new calling convension for system calls

[unix-history] / usr / src / sys / kern / kern_synch.c

diff --git a/usr/src/sys/kern/kern_synch.c b/usr/src/sys/kern/kern_synch.c
index c634975..61b3f79 100644 (file)
--- a/usr/src/sys/kern/kern_synch.c
+++ b/usr/src/sys/kern/kern_synch.c
@@ -1,21 +1,23 @@
-/*     kern_synch.c    6.9     85/06/07        */
+/*
+ * Copyright (c) 1982, 1986, 1990 Regents of the University of California.
+ * All rights reserved.  The Berkeley software License Agreement
+ * specifies the terms and conditions for redistribution.
+ *
+ *     @(#)kern_synch.c        7.11 (Berkeley) %G%
+ */

 
 

-#include "../machine/pte.h"
+#include "machine/pte.h"
+#include "machine/psl.h"
+#include "machine/mtpr.h"

 
 #include "param.h"
 #include "systm.h"
 
 #include "param.h"
 #include "systm.h"

-#include "dir.h"

 #include "user.h"
 #include "proc.h"
 #include "user.h"
 #include "proc.h"

-#include "file.h"
-#include "inode.h"

 #include "vm.h"
 #include "kernel.h"
 #include "buf.h"
 
 #include "vm.h"
 #include "kernel.h"
 #include "buf.h"
 

-#ifdef vax
-#include "../vax/mtpr.h"       /* XXX */
-#endif

 /*
  * Force switch among equal priority processes every 100ms.
  */
 /*
  * Force switch among equal priority processes every 100ms.
  */


@@ -27,20 +29,99 @@ roundrobin()
        timeout(roundrobin, (caddr_t)0, hz / 10);
 }
 
        timeout(roundrobin, (caddr_t)0, hz / 10);
 }
 

-/* fraction for digital decay to forget 90% of usage in 5*loadav sec */
-#define        filter(loadav) ((2 * (loadav)) / (2 * (loadav) + 1))
+/*
+ * constants for digital decay and forget
+ *     90% of (p_cpu) 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_cpu and p_cpticks independently.
+ *
+ * We wish to decay away 90% of p_cpu 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_cpu *= decay;
+ * will compute
+ *     p_cpu *= 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        get_b(loadav)           (2 * (loadav))
+#define        get_pcpu(b, cpu)        (((b) * ((cpu) & 0377)) / ((b) + FSCALE))

 
 

-double ccpu = 0.95122942450071400909;          /* exp(-1/20) */
+/* 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, once a second
  */
 schedcpu()
 {
 
 /*
  * Recompute process priorities, once a second
  */
 schedcpu()
 {

-       register double ccpu1 = (1.0 - ccpu) / (double)hz;
+       register fixpt_t b = get_b(averunnable[0]);

        register struct proc *p;
        register int s, a;
        register struct proc *p;
        register int s, a;

-       float scale = filter(avenrun[0]);

 
        wakeup((caddr_t)&lbolt);
        for (p = allproc; p != NULL; p = p->p_nxt) {
 
        wakeup((caddr_t)&lbolt);
        for (p = allproc; p != NULL; p = p->p_nxt) {


@@ -49,20 +130,27 @@ schedcpu()
                if (p->p_stat==SSLEEP || p->p_stat==SSTOP)
                        if (p->p_slptime != 127)
                                p->p_slptime++;
                if (p->p_stat==SSLEEP || p->p_stat==SSTOP)
                        if (p->p_slptime != 127)
                                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 the process has slept the entire second,
                 * stop recalculating its priority until it wakes up.
                 */

-               if (p->p_slptime > 1) {
-                       p->p_pctcpu *= ccpu;
+               if (p->p_slptime > 1)

                        continue;
                        continue;

-               }

                /*
                 * p_pctcpu is only for ps.
                 */
                /*
                 * p_pctcpu is only for ps.
                 */

-               p->p_pctcpu = ccpu * p->p_pctcpu + ccpu1 * p->p_cpticks;
+#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;
                p->p_cpticks = 0;

-               a = (int) (scale * (p->p_cpu & 0377)) + p->p_nice;
+               a = (int) get_pcpu(b, p->p_cpu) + p->p_nice;

                if (a < 0)
                        a = 0;
                if (a > 255)
                if (a < 0)
                        a = 0;
                if (a > 255)


@@ -101,11 +189,12 @@ updatepri(p)
        register struct proc *p;
 {
        register int a = p->p_cpu & 0377;
        register struct proc *p;
 {
        register int a = p->p_cpu & 0377;

-       float scale = filter(avenrun[0]);
+       register fixpt_t b = get_b(averunnable[0]);

 
        p->p_slptime--;         /* the first time was done in schedcpu */
        while (a && --p->p_slptime)
 
        p->p_slptime--;         /* the first time was done in schedcpu */
        while (a && --p->p_slptime)

