ref: 8177d20fb2709ba9290dfd41308b8e5bee4e00f8
dir: /sys/src/9/port/proc.c/
#include <u.h>
#include "../port/lib.h"
#include "mem.h"
#include "dat.h"
#include "fns.h"
#include "../port/error.h"
#include "edf.h"
#include <trace.h>
int schedgain = 30; /* units in seconds */
int nrdy;
void updatecpu(Proc*);
int reprioritize(Proc*);
ulong delayedscheds; /* statistics */
long skipscheds;
long preempts;
ulong load;
static struct Procalloc
{
Lock;
Proc* ht[128];
Proc* arena;
Proc* free;
} procalloc;
enum
{
Q=10,
DQ=4,
Scaling=2,
};
Schedq runq[Nrq];
ulong runvec;
char *statename[] =
{ /* BUG: generate automatically */
"Dead",
"Moribund",
"Ready",
"Scheding",
"Running",
"Queueing",
"QueueingR",
"QueueingW",
"Wakeme",
"Broken",
"Stopped",
"Rendez",
"Waitrelease",
};
static void pidfree(Proc*);
static void rebalance(void);
/*
* Always splhi()'ed.
*/
void
schedinit(void) /* never returns */
{
Edf *e;
setlabel(&m->sched);
if(up != nil) {
if((e = up->edf) != nil && (e->flags & Admitted))
edfrecord(up);
m->proc = nil;
switch(up->state) {
case Running:
ready(up);
break;
case Moribund:
up->state = Dead;
edfstop(up);
if(up->edf != nil)
free(up->edf);
up->edf = nil;
/*
* Holding locks from pexit:
* procalloc
* palloc
*/
mmurelease(up);
unlock(&palloc);
up->mach = nil;
updatecpu(up);
up->qnext = procalloc.free;
procalloc.free = up;
/* proc is free now, make sure unlock() wont touch it */
up = procalloc.Lock.p = nil;
unlock(&procalloc);
sched();
}
up->mach = nil;
updatecpu(up);
up = nil;
}
sched();
}
/*
* If changing this routine, look also at sleep(). It
* contains a copy of the guts of sched().
*/
void
sched(void)
{
Proc *p;
if(m->ilockdepth)
panic("cpu%d: ilockdepth %d, last lock %#p at %#p, sched called from %#p",
m->machno,
m->ilockdepth,
up != nil ? up->lastilock: nil,
(up != nil && up->lastilock != nil) ? up->lastilock->pc: 0,
getcallerpc(&p+2));
if(up != nil) {
/*
* Delay the sched until the process gives up the locks
* it is holding. This avoids dumb lock loops.
* Don't delay if the process is Moribund.
* It called sched to die.
* But do sched eventually. This avoids a missing unlock
* from hanging the entire kernel.
* But don't reschedule procs holding palloc or procalloc.
* Those are far too important to be holding while asleep.
*
* This test is not exact. There can still be a few instructions
* in the middle of taslock when a process holds a lock
* but Lock.p has not yet been initialized.
*/
if(up->nlocks)
if(up->state != Moribund)
if(up->delaysched < 20
|| palloc.Lock.p == up
|| procalloc.Lock.p == up){
up->delaysched++;
delayedscheds++;
return;
}
up->delaysched = 0;
splhi();
/* statistics */
m->cs++;
procsave(up);
if(setlabel(&up->sched)){
procrestore(up);
spllo();
return;
}
gotolabel(&m->sched);
}
p = runproc();
if(p->edf == nil){
updatecpu(p);
p->priority = reprioritize(p);
}
if(p != m->readied)
m->schedticks = m->ticks + HZ/10;
m->readied = nil;
up = p;
up->state = Running;
up->mach = MACHP(m->machno);
m->proc = up;
mmuswitch(up);
gotolabel(&up->sched);
}
int
anyready(void)
{
return runvec;
}
int
anyhigher(void)
{
return runvec & ~((1<<(up->priority+1))-1);
}
/*
* here once per clock tick to see if we should resched
*/
void
hzsched(void)
{
/* once a second, rebalance will reprioritize ready procs */
if(m->machno == 0)
rebalance();
/* unless preempted, get to run for at least 100ms */
if(anyhigher()
|| (!up->fixedpri && (long)(m->ticks - m->schedticks) > 0 && anyready())){
m->readied = nil; /* avoid cooperative scheduling */
up->delaysched++;
}
}
/*
* here at the end of non-clock interrupts to see if we should preempt the
* current process. Returns 1 if preempted, 0 otherwise.
*/
int
preempted(void)
{
if(up != nil && up->state == Running)
if(up->preempted == 0)
if(anyhigher())
if(!active.exiting){
m->readied = nil; /* avoid cooperative scheduling */
up->preempted = 1;
sched();
splhi();
up->preempted = 0;
return 1;
}
return 0;
}
/*
* Update the cpu time average for this particular process,
* which is about to change from up -> not up or vice versa.
