boot-qemu: Always use -no-reboot

[kees]

To make things easier for shutting down after a reboot (or a reboot on
panic), always use -no-reboot regardless of architecture.

Signed-off-by: Kees Cook [email protected]

[msfjarvis]

This looks reasonable from what I've read about -no-reboot and -no-shutdown but I'll let someone who uses this on the daily merge the PR.

[nathanchance]

Hmmm, we might want to rethink doing this unconditionally, as I think it will cause us to miss out on timeouts for certain crashes. android-4.14-stable has been failing to boot for two months but there is no failure because the machine does not continuously reboot, which would trigger the timeout.

[    0.902647] Freeing SMP alternatives memory: 132K
[    0.937461] Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in:         (ptrval)
[    0.937461] 
[    0.938272] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.14.305 #1
[    0.938448] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Arch Linux 1.16.1-1-1 04/01/2014
[    0.938786] Call Trace:
[    0.940000]  panic+0x11b/0x570
[    0.940000]  ? acpi_read_bit_register+0x66/0xa8
[    0.940000]  ? start_kernel+0xc5d/0xc90
[    0.940000]  ? early_idt_handler_array+0x120/0x120
[    0.940000]  __stack_chk_fail+0x14/0x20
[    0.940000]  start_kernel+0xc5d/0xc90
[    0.940000]  x86_64_start_reservations+0x30a/0x310
[    0.940000]  x86_64_start_kernel+0x44a/0x450
[    0.940000]  secondary_startup_64+0xa5/0xb0
[    0.940000] Rebooting in 5 seconds..

[kees]

But why didn't the running notice it never reached userspace?

[nathanchance]

But why didn't the running notice it never reached userspace?

Because QEMU exited cleanly, which seems expected to me in that case with -no-reboot (the machine goes to reboot but we told QEMU to just exit on reboot). Without timeout, we would have to always use pvpanic (I think?) to catch these issues, which won't scale with CI. It is possible that -action panic=exit-failure might do what we want but I am not sure that will catch everything, I will try to play around with it later.

[ojeda]

What about matching the log against some string that you print within QEMU at the end of the run (and checking it is there only once)? Poor man's solution, but it may be a good idea and can be done in addition to another approach too. We did something like that for tests etc.

[nathanchance]

What about matching the log against some string that you print within QEMU at the end of the run (and checking it is there only once)? Poor man's solution, but it may be a good idea and can be done in addition to another approach too. We did something like that for tests etc.

Thanks for the suggestion, I have gone ahead and filed #96 to further discuss that.