-               a = (int) (scale * a) /* + p->p_nice */;
+               a = (int) get_pcpu(b, a) /* + p->p_nice */;
+       p->p_slptime = 0;

        if (a < 0)
                a = 0;
        if (a > 255)
        if (a < 0)
                a = 0;
        if (a > 255)


@@ -122,90 +211,175 @@ struct slpque {
 } slpque[SQSIZE];
 
 /*
 } slpque[SQSIZE];
 
 /*

- * Give up the processor till a wakeup occurs
- * on chan, at which time the process
- * enters the scheduling queue at priority pri.
- * The most important effect of pri is that when
- * pri<=PZERO a signal cannot disturb the sleep;
- * if pri>PZERO signals will be processed.
- * Callers of this routine must be prepared for
- * premature return, and check that the reason for
- * sleeping has gone away.
+ * General sleep call.
+ * Suspends current process until a wakeup is made on chan.
+ * The process will then be made runnable with priority pri.
+ * 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).

  */
  */

-sleep(chan, pri)
+tsleep(chan, pri, wmesg, timo)

        caddr_t chan;
        int pri;
        caddr_t chan;
        int pri;

+       char *wmesg;
+       int timo;

 {
        register struct proc *rp;
        register struct slpque *qp;
        register s;
 {
        register struct proc *rp;
        register struct slpque *qp;
        register s;

+       int sig, catch = pri & PCATCH;
+       extern int cold;
+       int endtsleep();

 
        rp = u.u_procp;
        s = splhigh();
 
        rp = u.u_procp;
        s = splhigh();

-       if (panicstr) {
+       if (cold || panicstr) {

                /*
                /*

-                * After a panic, 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.
-                * The splnet should be spl0 if the network was being used
-                * by the filesystem, but for now avoid network interrupts
-                * that might cause another panic.
+                * 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.

                 */
                 */

-               (void) splnet();
+               (void) spl0();

                splx(s);
                splx(s);

-               return;
+               return (0);

        }
        }

-       if (chan==0 || rp->p_stat != SRUN || rp->p_rlink)
-               panic("sleep");
+#ifdef DIAGNOSTIC
+       if (chan == 0 || rp->p_stat != SRUN || rp->p_rlink)
+               panic("tsleep");
+#endif

        rp->p_wchan = chan;
        rp->p_wchan = chan;

+       rp->p_wmesg = wmesg;

        rp->p_slptime = 0;
        rp->p_slptime = 0;

-       rp->p_pri = pri;
+       rp->p_pri = pri & PRIMASK;

        qp = &slpque[HASH(chan)];
        if (qp->sq_head == 0)
                qp->sq_head = rp;
        else
                *qp->sq_tailp = rp;
        *(qp->sq_tailp = &rp->p_link) = 0;
        qp = &slpque[HASH(chan)];
        if (qp->sq_head == 0)
                qp->sq_head = rp;
        else
                *qp->sq_tailp = rp;
        *(qp->sq_tailp = &rp->p_link) = 0;

-       if (pri > PZERO) {
-               /*
-                * If we stop in issig(), wakeup may already have happened
-                * when we return (rp->p_wchan will then be 0).
-                */
-               if (ISSIG(rp)) {
+       /*
+        * If we stop in CURSIG/issig(), wakeup may already
+        * have happened when we return.
+        * rp->p_wchan will then be 0.
+        */
+       if (catch) {
+               if (sig = CURSIG(rp)) {

                        if (rp->p_wchan)
                                unsleep(rp);
                        rp->p_stat = SRUN;
                        if (rp->p_wchan)
                                unsleep(rp);
                        rp->p_stat = SRUN;

-                       (void) spl0();
-                       goto psig;
+                       splx(s);
+                       if (u.u_sigintr & sigmask(sig))
+                               return (EINTR);
+                       return (ERESTART);

                }
                }

-               if (rp->p_wchan == 0)
-                       goto out;
-               rp->p_stat = SSLEEP;
-               (void) spl0();
-               u.u_ru.ru_nvcsw++;
-               swtch();
-               if (ISSIG(rp))
-                       goto psig;
-       } else {
-               rp->p_stat = SSLEEP;
-               (void) spl0();
-               u.u_ru.ru_nvcsw++;
-               swtch();
+               if (rp->p_wchan == 0) {
+                       splx(s);
+                       return (0);
+               }
+               rp->p_flag |= SSINTR;

        }
        }

+       rp->p_stat = SSLEEP;
+       if (timo)
+               timeout(endtsleep, (caddr_t)rp, timo);
+       (void) spl0();
+       u.u_ru.ru_nvcsw++;
+       swtch();

        curpri = rp->p_usrpri;
        curpri = rp->p_usrpri;

-out:

        splx(s);
        splx(s);

