host-reboot: enforce host boot count on host crash

The host reboot path can be triggered via paths outside of the normal
RequestedHostTransition property set. For example, if the host crashes,
it just triggers an automatic reboot using the systemd targets, which
means the check to prevent continuous host reboots is not run.

This commit ensures we decrement the reboot counter when a host crash
occurs, and it also ensures we check that count when deciding if a
reboot should occur.

While testing, it was found when we end in the host Quiesce state (after
all reboot attempts have expired), the host stayed in Quiesce, even
after issuing a power off. The fix for that was to ensure the
obmc-host-quiesce@.target conflicted with obmc-host-stop@.target to
ensure it would properly run on the power off request and update the
host state to Off.

Tested:
- Verified that AttemptsLeft was properly decremented on host crashes
- Verified that when AttempsLeft reached 0, the reboot was halted
  and the host state was moved to Quiesced
- Verified that once in Quiesced, system could be powered off, and a
  boot worked fine (and AttempsLeft was back to 3)

Signed-off-by: Andrew Geissler <geissonator@yahoo.com>
Change-Id: I8d101b987517e160becc712ed70950f1d3204397
diff --git a/host_state_manager.cpp b/host_state_manager.cpp
index bec0f0a..f1709cd 100644
--- a/host_state_manager.cpp
+++ b/host_state_manager.cpp
@@ -127,6 +127,7 @@
          fmt::format("obmc-host-reboot@{}.target", id)}
     };
 #endif
+    hostCrashTarget = fmt::format("obmc-host-crash@{}.target", id);
 }
 
 const std::string& Host::getTarget(HostState state)
@@ -352,6 +353,20 @@
         info("Received signal that host is in diagnostice mode");
         this->currentHostState(server::Host::HostState::DiagnosticMode);
     }
+    else if ((newStateUnit == hostCrashTarget) &&
+             (server::Host::currentHostState() ==
+              server::Host::HostState::Running))
+    {
+        // Only decrease the boot count if host was running when the host crash
+        // target was started. Systemd will sometimes trigger multiple
+        // JobNew events for the same target. This seems to be related to
+        // how OpenBMC utilizes the targets in the reboot scenario
+        info("Received signal that host has crashed, decrement reboot count");
+
+        // A host crash can cause a reboot of the host so decrement the reboot
+        // count
+        decrementRebootCount();
+    }
 }
 
 uint32_t Host::decrementRebootCount()