from netbsd:

Deal with GCC's dead code elimination being suboptimal.
Modify splraise() to allow better optimization.
Make cpl, ipending, and astpending volatile.
Make sure interrupts are disabled before jumping to a resume point,
to prevent races.
Make FPU faults use INTRFASTEXIT, and remove INTREXIT.
Build the frame for recursive interrupts manually, and make sure to
disable interrupts to avoid races.
VS: ----------------------------------------------------------------------

5 files changed, 45 insertions, 77 deletions

diff --git a/sys/arch/i386/i386/locore.s b/sys/arch/i386/i386/locore.s
index 9b2bdc25188..3fe77734a0e 100644
--- a/sys/arch/i386/i386/locore.s
+++ b/sys/arch/i386/i386/locore.s
@@ -1,4 +1,4 @@
-/*	$NetBSD: locore.s,v 1.139.2.1 1995/10/24 16:32:42 mycroft Exp $	*/
+/*	$NetBSD: locore.s,v 1.142 1996/01/07 03:59:28 mycroft Exp $	*/
 
 #undef DIAGNOSTIC
 #define DIAGNOSTIC
@@ -101,8 +101,6 @@
 	movl	$GSEL(GDATA_SEL, SEL_KPL),%eax	; \
 	movl	%ax,%ds		; \
 	movl	%ax,%es
-#define	INTREXIT \
-	jmp	_Xdoreti
 #define	INTRFASTEXIT \
 	popl	%es		; \
 	popl	%ds		; \
@@ -1530,7 +1528,7 @@ ENTRY(cpu_switch)
 	movl	$0,_curproc
 
 	movl	$0,_cpl			# spl0()
-	call	_spllower		# process pending interrupts
+	call	_Xspllower		# process pending interrupts
 
 switch_search:
 	/*
@@ -1877,7 +1875,7 @@ IDTVEC(fpu)
 	incl	_cnt+V_TRAP
 	call	_npxintr
 	addl	$4,%esp
-	INTREXIT
+	INTRFASTEXIT
 #else
 	ZTRAP(T_ARITHTRAP)
 #endif
diff --git a/sys/arch/i386/include/psl.h b/sys/arch/i386/include/psl.h
index 4160fe81f04..52747892ae2 100644
--- a/sys/arch/i386/include/psl.h
+++ b/sys/arch/i386/include/psl.h
@@ -1,4 +1,4 @@
-/*	$NetBSD: psl.h,v 1.22 1995/10/11 04:20:23 mycroft Exp $	*/
+/*	$NetBSD: psl.h,v 1.25 1996/01/07 03:59:32 mycroft Exp $	*/
 
 /*-
  * Copyright (c) 1990 The Regents of the University of California.
@@ -99,7 +99,10 @@
 
 #ifndef LOCORE
 
-int cpl, ipending, astpending, imask[5];
+volatile int cpl, ipending, astpending;
+int imask[5];
+
+extern void Xspllower __P((void));
 
 /*
  * Add a mask to cpl, and return the old value of cpl.
@@ -109,28 +112,38 @@ splraise(ncpl)
 	register int ncpl;
 {
 	register int ocpl = cpl;
-	cpl |= ncpl;
+
+	cpl = ocpl | ncpl;
 	return (ocpl);
 }
 
-extern void spllower __P((void));
-
 /*
  * Restore a value to cpl (unmasking interrupts).  If any unmasked
- * interrupts are pending, call spllower() to process them.
- *
- * NOTE: We go to the trouble of returning the old value of cpl for
- * the benefit of some splsoftclock() callers.  This extra work is
- * usually optimized away by the compiler.
+ * interrupts are pending, call Xspllower() to process them.
  */
-static __inline int
+static __inline void
 splx(ncpl)
 	register int ncpl;
 {
+
+	cpl = ncpl;
+	if (ipending & ~ncpl)
+		Xspllower();
+}
+
+/*
+ * Same as splx(), but we return the old value of spl, for the
+ * benefit of some splsoftclock() callers.
+ */
+static __inline int
+spllower(ncpl)
+	register int ncpl;
+{
 	register int ocpl = cpl;
+
 	cpl = ncpl;
 	if (ipending & ~ncpl)
-		spllower();
+		Xspllower();
 	return (ocpl);
 }
 