* p->lastupdate is the last time an updatecpu happened.
*
* The cpu time average is a decaying average that lasts
* about D clock ticks. D is chosen to be approximately
* the cpu time of a cpu-intensive "quick job". A job has to run
* for approximately D clock ticks before we home in on its
* actual cpu usage. Thus if you manage to get in and get out
* quickly, you won't be penalized during your burst. Once you
* start using your share of the cpu for more than about D
* clock ticks though, your p->cpu hits 1000 (1.0) and you end up
* below all the other quick jobs. Interactive tasks, because
* they basically always use less than their fair share of cpu,
* will be rewarded.
*
* If the process has not been running, then we want to
* apply the filter
*
* cpu = cpu * (D-1)/D
*
* n times, yielding
*
* cpu = cpu * ((D-1)/D)^n
*
* but D is big enough that this is approximately
*
* cpu = cpu * (D-n)/D
*
* so we use that instead.
*
* If the process has been running, we apply the filter to
* 1 - cpu, yielding a similar equation. Note that cpu is
* stored in fixed point (* 1000).
*
* Updatecpu must be called before changing up, in order
* to maintain accurate cpu usage statistics. It can be called
* at any time to bring the stats for a given proc up-to-date.
*/
void
updatecpu(Proc *p)
{
ulong t, ocpu, n, D;
if(p->edf != nil)
return;
t = MACHP(0)->ticks*Scaling + Scaling/2;
n = t - p->lastupdate;
if(n == 0)
return;
p->lastupdate = t;
D = schedgain*HZ*Scaling;
if(n > D)
n = D;
ocpu = p->cpu;
if(p != up)
p->cpu = (ocpu*(D-n))/D;
else{
t = 1000 - ocpu;
t = (t*(D-n))/D;
p->cpu = 1000 - t;
}
//iprint("pid %lud %s for %lud cpu %lud -> %lud\n", p->pid,p==up?"active":"inactive",n, ocpu,p->cpu);
}
/*
* On average, p has used p->cpu of a cpu recently.
* Its fair share is conf.nmach/m->load of a cpu. If it has been getting
* too much, penalize it. If it has been getting not enough, reward it.
* I don't think you can get much more than your fair share that
* often, so most of the queues are for using less. Having a priority
* of 3 means you're just right. Having a higher priority (up to p->basepri)
* means you're not using as much as you could.
*/
int
reprioritize(Proc *p)
{
int fairshare, n, load, ratio;
load = MACHP(0)->load;
if(load == 0)
return p->basepri;
/*
* fairshare = 1.000 * conf.nmach * 1.000/load,
* except the decimal point is moved three places
* on both load and fairshare.
*/
fairshare = (conf.nmach*1000*1000)/load;
n = p->cpu;
if(n == 0)
n = 1;
ratio = (fairshare+n/2) / n;
if(ratio > p->basepri)
ratio = p->basepri;
if(ratio < 0)
panic("reprioritize");
//iprint("pid %lud cpu %lud load %d fair %d pri %d\n", p->pid, p->cpu, load, fairshare, ratio);
return ratio;
}
/*
* add a process to a scheduling queue
*/
void
queueproc(Schedq *rq, Proc *p)
{
int pri;
pri = rq - runq;
lock(runq);
p->priority = pri;
p->rnext = nil;
if(rq->tail != nil)
rq->tail->rnext = p;
else
rq->head = p;
rq->tail = p;
rq->n++;
nrdy++;
runvec |= 1<<pri;
unlock(runq);
}
/*
* try to remove a process from a scheduling queue (called splhi)
*/
Proc*
dequeueproc(Schedq *rq, Proc *tp)
{
Proc *l, *p;
if(!canlock(runq))
return nil;
/*
* the queue may have changed before we locked runq,
* refind the target process.
*/
l = nil;
for(p = rq->head; p != nil; p = p->rnext){
if(p == tp)
break;
l = p;
}
/*
* p->mach==0 only when process state is saved
*/
if(p == nil || p->mach != nil){
unlock(runq);
return nil;
}
if(p->rnext == nil)
rq->tail = l;
if(l != nil)
l->rnext = p->rnext;
else
rq->head = p->rnext;
if(rq->head == nil)
runvec &= ~(1<<(rq-runq));
rq->n--;
nrdy--;
if(p->state != Ready)
print("dequeueproc %s %lud %s\n", p->text, p->pid, statename[p->state]);
unlock(runq);
return p;
}
/*
* ready(p) picks a new priority for a process and sticks it in the
* runq for that priority.
