Fix stupid bug in atomic_{add,sub}_int_nv_mp, and stupider bug in

atomic_cas_uint_mp.
Also, make the interprocessor interlock the only thing on its cache line.

diff --git a/sys/arch/m88k/m88k/atomic.S b/sys/arch/m88k/m88k/atomic.S
index 262b662..a99df71 100644 (file)
--- a/sys/arch/m88k/m88k/atomic.S
+++ b/sys/arch/m88k/m88k/atomic.S
@@ -1,4 +1,4 @@
-/*     $OpenBSD: atomic.S,v 1.5 2014/07/13 08:13:07 miod Exp $ */
+/*     $OpenBSD: atomic.S,v 1.6 2014/07/15 16:26:28 miod Exp $ */
 
 /*
  * Copyright (c) 2009 Miodrag Vallat.
@@ -18,13 +18,21 @@
 
 #include <machine/asm.h>
 
+#ifdef M88110
+#define        CACHE_LINE      32
+#else
+#define        CACHE_LINE      16
+#endif
        .data
 
 /*
- * A __cpu_simple_lock_t used to provide the inter-processor interlock.
+ * A __cpu_simple_lock_t used to provide the inter-processor interlock,
+ * alone on its cache line.
  */
+       .balign CACHE_LINE
 ASLOCAL(__atomic_interlock)
        .word   0
+       .balign CACHE_LINE
 
        .text
 
@@ -68,7 +76,7 @@ ENTRY(atomic_add_int_nv_mp)
 
        or      %r9, %r2, %r0
        ld      %r2, %r9, %r0
-       addu    %r2, %r2, %r4
+       addu    %r2, %r2, %r3
        st      %r2, %r9, %r0
 
        br      _C_LABEL(__atomic_unlock)
@@ -79,7 +87,7 @@ ENTRY(atomic_sub_int_nv_mp)
 
        or      %r9, %r2, %r0
        ld      %r2, %r9, %r0
-       subu    %r2, %r2, %r4
+       subu    %r2, %r2, %r3
        st      %r2, %r9, %r0
 
        br      _C_LABEL(__atomic_unlock)
@@ -90,7 +98,7 @@ ENTRY(atomic_cas_uint_mp)
 
        ld      %r9, %r2, %r0
        cmp     %r3, %r3, %r9
-       bcnd    ne0, %r3, 1f
+       bb0     eq,  %r3, 1f
        st      %r4, %r2, %r0
 1:
        or      %r2, %r9, %r0
@@ -122,14 +130,14 @@ ASLOCAL(__atomic_lock_88110)
         * then grab the interlock.
         */
 
+       or.u    %r6, %r0, %hi16(_ASM_LABEL(__atomic_interlock))
+       or      %r6, %r6, %lo16(_ASM_LABEL(__atomic_interlock))
+
        ldcr    %r7, PSR
        set     %r8, %r7, 1<PSR_INTERRUPT_DISABLE_BIT>
        set     %r8, %r8, 1<PSR_SHADOW_FREEZE_BIT>
        stcr    %r8, PSR
        FLUSH_PIPELINE
-
-       or.u    %r6, %r0, %hi16(_ASM_LABEL(__atomic_interlock))
-       or      %r6, %r6, %lo16(_ASM_LABEL(__atomic_interlock))
 1:
        or      %r9, %r0, 1     /* __SIMPLELOCK_LOCKED */
        xmem    %r9, %r6, %r0
@@ -149,13 +157,14 @@ GLOBAL(__atomic_lock_88100)
         * the interlock.
         */
 
+       or.u    %r6, %r0, %hi16(_ASM_LABEL(__atomic_interlock))
+       or      %r6, %r6, %lo16(_ASM_LABEL(__atomic_interlock))
+
        ldcr    %r7, PSR
        set     %r8, %r7, 1<PSR_INTERRUPT_DISABLE_BIT>
        stcr    %r8, PSR
        FLUSH_PIPELINE
 
-       or.u    %r6, %r0, %hi16(_ASM_LABEL(__atomic_interlock))
-       or      %r6, %r6, %lo16(_ASM_LABEL(__atomic_interlock))
 1:
        or      %r9, %r0, 1     /* __SIMPLELOCK_LOCKED */
        xmem    %r9, %r6, %r0