clean up exit.

to clean up printing, it's important to first make the console
synchronous again.  that way we don't drop messages as we
lose the ability to run the console print process.  since we are
synchronous and not depending on interrupts, all the spl
manipulation and delays can be removed.

active.exiting does not need to be protected by a lock.  in fact,
active.machs is the only element that needs lock protection.
cleanup related to this should go back into the nix kernel.

finally, it's important to park the APs carefully.  first put them
in a delay loop with interrupts turned off.  then use launchhalt()
to turn them off, or whatever the architecture allows.  in our case
we send an INIT IPI to all-but-self.  (our case being amd64.)

Reference: /n/atom/patch/applied2013/fshalt
Date: Wed Sep  4 18:57:01 CES 2013
Signed-off-by: quanstro@quanstro.net

--- /sys/src/fs/amd64/fns.h	Wed Sep  4 18:50:47 2013
+++ /sys/src/fs/amd64/fns.h	Wed Sep  4 18:50:48 2013
@@ -21,6 +21,7 @@
 u64int	getcr3(void);
 u64int	getcr4(void);
 void	halt(void);
+void	halthi(void);
 void	hardhalt(void);
 void	i8042a20(void);
 void	i8042reset(void);
--- /sys/src/fs/amd64/l64v.s	Wed Sep  4 18:50:49 2013
+++ /sys/src/fs/amd64/l64v.s	Wed Sep  4 18:50:49 2013
@@ -356,6 +356,10 @@
 	HLT
 	RET
 
+TEXT halthi(SB), 1, $-4
+	HLT
+	RET
+
 TEXT _monitor(SB), 1, $-4			/* void monitor(void*); */
 	MOVQ	RARG, AX			/* linear address to monitor */
 	XORQ	CX, CX				/* no optional extensions yet */
--- /sys/src/fs/port/devcons.c	Wed Sep  4 18:50:50 2013
+++ /sys/src/fs/port/devcons.c	Wed Sep  4 18:50:51 2013
@@ -12,6 +12,7 @@
 	uint	n;
 	uint	seq;
 	uint	init;
+	uint	bufprint;
 } printq;
 
 static
@@ -73,6 +74,27 @@
 	early = 0;
 }
 
+void
+bufprint(int on)
+{
+	ilock(&printq);
+	switch(on<<1 | printq.bufprint){
+	case 1<<1 | 0:
+	case 1<<1 | 1:
+		printq.bufprint = 1;
+		break;
+	case 0<<1 | 0:
+		break;
+	case 0<<1 | 1:
+		printq.bufprint = 0;
+		waveprint("%.*s", printq.n, (char*)printq.buf);
+		printq.n = 0;
+		printq.seq++;
+		break;
+	}
+	iunlock(&printq);
+}
+
 /*
  * Print a string on the console.
  */
@@ -81,7 +103,7 @@
 {
 	uint t;
 
-	if(early || predawn){
+	if(early || printq.bufprint==0){
 		waveprint("%.*s", n, s);
 		return;
 	}
@@ -252,7 +274,7 @@
 	va_end(arg);
 	buf[n] = '\n';
 	putstrn(buf, n+1);
-	if(!predawn){
+	if(printq.bufprint){
 		spllo();
 		prflush();
 	}
@@ -267,10 +289,11 @@
 {
 	int i, lo;
 
-	if(predawn)
-		return;
-
 	ilock(&printq);
+	if(printq.bufprint == 0){
+		iunlock(&printq);
+		return;
+	}
 	i = printq.seq;
 	iunlock(&printq);
 
--- /sys/src/fs/port/main.c	Wed Sep  4 18:50:51 2013
+++ /sys/src/fs/port/main.c	Wed Sep  4 18:50:52 2013
@@ -76,7 +76,7 @@
 		clockinit();
 
 		print("Plan 9 fileserver\n");
-		print("\t%d-bit, %d-byte blocks\n", sizeof(Off)*8-1, RBUFSIZE);
+		print("\t%d-bit, %d-byte blocks, %s\n", sizeof(Off)*8-1, RBUFSIZE, Archname);
 
 		profinit();
 
@@ -111,6 +111,7 @@
 		userinit(touser, 0, "ini");
 
 	predawn = 0;
+	bufprint(1);
 	wakeup(&dawnrend);
 	launchinit();
 	schedinit();
@@ -292,35 +293,34 @@
 void
 exit(void)
 {
+	bufprint(0);
+	active.exiting = 1;
+
 	if(m->machno == 0)
 		doexits();
-	u = 0;
-	lock(&active);
-	active.machs &= ~(1<<m->machno);
-	active.exiting = 1;
-	unlock(&active);
-	if(!predawn)
-		spllo();
+	u = nil;
+	splhi();
 	print("cpu %d exiting\n", m->machno);
+
+	ilock(&active);
+	active.machs &= ~(1<<m->machno);
+	iunlock(&active);
+
 	while(active.machs)
 		delay(1);
 	while(m->machno != 0)
-		delay(1);
+		halthi();
+
+	launchhalt();
 
 	print("halted at %T.\npress a key to reboot sooner than %d mins.\n",
 		time(), Keydelay/60);
-	delay(500);		/* time to drain print q */
 
-	splhi();
 	/* reboot after delay (for debugging) or at key press */
 	rawchar(Keydelay);
 
-	spllo();
-	delay(500);		/* time to drain echo q */
 	print("rebooting...\n");
-	delay(500);		/* time to drain print q */
 
-	splhi();
 	consreset();
 	firmware();
 }