SCHED_4BSD scheduling quantum mechanism appears to be broken for some time.

With switchticks variable being reset each time thread preempted (that is
done regularly by interrupt threads) scheduling quantum may never expire.
It was not noticed in time because several other factors still regularly
trigger context switches.

Handle the problem by replacing that mechanism with its equivalent from
SCHED_ULE called time slice. It is effectively the same, just measured in
context of stathz instead of hz. Some unification is probably not bad.

1 files changed, 36 insertions, 33 deletions

diff --git a/sys/kern/sched_4bsd.c b/sys/kern/sched_4bsd.c
index 84fdb1f..91d00b2 100644
--- a/sys/kern/sched_4bsd.c
+++ b/sys/kern/sched_4bsd.c
@@ -94,6 +94,7 @@ struct td_sched {
 	fixpt_t		ts_pctcpu;	/* (j) %cpu during p_swtime. */
 	int		ts_cpticks;	/* (j) Ticks of cpu time. */
 	int		ts_slptime;	/* (j) Seconds !RUNNING. */
+	int		ts_slice;	/* Remaining part of time slice. */
 	int		ts_flags;
 	struct runq	*ts_runq;	/* runq the thread is currently on */
 #ifdef KTR
@@ -117,9 +118,9 @@ struct td_sched {
 static struct td_sched td_sched0;
 struct mtx sched_lock;
 
+static int	realstathz;	/* stathz is sometimes 0 and run off of hz. */
 static int	sched_tdcnt;	/* Total runnable threads in the system. */
-static int	sched_quantum;	/* Roundrobin scheduling quantum in ticks. */
-#define	SCHED_QUANTUM	(hz / 10)	/* Default sched quantum */
+static int	sched_slice = 1; /* Thread run time before rescheduling. */
 
 static void	setup_runqs(void);
 static void	schedcpu(void);
@@ -145,6 +146,10 @@ SYSINIT(schedcpu, SI_SUB_RUN_SCHEDULER, SI_ORDER_FIRST, kproc_start,
     &sched_kp);
 SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL);
 
+static void sched_initticks(void *dummy);
+SYSINIT(sched_initticks, SI_SUB_CLOCKS, SI_ORDER_THIRD, sched_initticks,
+    NULL);
+
 /*
  * Global run queue.
  */
@@ -179,31 +184,12 @@ setup_runqs(void)
 	runq_init(&runq);
 }
 
-static int
-sysctl_kern_quantum(SYSCTL_HANDLER_ARGS)
-{
-	int error, new_val;
-
-	new_val = sched_quantum * tick;
-	error = sysctl_handle_int(oidp, &new_val, 0, req);
-        if (error != 0 || req->newptr == NULL)
-		return (error);
-	if (new_val < tick)
-		return (EINVAL);
-	sched_quantum = new_val / tick;
-	hogticks = 2 * sched_quantum;
-	return (0);
-}
-
 SYSCTL_NODE(_kern, OID_AUTO, sched, CTLFLAG_RD, 0, "Scheduler");
 
 SYSCTL_STRING(_kern_sched, OID_AUTO, name, CTLFLAG_RD, "4BSD", 0,
     "Scheduler name");
-
-SYSCTL_PROC(_kern_sched, OID_AUTO, quantum, CTLTYPE_INT | CTLFLAG_RW,
-    0, sizeof sched_quantum, sysctl_kern_quantum, "I",
-    "Roundrobin scheduling quantum in microseconds");
-
+SYSCTL_INT(_kern_sched, OID_AUTO, slice, CTLFLAG_RW, &sched_slice, 0,
+    "Slice size for timeshare threads");
 #ifdef SMP
 /* Enable forwarding of wakeups to all other cpus */
 static SYSCTL_NODE(_kern_sched, OID_AUTO, ipiwakeup, CTLFLAG_RD, NULL,
@@ -471,9 +457,8 @@ schedcpu(void)
 	struct thread *td;
 	struct proc *p;
 	struct td_sched *ts;
-	int awake, realstathz;
+	int awake;
 
-	realstathz = stathz ? stathz : hz;
 	sx_slock(&allproc_lock);
 	FOREACH_PROC_IN_SYSTEM(p) {
 		PROC_LOCK(p);
@@ -645,14 +630,28 @@ sched_setup(void *dummy)
 {
 	setup_runqs();
 
-	if (sched_quantum == 0)
-		sched_quantum = SCHED_QUANTUM;
-	hogticks = 2 * sched_quantum;
+	/*
+	 * To avoid divide-by-zero, we set realstathz a dummy value
+	 * in case which sched_clock() called before sched_initticks().
+	 */
+	realstathz = hz;
+	sched_slice = realstathz / 10;	/* ~100ms */
 
 	/* Account for thread0. */
 	sched_load_add();
 }
 
+/*
+ * This routine determines the sched_slice after stathz and hz are setup.
+ */
+static void
+sched_initticks(void *dummy)
+{
+
+	realstathz = stathz ? stathz : hz;
+	sched_slice = realstathz / 10;	/* ~100ms */
+}
+
 /* External interfaces start here */
 
 /*
@@ -670,6 +669,7 @@ schedinit(void)
 	proc0.p_sched = NULL; /* XXX */
 	thread0.td_sched = &td_sched0;
 	thread0.td_lock = &sched_lock;
+	td_sched0.ts_slice = sched_slice;
 	mtx_init(&sched_lock, "sched lock", NULL, MTX_SPIN | MTX_RECURSE);
 }
 
@@ -686,9 +686,9 @@ sched_runnable(void)
 int
 sched_rr_interval(void)
 {
-	if (sched_quantum == 0)
-		sched_quantum = SCHED_QUANTUM;
-	return (sched_quantum);
+
+	/* Convert sched_slice from stathz to hz. */
+	return (hz / (realstathz / sched_slice));
 }
 
 /*
@@ -725,9 +725,10 @@ sched_clock(struct thread *td)
 	 * Force a context switch if the current thread has used up a full
 	 * quantum (default quantum is 100ms).
 	 */
-	if (!TD_IS_IDLETHREAD(td) &&
-	    ticks - PCPU_GET(switchticks) >= sched_quantum)
+	if (!TD_IS_IDLETHREAD(td) && (--ts->ts_slice <= 0)) {
+		ts->ts_slice = sched_slice;
 		td->td_flags |= TDF_NEEDRESCHED;
+	}
 
 	stat = DPCPU_PTR(idlestat);
 	stat->oldidlecalls = stat->idlecalls;
@@ -781,6 +782,7 @@ sched_fork_thread(struct thread *td, struct thread *childtd)
 	ts = childtd->td_sched;
 	bzero(ts, sizeof(*ts));
 	ts->ts_flags |= (td->td_sched->ts_flags & TSF_AFFINITY);
+	ts->ts_slice = 1;
 }
 
 void
@@ -1077,6 +1079,7 @@ sched_wakeup(struct thread *td)
 	}
 	td->td_slptick = 0;
 	ts->ts_slptime = 0;
+	ts->ts_slice = sched_slice;
 	sched_add(td, SRQ_BORING);
 }