Re: [PULL 16/45] vl: Add option to avoid stopping VM upon guest panic

[Paolo Bonzini]

On 19/01/21 22:34, Peter Maydell wrote:
> We used to default to 'pause' and now we default to 'poweroff'.
>
> We noticed this because it broke an in-flight test case for
> the pvpanic-pci device from Mihai (which was expecting to see
> the device in 'pause' state and found it was now in 'poweroff').
> Test cases aren't very exciting, but was it really intentional
> to change the default behaviour? It's part of the user-facing
> surface of QEMU, so if we did intend a default change that ought
> really to be more clearly stated (and noted in the Changelog) I think.

To sum up the difference:

1) before, without -no-shutdown:
{"event": "GUEST_PANICKED", "data": {"action": "pause"}}
{"event": "STOP"}
{"event": "GUEST_PANICKED", "data": {"action": "poweroff"}}
{"event": "SHUTDOWN", "data": {"guest": true, "reason": "guest-panic"}}

after, without -no-shutdown:
{"event": "GUEST_PANICKED", "data": {"action": "poweroff"}}
{"event": "STOP"}
{"event": "SHUTDOWN", "data": {"guest": true, "reason": "guest-panic"}}

2) before, with -no-shutdown:
{"event": "GUEST_PANICKED", "data": {"action": "pause"}}
{"event": "STOP"}

after, with -no-shutdown (aka -action panic=pause,shutdown=pause):
{"event": "GUEST_PANICKED", "data": {"action": "pause"}}
{"event": "STOP"}


I think the new behavior without -no-shutdown is (albeit not 
intentionally) preferrable, since QEMU used to report twice that the 
guest panicked.

Looking again at the -action CLI, though, there are some inconsistencies 
in the naming.  I'll send a patches for them, but they will not affect 
the HMP output.

Paolo