*/
void
ready(Proc *p)
{
int s, pri;
Schedq *rq;
void (*pt)(Proc*, int, vlong);
if(p->state == Ready){
print("double ready %s %lud pc %p\n", p->text, p->pid, getcallerpc(&p));
return;
}
s = splhi();
if(edfready(p)){
splx(s);
return;
}
if(up != p && (p->wired == nil || p->wired == MACHP(m->machno)))
m->readied = p; /* group scheduling */
updatecpu(p);
pri = reprioritize(p);
p->priority = pri;
rq = &runq[pri];
p->state = Ready;
queueproc(rq, p);
pt = proctrace;
if(pt != nil)
pt(p, SReady, 0);
splx(s);
}
/*
* yield the processor and drop our priority
*/
void
yield(void)
{
if(anyready()){
/* pretend we just used 1/2 tick */
up->lastupdate -= Scaling/2;
sched();
}
}
/*
* recalculate priorities once a second. We need to do this
* since priorities will otherwise only be recalculated when
* the running process blocks.
*/
ulong balancetime;
static void
rebalance(void)
{
int pri, npri, x;
Schedq *rq;
Proc *p;
ulong t;
t = m->ticks;
if(t - balancetime < HZ)
return;
balancetime = t;
for(pri=0, rq=runq; pri<Npriq; pri++, rq++){
another:
p = rq->head;
if(p == nil)
continue;
if(p->mp != MACHP(m->machno))
continue;
if(pri == p->basepri)
continue;
updatecpu(p);
npri = reprioritize(p);
if(npri != pri){
x = splhi();
p = dequeueproc(rq, p);
if(p != nil)
queueproc(&runq[npri], p);
splx(x);
goto another;
}
}
}
/*
* pick a process to run
*/
Proc*
runproc(void)
{
Schedq *rq;
Proc *p;
ulong start, now;
int i;
void (*pt)(Proc*, int, vlong);
start = perfticks();
/* cooperative scheduling until the clock ticks */
if((p = m->readied) != nil && p->mach == nil && p->state == Ready
&& (p->wired == nil || p->wired == MACHP(m->machno))
&& runq[Nrq-1].head == nil && runq[Nrq-2].head == nil){
skipscheds++;
rq = &runq[p->priority];
goto found;
}
preempts++;
loop:
/*
* find a process that last ran on this processor (affinity),
* or one that hasn't moved in a while (load balancing). Every
* time around the loop affinity goes down.
*/
spllo();
for(i = 0;; i++){
/*
* find the highest priority target process that this
* processor can run given affinity constraints.
*
*/
for(rq = &runq[Nrq-1]; rq >= runq; rq--){
for(p = rq->head; p != nil; p = p->rnext){
if(p->mp == nil || p->mp == MACHP(m->machno)
|| (p->wired == nil && i > 0))
goto found;
}
}
/* waste time or halt the CPU */
idlehands();
/* remember how much time we're here */
now = perfticks();
m->perf.inidle += now-start;
start = now;
}
found:
splhi();
p = dequeueproc(rq, p);
if(p == nil)
goto loop;
p->state = Scheding;
p->mp = MACHP(m->machno);
if(edflock(p)){
edfrun(p, rq == &runq[PriEdf]); /* start deadline timer and do admin */
edfunlock();
}
pt = proctrace;
if(pt != nil)
pt(p, SRun, 0);
return p;
}
int
canpage(Proc *p)
{
int ok = 0;
splhi();
lock(runq);
/* Only reliable way to see if we are Running */
if(p->mach == nil) {
p->newtlb = 1;
ok = 1;
}
unlock(runq);
spllo();
return ok;
}
Proc*
newproc(void)
{
char msg[64];
Proc *p;
lock(&procalloc);
for(;;) {
if((p = procalloc.free) != nil)
break;
snprint(msg, sizeof msg, "no procs; %s forking",
up != nil ? up->text: "kernel");
unlock(&procalloc);
resrcwait(msg);
lock(&procalloc);
}
procalloc.free = p->qnext;
unlock(&procalloc);
p->state = Scheding;
p->psstate = "New";
p->mach = nil;
p->eql = nil;
p->qnext = nil;
p->nchild = 0;
p->nwait = 0;
p->waitq = nil;
p->parent = nil;
p->pgrp = nil;
p->egrp = nil;
p->fgrp = nil;
p->rgrp = nil;
p->pdbg = nil;
p->fpstate = FPinit;
p->kp = 0;
p->procctl = 0;
p->syscalltrace = nil;
p->notepending = 0;
p->ureg = nil;
p->privatemem = 0;
p->noswap = 0;
p->errstr = p->errbuf0;
p->syserrstr = p->errbuf1;
p->errbuf0[0] = '\0';
p->errbuf1[0] = '\0';
p->nlocks = 0;
p->delaysched = 0;
p->trace = 0;
kstrdup(&p->user, "*nouser");
kstrdup(&p->text, "*notext");
kstrdup(&p->args, "");
p->nargs = 0;
p->setargs = 0;
memset(p->seg, 0, sizeof p->seg);
