)		/* priorities per queue */
			if ((p != curproc) &&
			    p->p_stat == SRUN &&
			    (p->p_flag & P_INMEM) &&
			    (p->p_priority / PPQ) != (p->p_usrpri / PPQ)) {
				remrq(p);
				p->p_priority = p->p_usrpri;
				setrunqueue(p);
			} else
				p->p_priority = p->p_usrpri;
		}
		splx(s);
	}
	vmmeter();
	if (bclnlist != NULL)
		wakeup((caddr_t)pageproc);
	timeout(schedcpu, (void *)0, hz);
}

/*
 * Recalculate the priority of a process after it has slept for a while.
 * For all load averages >= 1 and max p_estcpu of 255, sleeping for at
 * least six times the loadfactor will decay p_estcpu to zero.
 */
void
updatepri(p)
	register struct proc *p;
{
	register unsigned int newcpu = p->p_estcpu;
	register fixpt_t loadfac = loadfactor(averunnable.ldavg[0]);

	if (p->p_slptime > 5 * loadfac)
		p->p_estcpu = 0;
	else {
		p->p_slptime--;	/* the first time was done in schedcpu */
		while (newcpu && --p->p_slptime)
			newcpu = (int) decay_cpu(loadfac, newcpu);
		p->p_estcpu = min(newcpu, UCHAR_MAX);
	}
	resetpriority(p);
}

/*
 * We're only looking at 7 bits of the address; everything is
 * aligned to 4, lots of things are aligned to greater powers
 * of 2.  Shift right by 8, i.e. drop the bottom 256 worth.
 */
#define TABLESIZE	128
#define LOOKUP(x)	(((int)(x) >> 8) & (TABLESIZE - 1))
struct slpque {
	struct proc *sq_head;
	struct proc **sq_tailp;
} slpque[TABLESIZE];

/*
 * During autoconfiguration or after a panic, a sleep will simply
 * lower the priority briefly to allow interrupts, then return.
 * The priority to be used (safepri) is machine-dependent, thus this
 * value is initialized and maintained in the machine-dependent layers.
 * This priority will typically be 0, or the lowest priority
 * that is safe for use on the interrupt stack; it can be made
 * higher to block network software interrupts after panics.
 */
int safepri;

/*
 * General sleep call.  Suspends the current process until a wakeup is
 * performed on the specified identifier.  The process will then be made
 * runnable with the specified priority.  Sleeps at most timo/hz seconds
 * (0 means no timeout).  If pri includes PCATCH flag, signals are checked
 * before and after sleeping, else signals are not checked.  Returns 0 if
 * awakened, EWOULDBLOCK if the timeout expires.  If PCATCH is set and a
 * signal needs to be delivered, ERESTART is returned if the current system
 * call should be restarted if possible, and EINTR is returned if the system
 * call should be interrupted by the signal (return EINTR).
 */
int
tsleep(ident, priority, wmesg, timo)
	void *ident;
	int priority, timo;
	char *wmesg;
{
	register struct proc *p = curproc;
	register struct slpque *qp;
	register s;
	int sig, catch = priority & PCATCH;
	extern int cold;
	void endtsleep __P((void *));

#ifdef KTRACE
	if (KTRPOINT(p, KTR_CSW))
		ktrcsw(p->p_tracep, 1, 0);
#endif
	s = splhigh();
	if (cold || panicstr) {
		/*
		 * After a panic, or during autoconfiguration,
		 * just give interrupts a chance, then just return;
		 * don't run any other procs or panic below,
		 * in case this is the idle process and already asleep.
		 */
		splx(safepri);
		splx(s);
		return (0);
	}
#ifdef DIAGNOSTIC
	if (ident == NULL || p->p_stat != SRUN || p->p_back)
		panic("tsleep");
#endif
	p->p_wchan = ident;
	p->p_wmesg = wmesg;
	p->p_slptime = 0;
	p->p_priority = priority & PRIMASK;
	qp = &slpque[LOOKUP(ident)];
	if (qp->sq_head == 0)
		qp->sq_head = p;
	else
		*qp->sq_tailp = p;
	*(qp->sq_tailp = &p->p_forw) = 0;
	if (timo)
		timeout(endtsleep, (void *)p, timo);
	/*
	 * We put ourselves on the sleep queue and start our timeout
	 * before calling CURSIG, as we could stop there, and a wakeup
	 * or a SIGCONT (or both) could occur while we were stopped.
	 * A SIGCONT would cause us to be marked as SSLEEP
	 * without resuming us, thus we must be ready for sleep
	 * when CURSIG is called.  If the wakeup happens while we're
	 * stopped, p->p_wchan will be 0 upon return from CURSIG.
	 */
	if (catch) {