-       return;
+       rp->p_flag &= ~SSINTR;
+       if (rp->p_flag & STIMO) {
+               rp->p_flag &= ~STIMO;
+               return (EWOULDBLOCK);
+       }
+       if (timo)
+               untimeout(endtsleep, (caddr_t)rp);
+       if (catch && (sig = CURSIG(rp))) {
+               if (u.u_sigintr & sigmask(sig))
+                       return (EINTR);
+               return (ERESTART);
+       }
+       return (0);
+}

 
 

-       /*
-        * If priority was low (>PZERO) and
-        * there has been a signal, execute non-local goto through
-        * u.u_qsave, aborting the system call in progress (see trap.c)
-        */
-psig:
-       longjmp(&u.u_qsave);
-       /*NOTREACHED*/
+/*
+ * Implement timeout for tsleep.
+ * If process hasn't been awakened (wchan non-zero),
+ * set timeout flag and undo the sleep.  If proc
+ * is stopped, just unsleep so it will remain stopped.
+ */
+endtsleep(p)
+       register struct proc *p;
+{
+       int s = splhigh();
+
+       if (p->p_wchan) {
+               if (p->p_stat == SSLEEP)
+                       setrun(p);
+               else
+                       unsleep(p);
+               p->p_flag |= STIMO;
+       }
+       splx(s);
+}
+
+/*
+ * Short-term, non-interruptable sleep.
+ */
+sleep(chan, pri)
+       caddr_t chan;
+       int pri;
+{
+       register struct proc *rp;
+       register struct slpque *qp;
+       register s;
+       extern int cold;
+
+#ifdef DIAGNOSTIC
+       if (pri > PZERO) {
+               printf("sleep called with pri %d > PZERO, wchan: %x\n",
+                       pri, chan);
+               panic("old sleep");
+       }
+#endif
+       rp = u.u_procp;
+       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.
+                */
+               (void) spl0();
+               splx(s);
+               return;
+       }
+#ifdef DIAGNOSTIC
+       if (chan==0 || rp->p_stat != SRUN || rp->p_rlink)
+               panic("sleep");
+#endif
+       rp->p_wchan = chan;
+       rp->p_wmesg = NULL;
+       rp->p_slptime = 0;
+       rp->p_pri = pri;
+       qp = &slpque[HASH(chan)];
+       if (qp->sq_head == 0)
+               qp->sq_head = rp;
+       else
+               *qp->sq_tailp = rp;
+       *(qp->sq_tailp = &rp->p_link) = 0;
+       rp->p_stat = SSLEEP;
+       (void) spl0();
+       u.u_ru.ru_nvcsw++;
+       swtch();
+       curpri = rp->p_usrpri;
+       splx(s);

 }
 
 /*
 }
 
 /*


@@ -245,8 +419,10 @@ wakeup(chan)
        qp = &slpque[HASH(chan)];
 restart:
        for (q = &qp->sq_head; p = *q; ) {
        qp = &slpque[HASH(chan)];
 restart:
        for (q = &qp->sq_head; p = *q; ) {

+#ifdef DIAGNOSTIC

                if (p->p_rlink || p->p_stat != SSLEEP && p->p_stat != SSTOP)
                        panic("wakeup");
                if (p->p_rlink || p->p_stat != SSLEEP && p->p_stat != SSTOP)
                        panic("wakeup");

+#endif

                if (p->p_wchan==chan) {
                        p->p_wchan = 0;
                        *q = p->p_link;
                if (p->p_wchan==chan) {
                        p->p_wchan = 0;
                        *q = p->p_link;


@@ -256,7 +432,6 @@ restart:
                                /* OPTIMIZED INLINE EXPANSION OF setrun(p) */
                                if (p->p_slptime > 1)
                                        updatepri(p);
                                /* OPTIMIZED INLINE EXPANSION OF setrun(p) */
                                if (p->p_slptime > 1)
                                        updatepri(p);

-                               p->p_slptime = 0;

                                p->p_stat = SRUN;
                                if (p->p_flag & SLOAD)
                                        setrq(p);
                                p->p_stat = SRUN;
                                if (p->p_flag & SLOAD)
                                        setrq(p);


@@ -276,7 +451,6 @@ restart:
                                /* END INLINE EXPANSION */
                                goto restart;
                        }
                                /* END INLINE EXPANSION */
                                goto restart;
                        }

-                       p->p_slptime = 0;

                } else
                        q = &p->p_link;
        }
                } else
                        q = &p->p_link;
        }


@@ -322,12 +496,12 @@ setrun(p)
        case SIDL:
                break;
        }
        case SIDL:
                break;
        }

-       if (p->p_slptime > 1)
-               updatepri(p);

        p->p_stat = SRUN;
        if (p->p_flag & SLOAD)
                setrq(p);
        splx(s);
        p->p_stat = SRUN;
        if (p->p_flag & SLOAD)
                setrq(p);
        splx(s);

+       if (p->p_slptime > 1)
+               updatepri(p);

        if (p->p_pri < curpri) {
                runrun++;
                aston();
        if (p->p_pri < curpri) {
                runrun++;
                aston();