p->parentpid = 0;
p->noteid = pidalloc(p);
if(p->kstack == nil)
p->kstack = smalloc(KSTACK);
/* sched params */
p->mp = nil;
p->wired = nil;
procpriority(p, PriNormal, 0);
p->cpu = 0;
p->lastupdate = MACHP(0)->ticks*Scaling;
p->edf = nil;
return p;
}
/*
* wire this proc to a machine
*/
void
procwired(Proc *p, int bm)
{
Proc *pp;
int i;
char nwired[MAXMACH];
Mach *wm;
if(bm < 0){
/* pick a machine to wire to */
memset(nwired, 0, sizeof(nwired));
p->wired = nil;
pp = proctab(0);
for(i=0; i<conf.nproc; i++, pp++){
wm = pp->wired;
if(wm != nil && pp->pid)
nwired[wm->machno]++;
}
bm = 0;
for(i=0; i<conf.nmach; i++)
if(nwired[i] < nwired[bm])
bm = i;
} else {
/* use the virtual machine requested */
bm = bm % conf.nmach;
}
p->wired = MACHP(bm);
p->mp = p->wired;
}
void
procpriority(Proc *p, int pri, int fixed)
{
if(pri >= Npriq)
pri = Npriq - 1;
else if(pri < 0)
pri = 0;
p->basepri = pri;
p->priority = pri;
if(fixed){
p->fixedpri = 1;
} else {
p->fixedpri = 0;
}
}
void
procinit0(void) /* bad planning - clashes with devproc.c */
{
Proc *p;
int i;
p = xalloc(conf.nproc*sizeof(Proc));
if(p == nil){
xsummary();
panic("cannot allocate %lud procs (%ludMB)", conf.nproc, conf.nproc*sizeof(Proc)/(1024*1024));
}
procalloc.arena = p;
procalloc.free = p;
for(i=0; i<conf.nproc-1; i++, p++)
p->qnext = p+1;
p->qnext = nil;
}
/*
* sleep if a condition is not true. Another process will
* awaken us after it sets the condition. When we awaken
* the condition may no longer be true.
*
* we lock both the process and the rendezvous to keep r->p
* and p->r synchronized.
*/
void
sleep(Rendez *r, int (*f)(void*), void *arg)
{
int s;
void (*pt)(Proc*, int, vlong);
s = splhi();
if(up->nlocks)
print("process %lud sleeps with %d locks held, last lock %#p locked at pc %#p, sleep called from %#p\n",
up->pid, up->nlocks, up->lastlock, up->lastlock->pc, getcallerpc(&r));
lock(r);
lock(&up->rlock);
if(r->p != nil){
print("double sleep called from %#p, %lud %lud\n", getcallerpc(&r), r->p->pid, up->pid);
dumpstack();
}
/*
* Wakeup only knows there may be something to do by testing
* r->p in order to get something to lock on.
* Flush that information out to memory in case the sleep is
* committed.
*/
r->p = up;
if((*f)(arg) || up->notepending){
/*
* if condition happened or a note is pending
* never mind
*/
r->p = nil;
unlock(&up->rlock);
unlock(r);
} else {
/*
* now we are committed to
* change state and call scheduler
*/
pt = proctrace;
if(pt != nil)
pt(up, SSleep, 0);
up->state = Wakeme;
up->r = r;
/* statistics */
m->cs++;
procsave(up);
if(setlabel(&up->sched)) {
/*
* here when the process is awakened
*/
procrestore(up);
spllo();
} else {
/*
* here to go to sleep (i.e. stop Running)
*/
unlock(&up->rlock);
unlock(r);
gotolabel(&m->sched);
}
}
if(up->notepending) {
up->notepending = 0;
splx(s);
interrupted();
}
splx(s);
}
void
interrupted(void)
{
if(up->procctl == Proc_exitme && up->closingfgrp != nil)
forceclosefgrp();
error(Eintr);
}
static int
tfn(void *arg)
{
return up->trend == nil || up->tfn(arg);
}
static void
twakeup(Ureg*, Timer *t)
{
Proc *p;
Rendez *trend;
ilock(t);
p = t->ta;
trend = p->trend;
if(trend != nil){
wakeup(trend);
p->trend = nil;
}
iunlock(t);
}
static void
stoptimer(void)
{
if(up->trend != nil){
up->trend = nil;
timerdel(up);
}
}
void
tsleep(Rendez *r, int (*fn)(void*), void *arg, ulong ms)
{
if(up->tt != nil){
print("%s %lud: tsleep timer active: mode %d, tf %#p, pc %#p\n",
up->text, up->pid, up->tmode, up->tf, getcallerpc(&r));
timerdel(up);
}
up->tns = MS2NS(ms);
up->tf = twakeup;
up->tmode = Trelative;
up->ta = up;
up->trend = r;
up->tfn = fn;
timeradd(up);
if(waserror()){
stoptimer();
nexterror();
}
sleep(r, tfn, arg);
stoptimer();
poperror();
}
/*
* Expects that only one process can call wakeup for any given Rendez.
