When -icount is used and a TB is recompiled due to an IO access

shortly after an IRQ has been raised, env->exception_index will still be set
to EXCP_IRQ when cpu_io_recompile calls cpu_resume_from_signal.
This causes qemu to repeat the IRQ trap, with disasterous consequences.

I suspect this "works" most of the time because linux tends to drop back to
svc mode before doing actual IRQ processing, and be fairly
tolerant of spurious IRQ traps.

Signed-off-by: Paul Brook <paul@codesourcery.com>

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@6100 c046a42c-6fe2-441c-8c8c-71466251a162

diff --git a/cpu-exec.c b/cpu-exec.c
index ed1545b..0a49a7e 100644 (file)
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -86,6 +86,7 @@ void cpu_resume_from_signal(CPUState *env1, void *puc)
 #endif
     }
 #endif
+    env->exception_index = -1;
     longjmp(env->jmp_env, 1);
 }