boot-qemu.py: Do not add '-no-reboot' unconditionally

[nathanchance]

This effectively reverts commit 8c78e38 ("boot-qemu: Always use
-no-reboot") against the rewrite that occurred in #91.

By applying this unconditionally, we potentially miss out on catching
panics. Prior to that change, if a machine panicked and rebooted, it
would keep rebooting until timeout killed QEMU, resulting in a non-zero
exit code. After that change, QEMU just exits cleanly on reboot.

cc @kees, I cannot remember what the original reason for unconditionally adding
-no-reboot was.

[nickdesaulniers]

LGTM; thanks! 🌥️ 🌔

[kees]

It was exactly that: to convert panics into errors instead of timeouts (and to avoid the overhead of waiting for the timeouts)

[nickdesaulniers]

So I think we're still missing panics: https://github.com/ClangBuiltLinux/continuous-integration2/actions/runs/4729226155/jobs/8392326831

[nathanchance]

As mentioned in the CI issue, boot-utils in CI has not been updated to include this, so it is expected that we are still missing panics.

[nickdesaulniers]

It was exactly that: to convert panics into errors instead of timeouts (and to avoid the overhead of waiting for the timeouts)

The issue is that setting -no-reboot does not set the return code of qemu to indicate failure.

I just hit this where we have a boot loop in ClangBuiltLinux/linux#1848, so I added

diff --git a/boot-qemu.py b/boot-qemu.py
index 24a0f0f59aa4..1052ca1ddf16 100755
--- a/boot-qemu.py
+++ b/boot-qemu.py
@@ -399,6 +399,7 @@ class ARM64QEMURunner(QEMURunner):
         self._default_kernel_path = Path('arch/arm64/boot/Image.gz')
         self._initrd_arch = 'arm64'
         self._qemu_arch = 'aarch64'
+        self._qemu_args.append('-no-reboot')
 
     def _get_cpu_val(self):
         cpu = ['max']

This results in the machine exiting cleanly rather than looping but:

$ /android1/boot-utils/boot-qemu.py -k .
...
[    0.027803][    T0] ---[ end trace 0000000000000000 ]---
[    0.028222][    T0] Kernel panic - not syncing: Oops - Undefined instruction: Fatal exception
$ echo $?
0

Based on https://unix.stackexchange.com/a/443060, mentions pvpanic, but that requires multiple configs to be enabled (PVPANIC, PVPANIC_PCI) and setting -device pvpanic-pci (I haven't found which -action value works best yet). https://blogs.oracle.com/linux/post/an-introduction-to-pvpanic . I wonder if we should consider enabling those in arch/arm64/configs/virt.config?