* We hold both locks to ensure that r->p and p->r remain consistent.
* Richard Miller has a better solution that doesn't require both to
* be held simultaneously, but I'm a paranoid - presotto.
*/
Proc*
wakeup(Rendez *r)
{
Proc *p;
int s;
s = splhi();
lock(r);
p = r->p;
if(p != nil){
lock(&p->rlock);
if(p->state != Wakeme || p->r != r){
iprint("%p %p %d\n", p->r, r, p->state);
panic("wakeup: state");
}
r->p = nil;
p->r = nil;
ready(p);
unlock(&p->rlock);
}
unlock(r);
splx(s);
return p;
}
/*
* if waking a sleeping process, this routine must hold both
* p->rlock and r->lock. However, it can't know them in
* the same order as wakeup causing a possible lock ordering
* deadlock. We break the deadlock by giving up the p->rlock
* lock if we can't get the r->lock and retrying.
*/
int
postnote(Proc *p, int dolock, char *n, int flag)
{
int s, ret;
QLock *q;
if(p == nil)
return 0;
if(dolock)
qlock(&p->debug);
if(p->pid == 0){
if(dolock)
qunlock(&p->debug);
return 0;
}
if(n != nil && flag != NUser && (p->notify == 0 || p->notified))
p->nnote = 0;
ret = 0;
if(p->nnote < NNOTE && n != nil) {
kstrcpy(p->note[p->nnote].msg, n, ERRMAX);
p->note[p->nnote++].flag = flag;
ret = 1;
}
p->notepending = 1;
if(dolock)
qunlock(&p->debug);
/* this loop is to avoid lock ordering problems. */
for(;;){
Rendez *r;
s = splhi();
lock(&p->rlock);
r = p->r;
/* waiting for a wakeup? */
if(r == nil)
break; /* no */
/* try for the second lock */
if(canlock(r)){
if(p->state != Wakeme || r->p != p)
panic("postnote: state %d %d %d", r->p != p, p->r != r, p->state);
p->r = nil;
r->p = nil;
ready(p);
unlock(r);
break;
}
/* give other process time to get out of critical section and try again */
unlock(&p->rlock);
splx(s);
sched();
}
unlock(&p->rlock);
splx(s);
switch(p->state){
case Queueing:
/* Try and pull out of a eqlock */
if((q = p->eql) != nil){
lock(&q->use);
if(p->state == Queueing && p->eql == q){
Proc *d, *l;
for(l = nil, d = q->head; d != nil; l = d, d = d->qnext){
if(d == p){
if(l != nil)
l->qnext = p->qnext;
else
q->head = p->qnext;
if(p->qnext == nil)
q->tail = l;
p->qnext = nil;
p->eql = nil; /* not taken */
ready(p);
break;
}
}
}
unlock(&q->use);
}
break;
case Rendezvous:
/* Try and pull out of a rendezvous */
lock(p->rgrp);
if(p->state == Rendezvous) {
Proc *d, **l;
l = &REND(p->rgrp, p->rendtag);
for(d = *l; d != nil; d = d->rendhash) {
if(d == p) {
*l = p->rendhash;
p->rendval = ~0;
ready(p);
break;
}
l = &d->rendhash;
}
}
unlock(p->rgrp);
break;
}
return ret;
}
/*
* weird thing: keep at most NBROKEN around
*/
#define NBROKEN 4
struct
{
QLock;
int n;
Proc *p[NBROKEN];
}broken;
void
addbroken(Proc *p)
{
qlock(&broken);
if(broken.n == NBROKEN) {
ready(broken.p[0]);
memmove(&broken.p[0], &broken.p[1], sizeof(Proc*)*(NBROKEN-1));
--broken.n;
}
broken.p[broken.n++] = p;
qunlock(&broken);
edfstop(up);
p->state = Broken;
p->psstate = nil;
sched();
}
void
unbreak(Proc *p)
{
int b;
qlock(&broken);
for(b=0; b < broken.n; b++)
if(broken.p[b] == p) {
broken.n--;
memmove(&broken.p[b], &broken.p[b+1],
sizeof(Proc*)*(NBROKEN-(b+1)));
ready(p);
break;
}
qunlock(&broken);
}
int
freebroken(void)
{
int i, n;
qlock(&broken);
n = broken.n;
for(i=0; i<n; i++) {
ready(broken.p[i]);
broken.p[i] = nil;
}
broken.n = 0;
qunlock(&broken);
return n;
}
void
pexit(char *exitstr, int freemem)
{
Proc *p;
Segment **s, **es;
ulong utime, stime;
Waitq *wq;
Fgrp *fgrp;
Egrp *egrp;
Rgrp *rgrp;
Pgrp *pgrp;
Chan *dot;
void (*pt)(Proc*, int, vlong);
up->alarm = 0;
if(up->tt != nil)
timerdel(up);
pt = proctrace;
if(pt != nil)
pt(up, SDead, 0);
/* nil out all the resources under lock (free later) */
qlock(&up->debug);
fgrp = up->fgrp;
up->fgrp = nil;
egrp = up->egrp;
up->egrp = nil;
rgrp = up->rgrp;
up->rgrp = nil;
pgrp = up->pgrp;
up->pgrp = nil;
dot = up->dot;
up->dot = nil;
qunlock(&up->debug);
if(fgrp != nil)
closefgrp(fgrp);
if(egrp != nil)
closeegrp(egrp);
if(rgrp != nil)
closergrp(rgrp);
if(dot != nil)
cclose(dot);
if(pgrp != nil)
closepgrp(pgrp);
/*
* if not a kernel process and have a parent,
* do some housekeeping.
