From b5edcab20bb6cf38c86e2113bd75f59ae5e65982 Mon Sep 17 00:00:00 2001
From: Sebastian Ene <sebastianene@google.com>
Date: Fri, 7 Oct 2022 14:27:08 +0000
Subject: [PATCH] devices: vmwdt: Add watchdog reset reboot reason

When we detect a VCPU stall we propagate the watchdog reboot error code
to the event tube. The newly added error code WATCHDOG_REBOOT = 36 is
returned from the crovm process.

Bug: 245900797
Test: manual testing, build and Virtualization apk on a device
Signed-off-by: Sebastian Ene <sebastianene@google.com>
Change-Id: Ib1ce97de911784b33d130d40536be26813edc3d7
Reviewed-on: https://chromium-review.googlesource.com/c/crosvm/crosvm/+/3936647
Reviewed-by: Daniel Verkamp <dverkamp@chromium.org>
---
 base/src/lib.rs             | 1 +
 devices/src/vmwdt.rs        | 4 ++--
 src/crosvm/sys/unix.rs      | 5 +++++
 src/crosvm/sys/unix/vcpu.rs | 1 +
 src/main.rs                 | 7 +++++++
 src/sys/windows.rs          | 5 +++++
 6 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/base/src/lib.rs b/base/src/lib.rs
index e8685443bc..136d43f00e 100644
--- a/base/src/lib.rs
+++ b/base/src/lib.rs
@@ -191,4 +191,5 @@ pub enum VmEventType {
     Reset,
     Crash,
     Panic(u8),
+    WatchdogReset,
 }
diff --git a/devices/src/vmwdt.rs b/devices/src/vmwdt.rs
index 72a990a067..994632d76b 100644
--- a/devices/src/vmwdt.rs
+++ b/devices/src/vmwdt.rs
@@ -176,7 +176,7 @@ impl Vmwdt {
                         } else {
                             // The guest ran but it did not send the periodic event
                             if let Err(_e) =
-                                reset_evt_wrtube.send::<VmEventType>(&VmEventType::Reset)
+                                reset_evt_wrtube.send::<VmEventType>(&VmEventType::WatchdogReset)
                             {
                                 error!("failed to send reset event from vcpu {}", cpu_id)
                             }
@@ -409,7 +409,7 @@ mod tests {
         // Verify that our timer expired and the next_expiration_interval_ms changed
         match vm_evt_rdtube.recv::<VmEventType>() {
             Ok(vm_event) => {
-                assert!(vm_event == VmEventType::Reset);
+                assert!(vm_event == VmEventType::WatchdogReset);
             }
             Err(_e) => {
                 panic!();
diff --git a/src/crosvm/sys/unix.rs b/src/crosvm/sys/unix.rs
index 0f7f567d96..981419aa70 100644
--- a/src/crosvm/sys/unix.rs
+++ b/src/crosvm/sys/unix.rs
@@ -1213,6 +1213,7 @@ pub enum ExitState {
     Stop,
     Crash,
     GuestPanic,
+    WatchdogReset,
 }
 // Remove ranges in `guest_mem_layout` that overlap with ranges in `file_backed_mappings`.
 // Returns the updated guest memory layout.
@@ -2465,6 +2466,10 @@ fn run_control<V: VmArch + 'static, Vcpu: VcpuArch + 'static>(
                                 info!("Guest reported panic [Code: {}]", pvpanic_code);
                                 break_to_wait = false;
                             }
+                            VmEventType::WatchdogReset => {
+                                info!("vcpu stall detected");
+                                exit_state = ExitState::WatchdogReset;
+                            }
                         },
                         Err(e) => {
                             warn!("failed to recv VmEvent: {}", e);
diff --git a/src/crosvm/sys/unix/vcpu.rs b/src/crosvm/sys/unix/vcpu.rs
index f42aed3e3c..c0a1595fb3 100644
--- a/src/crosvm/sys/unix/vcpu.rs
+++ b/src/crosvm/sys/unix/vcpu.rs
@@ -733,6 +733,7 @@ where
                 ExitState::Crash => VmEventType::Crash,
                 // vcpu_loop doesn't exit with GuestPanic.
                 ExitState::GuestPanic => unreachable!(),
+                ExitState::WatchdogReset => VmEventType::WatchdogReset,
             };
             if let Err(e) = vm_evt_wrtube.send::<VmEventType>(&final_event_data) {
                 error!(
diff --git a/src/main.rs b/src/main.rs
index f984521457..da10f1d0f2 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -90,6 +90,8 @@ enum CommandStatus {
     GuestPanic = 34,
     /// Invalid argument was given to crosvm.
     InvalidArgs = 35,
+    /// VM exit due to vcpu stall detection.
+    WatchdogReset = 36,
 }
 
 impl CommandStatus {
@@ -100,6 +102,7 @@ impl CommandStatus {
             Self::VmCrash => "exiting with crash",
             Self::GuestPanic => "exiting with guest panic",
             Self::InvalidArgs => "invalid argument",
+            Self::WatchdogReset => "exiting with watchdog reset",
         }
     }
 }
@@ -122,6 +125,10 @@ fn to_command_status(result: Result<sys::ExitState>) -> Result<CommandStatus> {
             info!("crosvm has exited due to a kernel panic in guest");
             Ok(CommandStatus::GuestPanic)
         }
+        Ok(sys::ExitState::WatchdogReset) => {
+            info!("crosvm has exited due to watchdog reboot");
+            Ok(CommandStatus::WatchdogReset)
+        }
         Err(e) => {
             error!("crosvm has exited with error: {:#}", e);
             Err(e)
diff --git a/src/sys/windows.rs b/src/sys/windows.rs
index cf0f708b0a..20fdfb5cfa 100644
--- a/src/sys/windows.rs
+++ b/src/sys/windows.rs
@@ -233,6 +233,7 @@ pub enum ExitState {
     Crash,
     #[allow(dead_code)]
     GuestPanic,
+    WatchdogReset,
 }
 
 type DeviceResult<T = VirtioDeviceStub> = Result<T>;
@@ -966,6 +967,10 @@ fn run_control<V: VmArch + 'static, Vcpu: VcpuArch + 'static>(
                             VmEventType::Panic(_) => {
                                 error!("got pvpanic event. this event is not expected on Windows.");
                             }
+                            VmEventType::WatchdogReset => {
+                                info!("vcpu stall detected");
+                                exit_state = ExitState::WatchdogReset;
+                            }
                         }
                         break 'poll;
                     }