@@ -150,7 +163,7 @@ splx(ncpl)
  * NOTE: splsoftclock() is used by hardclock() to lower the priority from
  * clock to softclock before it calls softclock().
  */
-#define	splsoftclock()	splx(SIR_CLOCKMASK)
+#define	splsoftclock()	spllower(SIR_CLOCKMASK)
 #define	splsoftnet()	splraise(SIR_NETMASK)
 #define	splsofttty()	splraise(SIR_TTYMASK)
 
@@ -158,7 +171,7 @@ splx(ncpl)
  * Miscellaneous
  */
 #define	splhigh()	splraise(-1)
-#define	spl0()		splx(0)
+#define	spl0()		spllower(0)
 
 /*
  * Software interrupt registration
diff --git a/sys/arch/i386/isa/icu.h b/sys/arch/i386/isa/icu.h
index d9a7e144d28..b2cc5068c34 100644
--- a/sys/arch/i386/isa/icu.h
+++ b/sys/arch/i386/isa/icu.h
@@ -1,4 +1,4 @@
-/*	$NetBSD: icu.h,v 1.17 1994/11/04 19:13:49 mycroft Exp $	*/
+/*	$NetBSD: icu.h,v 1.18 1996/01/07 02:03:20 mycroft Exp $	*/
 
 /*-
  * Copyright (c) 1990 The Regents of the University of California.
@@ -53,19 +53,7 @@
  */
 extern	unsigned imen;		/* interrupt mask enable */
 
-#define	INTRUNMASK(msk,s)	(msk &= ~(s))
-#define	INTREN(s)		(INTRUNMASK(imen, s), SET_ICUS())
-#define	INTRMASK(msk,s)		(msk |= (s))
-#define	INTRDIS(s)		(INTRMASK(imen, s), SET_ICUS())
-#if 0
-#define SET_ICUS()	(outb(IO_ICU1 + 1, imen), outb(IU_ICU2 + 1, imen >> 8))
-#else
-/*
- * XXX - IO_ICU* are defined in isa.h, not icu.h, and nothing much bothers to
- * include isa.h, while too many things include icu.h.
- */
-#define SET_ICUS()	(outb(0x21, imen), outb(0xa1, imen >> 8))
-#endif
+#define SET_ICUS()	(outb(IO_ICU1 + 1, imen), outb(IO_ICU2 + 1, imen >> 8))
 
 #endif /* !LOCORE */
 
diff --git a/sys/arch/i386/isa/icu.s b/sys/arch/i386/isa/icu.s
index 1f5de650230..b6ffcd13c79 100644
--- a/sys/arch/i386/isa/icu.s
+++ b/sys/arch/i386/isa/icu.s
@@ -1,4 +1,4 @@
-/*	$NetBSD: icu.s,v 1.43 1995/10/11 04:20:31 mycroft Exp $	*/
+/*	$NetBSD: icu.s,v 1.45 1996/01/07 03:59:34 mycroft Exp $	*/
 
 /*-
  * Copyright (c) 1993, 1994, 1995 Charles M. Hannum.  All rights reserved.
@@ -64,7 +64,6 @@ _splx:
  *   esi - address to resume loop at
  *   edi - scratch for Xsoftnet
  */
-ENTRY(spllower)
 IDTVEC(spllower)
 	pushl	%ebx
 	pushl	%esi
@@ -103,6 +102,7 @@ IDTVEC(doreti)
 	bsfl    %eax,%eax               # slow, but not worth optimizing
 	btrl    %eax,_ipending
 	jnc     1b			# some intr cleared the in-memory bit
+	cli
 	jmp	*_Xresume(,%eax,4)
 2:	/* Check for ASTs on exit to user mode. */
 	cli
diff --git a/sys/arch/i386/isa/vector.s b/sys/arch/i386/isa/vector.s
index 0558ebbde2b..01acbfbe80d 100644
--- a/sys/arch/i386/isa/vector.s
+++ b/sys/arch/i386/isa/vector.s
@@ -1,4 +1,4 @@
-/*	$NetBSD: vector.s,v 1.29 1995/05/08 18:00:20 mycroft Exp $	*/
+/*	$NetBSD: vector.s,v 1.31 1996/01/07 03:59:35 mycroft Exp $	*/
 