*/
if(up->kp == 0 && up->parentpid != 0) {
wq = smalloc(sizeof(Waitq));
wq->w.pid = up->pid;
utime = up->time[TUser] + up->time[TCUser];
stime = up->time[TSys] + up->time[TCSys];
wq->w.time[TUser] = tk2ms(utime);
wq->w.time[TSys] = tk2ms(stime);
wq->w.time[TReal] = tk2ms(MACHP(0)->ticks - up->time[TReal]);
if(exitstr != nil && exitstr[0])
snprint(wq->w.msg, sizeof(wq->w.msg), "%s %lud: %s", up->text, up->pid, exitstr);
else
wq->w.msg[0] = '\0';
p = up->parent;
lock(&p->exl);
/*
* Check that parent is still alive.
*/
if(p->pid == up->parentpid && p->state != Broken) {
p->nchild--;
p->time[TCUser] += utime;
p->time[TCSys] += stime;
/*
* If there would be more than 128 wait records
* processes for my parent, then don't leave a wait
* record behind. This helps prevent badly written
* daemon processes from accumulating lots of wait
* records.
*/
if(p->nwait < 128) {
wq->next = p->waitq;
p->waitq = wq;
p->nwait++;
wq = nil;
wakeup(&p->waitr);
}
}
unlock(&p->exl);
if(wq != nil)
free(wq);
}
else if(up->kp == 0 && up->parent == nil){
if(exitstr == nil)
exitstr = "unknown";
panic("boot process died: %s", exitstr);
}
if(!freemem)
addbroken(up);
qlock(&up->seglock);
es = &up->seg[NSEG];
for(s = up->seg; s < es; s++) {
if(*s != nil) {
putseg(*s);
*s = nil;
}
}
qunlock(&up->seglock);
lock(&up->exl); /* Prevent my children from leaving waits */
pidfree(up);
up->pid = 0;
wakeup(&up->waitr);
unlock(&up->exl);
while((wq = up->waitq) != nil){
up->waitq = wq->next;
free(wq);
}
/* release debuggers */
qlock(&up->debug);
if(up->pdbg != nil) {
wakeup(&up->pdbg->sleep);
up->pdbg = nil;
}
if(up->syscalltrace != nil) {
free(up->syscalltrace);
up->syscalltrace = nil;
}
qunlock(&up->debug);
/* Sched must not loop for these locks */
lock(&procalloc);
lock(&palloc);
edfstop(up);
up->state = Moribund;
sched();
panic("pexit");
}
static int
haswaitq(void *x)
{
return ((Proc*)x)->waitq != nil;
}
ulong
pwait(Waitmsg *w)
{
ulong cpid;
Waitq *wq;
if(!canqlock(&up->qwaitr))
error(Einuse);
if(waserror()) {
qunlock(&up->qwaitr);
nexterror();
}
lock(&up->exl);
while(up->waitq == nil) {
if(up->nchild == 0) {
unlock(&up->exl);
error(Enochild);
}
unlock(&up->exl);
sleep(&up->waitr, haswaitq, up);
lock(&up->exl);
}
wq = up->waitq;
up->waitq = wq->next;
up->nwait--;
unlock(&up->exl);
qunlock(&up->qwaitr);
poperror();
if(w != nil)
memmove(w, &wq->w, sizeof(Waitmsg));
cpid = wq->w.pid;
free(wq);
return cpid;
}
Proc*
proctab(int i)
{
return &procalloc.arena[i];
}
void
dumpaproc(Proc *p)
{
ulong bss;
char *s;
if(p == nil)
return;
bss = 0;
if(p->seg[BSEG] != nil)
bss = p->seg[BSEG]->top;
s = p->psstate;
if(s == nil)
s = statename[p->state];
