[Bug 47800] New: Better detect Windows reboots

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47800

Bug ID: 47800
           Summary: Better detect Windows reboots
           Product: Wine-Testbot
           Version: unspecified
          Hardware: x86
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: unknown
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

When a test crashes Windows in a VM we get typically get an error like this:

channel 2: open failed: connect failed: Connection refused
channel 2: open failed: connect failed: Connection refused
channel 2: open failed: connect failed: Connection refused
An error occurred while waiting for the test to complete: network read got a
premature EOF (wait2/connect:AgentVersion.h:0/9)
The test VM has crashed, rebooted or lost connectivity (or the TestAgent server
died)
The previous 2 run(s) terminated abnormally

Nowadays most VMs autologin and autostart the TestAgent server with the
--show-restarts option. What this does is pop up a dialog saying:

TestAgentd.exe was restarted (2). Did Windows reboot?

This is a telltale sign that Windows indeed rebooted (or that the TestBot
administrator incorrectly set up the VM but then you'd see that dialog in all
screenshots of that VM).

The way this works is that when given the --show-restarts option the TestAgent
server increases a persistent counter every time it starts. Normally the
TestBot administrator will have reset that counter so it will be set to one
such that a reboot will push it to 2 prompting this dialog.

What's interesting is that this counter is accessible from the client with
getproperties("start.counter"). So the WineRun*.pl script can know if the VM
was indeed rebooted.

But from the log above we also notice that they don't leave enough time for the
VM to reboot and thus fail to reestablish the connection to the TestAgent
server. This could be solved by adding a call to
SetConnectionTimeout(,,$WaitForBoot) in the strategic places.

Then in most cases the WineRun*.pl scripts could conclusively say that Windows
crashed and rebooted.

[Bug 47800] Better detect Windows reboots

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=47800

--- Comment #1 from François Gouget  ---
This counter is actually a really annoying when setting up VMs: forget to reset
it before taking a snapshot and LibvirtTool will rightly complain that it is
set wrong.

Maybe that could be avoided by having LibvirtTool match changes in this counter
to the reboots it triggered and reset it when appropriate. Or maybe another
mechanism should be used.

One thing that often prevents detecting that the VM rebooted is Windows'
default 30 second boot delay which causes the test to time out and give up
before it had time to reconnect to the TestAgent server. So it would help if
LibvirtTool could automatically reduce this delay to ~2 seconds when setting up
"-live" snapshots.

One more thing is that it's possible to get information about the reason for
the reboot from the event log. If there was a BSOD one can even get some
information about it... if there's at least a small pagefile set up.

Removing "options kvm ignore_msrs=1" from /etc/modprobe.d and running
ntdll:exception on a Windows 10 VM is a nice way to produce BSODs to test this.