 /*
  * Copyright (c) 1993, 1994, 1995 Charles M. Hannum.  All rights reserved.
@@ -111,7 +111,7 @@
  * XXX
  * The interrupt frame is set up to look like a trap frame.  This may be a
  * waste.  The only handler which needs a frame is the clock handler, and it
- * only needs a few bits.  doreti() needs a trap frame for handling ASTs, but
+ * only needs a few bits.  Xdoreti() needs a trap frame for handling ASTs, but
  * it could easily convert the frame on demand.
  *
  * The direct costs of setting up a trap frame are two pushl's (error code and
@@ -199,10 +199,15 @@ FAST(15, IO_ICU2, ENABLE_ICU1_AND_2)
  * interrupt.  On a system with level-triggered interrupts, we could terminate
  * immediately when one of them returns 1; but this is a PC.
  *
- * On exit, we jump to doreti, to process soft interrupts and ASTs.
+ * On exit, we jump to Xdoreti(), to process soft interrupts and ASTs.
  */
 #define	INTR(irq_num, icu, enable_icus) \
-IDTVEC(intr/**/irq_num)							;\
+IDTVEC(recurse/**/irq_num)						;\
+	pushfl								;\
+	pushl	%cs							;\
+	pushl	%esi							;\
+	cli								;\
+_Xintr/**/irq_num/**/:							;\
 	pushl	$0			/* dummy error code */		;\
 	pushl	$T_ASTFLT		/* trap # for doing ASTs */	;\
 	INTRENTRY							;\
@@ -235,7 +240,7 @@ _Xresume/**/irq_num/**/:						;\
 	jnz	7b							;\
 	STRAY_TEST			/* see if it's a stray */	;\
 5:	UNMASK(irq_num, icu)		/* unmask it in hardware */	;\
-	INTREXIT			/* lower spl and do ASTs */	;\
+	jmp	_Xdoreti		/* lower spl and do ASTs */	;\
 IDTVEC(stray/**/irq_num)						;\
 	pushl	$irq_num						;\
 	call	_isa_strayintr						;\
@@ -284,42 +289,6 @@ INTR(14, IO_ICU2, ENABLE_ICU1_AND_2)
 INTR(15, IO_ICU2, ENABLE_ICU1_AND_2)
 
 /*
- * Recursive interrupts.
- *
- * This is a somewhat nasty hack to deal with resuming interrupts from splx().
- * We can't just jump to the resume point, because some handlers require an
- * interrupt frame.  Instead, we just recursively interrupt.
- *
- * On entry, %esi contains a pointer to where we need to return.  This is a
- * bit faster than a call/ret/jmp to continue the loop.
- *
- * XXX
- * It might be a little faster to build the interrupt frame manually and jump
- * to the resume point.  The code would be larger, though.
- */
-#define	RECURSE(irq_num) \
-IDTVEC(recurse/**/irq_num)						;\
-	int	$(ICU_OFFSET + irq_num)					;\
-	jmp	%esi
-
-RECURSE(0)
-RECURSE(1)
-RECURSE(2)
-RECURSE(3)
-RECURSE(4)
-RECURSE(5)
-RECURSE(6)
-RECURSE(7)
-RECURSE(8)
-RECURSE(9)
-RECURSE(10)
-RECURSE(11)
-RECURSE(12)
-RECURSE(13)
-RECURSE(14)
-RECURSE(15)
-
-/*
  * These tables are used by the ISA configuration code.
  */
 /* interrupt service routine entry points */
@@ -334,7 +303,7 @@ IDTVEC(fast)
 	.long   _Xfast13, _Xfast14, _Xfast15
 
 /*
- * These tables are used by doreti() and spllower().
+ * These tables are used by Xdoreti() and Xspllower().
  */
 /* resume points for suspended interrupts */
 IDTVEC(resume)