> While adding MP support to ddb on sparc64 I notice that ddb was
> inserting breakpoints even if you only entered ddb and used continue
> again. So I wonder if there is a cache synchronization issue of some
> sort and by adding those additional instructions you just manage to
> avoid it.
Locore is not consistent in its use of `isync' after disabling
exceptions. With the following diff, and without expanding the stmw
instructions, I can quit ddb without problems.
Index: locore.S
===================================================================
RCS file: /OpenBSD/src/sys/arch/macppc/macppc/locore.S,v
retrieving revision 1.50
diff -u -p -r1.50 locore.S
--- locore.S 29 Jul 2015 18:52:44 -0000 1.50
+++ locore.S 31 Oct 2015 08:20:16 -0000
@@ -175,6 +175,7 @@ switch_exited:
mfmsr %r30
andi. %r30,%r30,~PSL_EE@l
mtmsr %r30
+ isync
stw %r4,CI_CURPROC(%r5) /* record new process */
@@ -223,6 +224,7 @@ _ENTRY(_C_LABEL(cpu_idle_enter))
mfmsr %r3
andi. %r3,%r3,~PSL_EE@l
mtmsr %r3
+ isync
1:
blr
@@ -890,6 +892,7 @@ trapexit:
mfmsr %r3
andi. %r3,%r3,~PSL_EE@l
mtmsr %r3
+ isync
/* Test AST pending: */
lwz %r5,FRAME_SRR1+8(%r1)
mtcr %r5
