On 08/24/2011 03:38 PM, Anthony Liguori wrote:
diff --git a/hw/pc.c b/hw/pc.c
index 1c9d89a..4b07b35 100644
--- a/hw/pc.c
+++ b/hw/pc.c
@@ -549,8 +549,7 @@ static void bochs_bios_write(void *opaque,
uint32_t addr, uint32_t val)
/* LGPL'ed VGA BIOS messages */
case 0x501:
case 0x502:
- fprintf(stderr, "VGA BIOS panic, line %d\n", val);
- exit(1);
+ exit((val<< 1) | 1);
This code (before the patch) circumvents -no-shutdown.
Indeed. I believe that's a feature though?
Depends on what the user of -no-shutdown expects.
Shifting val left is surprising. What's wrong with even exit codes?
Hrm, the '| 1' is wrong. My intention was to make the bottom bit of
the exit code mean "if set to 0, this is an exit coming from a guest'.
Too subtle, IMO. I understand that we want to avoid a full qmp parser
for one-off unit tests, but using bit fields in the exit code?
Perhaps a python script that launches qemu with qmp and -no-shutdown,
listens for the guest shutdown event, and prints out the result? That
can be easily reused in test scripts.