print("%3lud:%10s pc %#p dbgpc %#p %8s (%s) ut %ld st %ld bss %lux qpc %#p nl %d nd %lud lpc %#p pri %lud\n",
p->pid, p->text, p->pc, dbgpc(p), s, statename[p->state],
p->time[0], p->time[1], bss, p->qpc, p->nlocks, p->delaysched,
p->lastlock ? p->lastlock->pc : 0, p->priority);
}
void
procdump(void)
{
int i;
Proc *p;
if(up != nil)
print("up %lud\n", up->pid);
else
print("no current process\n");
for(i=0; i<conf.nproc; i++) {
p = &procalloc.arena[i];
if(p->state != Dead)
dumpaproc(p);
}
}
/*
* wait till all matching processes have flushed their mmu
*/
static void
procflushmmu(int (*match)(Proc*, void*), void *a)
{
int i, nm, nwait;
Proc *p;
/*
* tell all matching processes to flush their mmu's
*/
nwait = 0;
for(i=0; i<conf.nproc; i++) {
p = &procalloc.arena[i];
if(p->state != Dead && (*match)(p, a)){
p->newtlb = 1;
for(nm = 0; nm < conf.nmach; nm++){
if(MACHP(nm)->proc == p){
MACHP(nm)->flushmmu = 1;
nwait++;
}
}
}
}
if(nwait == 0)
return;
/*
* wait for all other processors to take a clock interrupt
* and flush their mmu's
*/
for(nm = 0; nm < conf.nmach; nm++)
while(m->machno != nm && MACHP(nm)->flushmmu)
sched();
}
static int
matchseg(Proc *p, void *a)
{
int ns;
for(ns = 0; ns < NSEG; ns++){
if(p->seg[ns] == a)
return 1;
}
return 0;
}
void
procflushseg(Segment *s)
{
procflushmmu(matchseg, s);
}
static int
matchpseg(Proc *p, void *a)
{
Segment *s;
int ns;
for(ns = 0; ns < NSEG; ns++){
s = p->seg[ns];
if(s != nil && s->pseg == a)
return 1;
}
return 0;
}
void
procflushpseg(Physseg *ps)
{
procflushmmu(matchpseg, ps);
}
void
scheddump(void)
{
Proc *p;
Schedq *rq;
for(rq = &runq[Nrq-1]; rq >= runq; rq--){
if(rq->head == nil)
continue;
print("rq%zd:", rq-runq);
for(p = rq->head; p != nil; p = p->rnext)
print(" %lud(%lud)", p->pid, m->ticks - p->readytime);
print("\n");
delay(150);
}
print("nrdy %d\n", nrdy);
}
void
kproc(char *name, void (*func)(void *), void *arg)
{
Proc *p;
static Pgrp *kpgrp;
p = newproc();
p->psstate = nil;
p->procmode = 0640;
p->kp = 1;
p->noswap = 1;
p->scallnr = up->scallnr;
p->s = up->s;
p->nerrlab = 0;
p->slash = up->slash;
p->dot = up->slash; /* unlike fork, do not inherit the dot for kprocs */
if(p->dot != nil)
incref(p->dot);
memmove(p->note, up->note, sizeof(p->note));
p->nnote = up->nnote;
p->notified = 0;
p->lastnote = up->lastnote;
p->notify = up->notify;
p->ureg = nil;
p->dbgreg = nil;
procpriority(p, PriKproc, 0);
kprocchild(p, func, arg);
kstrdup(&p->user, eve);
kstrdup(&p->text, name);
if(kpgrp == nil)
kpgrp = newpgrp();
p->pgrp = kpgrp;
incref(kpgrp);
memset(p->time, 0, sizeof(p->time));
p->time[TReal] = MACHP(0)->ticks;
ready(p);
}
/*
* called splhi() by notify(). See comment in notify for the
* reasoning.
