*use atomic instructions for counting, rather than locking
*don't sched unless there's just a single processor when there is process priority inversion
*iprint in atomic context

Reference: /n/atom/patch/applied/qiosched
Date: Mon Aug 17 06:15:32 CES 2015
Signed-off-by: quanstro@quanstro.net

--- /sys/src/nix/port/qio.c	Mon Aug 17 06:14:13 2015
+++ /sys/src/nix/port/qio.c	Mon Aug 17 06:14:15 2015
@@ -31,7 +31,6 @@
 };
 
 static uvlong count[Ncount];
-static Lock countlk;
 
 static int debugging;
 
@@ -76,20 +75,19 @@
 
 uint	qiomaxatomic = Maxatomic;
 
+extern uvlong ainc64(uvlong*);	/* sic */
+extern uvlong aadd64(uvlong*, uvlong);	/* sic */
+
 static void
 counter(int i)
 {
-	ilock(&countlk);
-	count[i]++;
-	iunlock(&countlk);
+	ainc64(count+i);
 }
 
 static void
 bytes(int i, int n)
 {
-	ilock(&countlk);
-	count[i] += n;
-	iunlock(&countlk);
+	aadd64(count+i, n);
 }
 
 char*
@@ -218,7 +216,7 @@
 	int len;
 	Block *nb, *f;
 
-	if(bp->next == 0)
+	if(bp->next == nil)
 		return bp;
 
 	nb = allocb(blocklen(bp));
@@ -1277,7 +1275,7 @@
 		p = wakeup(&q->rr);
 
 		/* if we just wokeup a higher priority process, let it run */
-		if(p != nil && p->priority > up->priority){
+		if(p != nil && p->priority > up->priority && sys->nproc == 1){
 			counter(Cqbwritepricnt);
 			sched();
 		}
@@ -1322,7 +1320,7 @@
 	uchar *p = vp;
 
 	QDEBUG if(!islo())
-		print("qwrite hi %#p\n", getcallerpc(&q));
+		iprint("qwrite hi %#p\n", getcallerpc(&q));
 
 	sofar = 0;
 	do {