Add more chance to process IPI in the interrupt service routine.

This prevents "luna88k_ext_int: cpu0 level 1 interrupt" message on
heavy load.

"This makes sense" miod@, tested by me.

diff --git a/sys/arch/luna88k/luna88k/machdep.c b/sys/arch/luna88k/luna88k/machdep.c
index 44bc63c..e2f6350 100644 (file)
--- a/sys/arch/luna88k/luna88k/machdep.c
+++ b/sys/arch/luna88k/luna88k/machdep.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: machdep.c,v 1.139 2022/10/14 20:53:18 aoyama Exp $    */
+/*     $OpenBSD: machdep.c,v 1.140 2022/10/25 11:39:33 aoyama Exp $    */
 /*
  * Copyright (c) 1998, 1999, 2000, 2001 Steve Murphree, Jr.
  * Copyright (c) 1996 Nivas Madhur
@@ -859,6 +859,17 @@ luna88k_ext_int(struct trapframe *eframe)
                        if (CPU_IS_PRIMARY(ci))
                                isrdispatch_autovec(cur_int_level);
                        break;
+#ifdef MULTIPROCESSOR
+               case 1:
+                       /*
+                        * Another processor may have sent us an IPI
+                        * while we were servicing a device interrupt.
+                        */
+                       set_psr(get_psr() | PSR_IND);
+                       luna88k_ipi_handler(eframe);
+                       set_psr(get_psr() & ~PSR_IND);
+                       break;
+#endif
                default:
                        printf("%s: cpu%d level %d interrupt.\n",
                                __func__, ci->ci_cpuid, cur_int_level);