*/
void
procctl(void)
{
char *state;
ulong s;
switch(up->procctl) {
case Proc_exitbig:
spllo();
pprint("Killed: Insufficient physical memory\n");
pexit("Killed: Insufficient physical memory", 1);
case Proc_exitme:
spllo(); /* pexit has locks in it */
pexit("Killed", 1);
case Proc_traceme:
if(up->nnote == 0)
return;
/* No break */
case Proc_stopme:
up->procctl = 0;
state = up->psstate;
up->psstate = "Stopped";
/* free a waiting debugger */
s = spllo();
qlock(&up->debug);
if(up->pdbg != nil) {
wakeup(&up->pdbg->sleep);
up->pdbg = nil;
}
qunlock(&up->debug);
splhi();
up->state = Stopped;
sched();
up->psstate = state;
splx(s);
return;
}
}
#include "errstr.h"
void
error(char *err)
{
spllo();
assert(up->nerrlab < NERR);
kstrcpy(up->errstr, err, ERRMAX);
setlabel(&up->errlab[NERR-1]);
nexterror();
}
void
nexterror(void)
{
assert(up->nerrlab > 0);
gotolabel(&up->errlab[--up->nerrlab]);
}
void
exhausted(char *resource)
{
char buf[ERRMAX];
snprint(buf, sizeof buf, "no free %s", resource);
iprint("%s\n", buf);
error(buf);
}
ulong
procpagecount(Proc *p)
{
Segment *s;
ulong pages;
int i;
eqlock(&p->seglock);
if(waserror()){
qunlock(&p->seglock);
nexterror();
}
pages = 0;
for(i=0; i<NSEG; i++){
if((s = p->seg[i]) != nil){
eqlock(s);
pages += mcountseg(s);
qunlock(s);
}
}
qunlock(&p->seglock);
poperror();
return pages;
}
void
killbig(char *why)
{
int i;
Segment *s;
ulong l, max;
Proc *p, *ep, *kp;
max = 0;
kp = nil;
ep = procalloc.arena+conf.nproc;
for(p = procalloc.arena; p < ep; p++) {
if(p->state == Dead || p->kp)
continue;
if((p->noswap || (p->procmode & 0222) == 0) && strcmp(eve, p->user) == 0)
continue;
l = procpagecount(p);
if(l > max){
kp = p;
max = l;
}
}
if(kp == nil)
return;
print("%lud: %s killed: %s\n", kp->pid, kp->text, why);
qlock(&kp->seglock);
for(p = procalloc.arena; p < ep; p++) {
if(p->state == Dead || p->kp)
continue;
if(p != kp && p->seg[BSEG] != nil && p->seg[BSEG] == kp->seg[BSEG])
p->procctl = Proc_exitbig;
}
kp->procctl = Proc_exitbig;
for(i = 0; i < NSEG; i++) {
s = kp->seg[i];
if(s == nil)
continue;
switch(s->type & SG_TYPE){
case SG_SHARED:
case SG_PHYSICAL:
case SG_FIXED:
continue;
}
qlock(s);
mfreeseg(s, s->base, (s->top - s->base)/BY2PG);
qunlock(s);
}
qunlock(&kp->seglock);
}
/*
* change ownership to 'new' of all processes owned by 'old'. Used when
* eve changes.
*/
void
renameuser(char *old, char *new)
{
Proc *p, *ep;
ep = procalloc.arena+conf.nproc;
for(p = procalloc.arena; p < ep; p++)
if(p->user != nil && strcmp(old, p->user) == 0)
kstrdup(&p->user, new);
}
/*
* time accounting called by clock() splhi'd
*/
void
accounttime(void)
{
Proc *p;
ulong n, per;
static ulong nrun;
p = m->proc;
if(p != nil) {
nrun++;
p->time[p->insyscall]++;
}
/* calculate decaying duty cycles */
n = perfticks();
per = n - m->perf.last;
m->perf.last = n;
per = ((uvlong)m->perf.period*(HZ-1) + per)/HZ;
if(per != 0)
m->perf.period = per;
m->perf.avg_inidle = ((uvlong)m->perf.avg_inidle*(HZ-1)+m->perf.inidle)/HZ;
m->perf.inidle = 0;
m->perf.avg_inintr = ((uvlong)m->perf.avg_inintr*(HZ-1)+m->perf.inintr)/HZ;
m->perf.inintr = 0;
/* only one processor gets to compute system load averages */
if(m->machno != 0)
return;
/*
* calculate decaying load average.
* if we decay by (n-1)/n then it takes
* n clock ticks to go from load L to .36 L once
* things quiet down. it takes about 5 n clock
* ticks to go to zero. so using HZ means this is
* approximately the load over the last second,
* with a tail lasting about 5 seconds.
*/
n = nrun;
nrun = 0;
n = (nrdy+n)*1000*100;
load = ((uvlong)load*(HZ-1)+n)/HZ;
m->load = load/100;
}
int
pidalloc(Proc *p)
{
static int gen, wrapped;
int pid, h;
Proc *x;
lock(&procalloc);
Retry:
pid = ++gen & 0x7FFFFFFF;
if(pid == 0){
wrapped = 1;
goto Retry;
}
h = pid % nelem(procalloc.ht);
if(wrapped)
for(x = procalloc.ht[h]; x != nil; x = x->pidhash)
if(x->pid == pid)
goto Retry;
if(p != nil){
p->pid = pid;
p->pidhash = procalloc.ht[h];
procalloc.ht[h] = p;
}
unlock(&procalloc);
return pid;
}
static void
pidfree(Proc *p)
{
int h;
Proc **l;
h = p->pid % nelem(procalloc.ht);
lock(&procalloc);
for(l = &procalloc.ht[h]; *l != nil; l = &(*l)->pidhash)
if(*l == p){
*l = p->pidhash;
break;
}
unlock(&procalloc);
}
int
procindex(ulong pid)
{
Proc *p;
int h;
int s;
s = -1;
h = pid % nelem(procalloc.ht);
lock(&procalloc);
for(p = procalloc.ht[h]; p != nil; p = p->pidhash)
if(p->pid == pid){
s = p - procalloc.arena;
break;
}
unlock(&procalloc);
return s;
}