feat: enable hardware watchdog on 700/800 series controllers (#6954)

packages/core/src/error/ZWaveError.ts

@@ -14,7 +14,7 @@ export enum ZWaveErrorCodes {
 
 	/** The driver failed to start */
 	Driver_Failed = 100,
-	Driver_Reset,
+	Driver_Reset, // FIXME: This is not used
 	Driver_Destroyed,
 	Driver_NotReady,
 	Driver_InvalidDataReceived,
@@ -35,6 +35,9 @@ export enum ZWaveErrorCodes {
 	Controller_ResponseNOK,
 	Controller_CallbackNOK,
 	Controller_Jammed,
+	/** The controller was reset in the middle of a Serial API command */
+	Controller_Reset,
+
 	Controller_InclusionFailed,
 	Controller_ExclusionFailed,
 
@@ -294,6 +297,15 @@ export function isMissingControllerACK(
 		&& e.context === "ACK";
 }
 
+export function wasControllerReset(
+	e: unknown,
+): e is ZWaveError & {
+	code: ZWaveErrorCodes.Controller_Reset;
+} {
+	return isZWaveError(e)
+		&& e.code === ZWaveErrorCodes.Controller_Reset;
+}
+
 export function isMissingControllerResponse(
 	e: unknown,
 ): e is ZWaveError & {

packages/serial/src/message/Constants.ts

@@ -157,9 +157,9 @@ export enum FunctionType {
 
 	UNKNOWN_FUNC_UNKNOWN_0xB4 = 0xb4, // ??
 
-	UNKNOWN_FUNC_WATCH_DOG_ENABLE = 0xb6,
-	UNKNOWN_FUNC_WATCH_DOG_DISABLE = 0xb7,
-	UNKNOWN_FUNC_WATCH_DOG_KICK = 0xb8,
+	EnableWatchdog500 = 0xb6, // Enable Watchdog (500 series and older)
+	DisableWatchdog500 = 0xb7, // Disable Watchdog (500 series and older)
+	KickWatchdog500 = 0xb8, // Kick Watchdog (500 series and older)
 	UNKNOWN_FUNC_UNKNOWN_0xB9 = 0xb9, // ??
 	UNKNOWN_FUNC_RF_POWERLEVEL_GET = 0xba, // Get RF Power level
 
@@ -170,8 +170,9 @@ export enum FunctionType {
 	FUNC_ID_ZW_SET_PROMISCUOUS_MODE = 0xd0, // Set controller into promiscuous mode to listen to all messages
 	FUNC_ID_PROMISCUOUS_APPLICATION_COMMAND_HANDLER = 0xd1,
 
-	UNKNOWN_FUNC_UNKNOWN_0xD2 = 0xd2, // ??
-	UNKNOWN_FUNC_UNKNOWN_0xD3 = 0xd3, // ??
+	StartWatchdog = 0xd2, // Start Hardware Watchdog (700 series and newer)
+	StopWatchdog = 0xd3, // Stop Hardware Watchdog (700 series and newer)
+
 	UNKNOWN_FUNC_UNKNOWN_0xD4 = 0xd4, // ??
 
 	Shutdown = 0xd9, // Instruct the Z-Wave API to shut down in order to safely remove the power

packages/serial/src/message/Message.ts

@@ -330,6 +330,12 @@ export class Message {
 		}
 	}
 
+	/** Tests whether this message expects an ACK from the controller */
+	public expectsAck(): boolean {
+		// By default, all commands expect an ACK
+		return true;
+	}
+
 	/** Tests whether this message expects a response from the controller */
 	public expectsResponse(): boolean {
 		return !!this.expectedResponse;

packages/zwave-js/src/lib/controller/Controller.ts

@@ -130,10 +130,7 @@ import {
 	ApplicationUpdateRequestSmartStartHomeIDReceived,
 	ApplicationUpdateRequestSmartStartLongRangeHomeIDReceived,
 } from "../serialapi/application/ApplicationUpdateRequest";
-import {
-	type SerialAPIStartedRequest,
-	SerialAPIWakeUpReason,
-} from "../serialapi/application/SerialAPIStartedRequest";
+
 import {
 	ShutdownRequest,
 	type ShutdownResponse,
@@ -216,6 +213,10 @@ import {
 	SetSerialApiTimeoutsRequest,
 	type SetSerialApiTimeoutsResponse,
 } from "../serialapi/misc/SetSerialApiTimeoutsMessages";
+import {
+	StartWatchdogRequest,
+	StopWatchdogRequest,
+} from "../serialapi/misc/WatchdogMessages";
 import {
 	AddNodeDSKToNetworkRequest,
 	AddNodeStatus,
@@ -449,10 +450,6 @@ export class ZWaveController
 			FunctionType.ReplaceFailedNode,
 			this.handleReplaceNodeStatusReport.bind(this),
 		);
-		driver.registerRequestHandler(
-			FunctionType.SerialAPIStarted,
-			this.handleSerialAPIStartedUnexpectedly.bind(this),
-		);
 	}
 
 	private _type: MaybeNotKnown<ZWaveLibraryTypes>;
@@ -739,6 +736,10 @@ export class ZWaveController
 	public get nodeIdType(): NodeIDType {
 		return this._nodeIdType;
 	}
+	/** @internal */
+	public set nodeIdType(value: NodeIDType) {
+		this._nodeIdType = value;
+	}
 
 	/** Returns the node with the given DSK */
 	public getNodeByDSK(dsk: Buffer | string): ZWaveNode | undefined {
@@ -1888,6 +1889,63 @@ export class ZWaveController
 		}
 	}
 
+	/**
+	 * Starts the hardware watchdog on supporting 700+ series controllers.
+	 * Returns whether the operation was successful.
+	 */
+	public async startWatchdog(): Promise<boolean> {
+		if (
+			this.sdkVersionGte("7.0")
+			&& this.isFunctionSupported(FunctionType.StartWatchdog)
+		) {
+			try {
+				this.driver.controllerLog.print(
+					"Starting hardware watchdog...",
+				);
+				await this.driver.sendMessage(
+					new StartWatchdogRequest(this.driver),
+				);
+
+				return true;
+			} catch (e) {
+				this.driver.controllerLog.print(
+					`Starting the hardware watchdog failed: ${
+						getErrorMessage(e)
+					}`,
+					"error",
+				);
+			}
+		}
+		return false;
+	}
+
+	/**
+	 * Stops the hardware watchdog on supporting controllers.
+	 * Returns whether the operation was successful.
+	 */
+	public async stopWatchdog(): Promise<boolean> {
+		if (this.isFunctionSupported(FunctionType.StopWatchdog)) {
+			try {
+				this.driver.controllerLog.print(
+					"Stopping hardware watchdog...",
+				);
+				await this.driver.sendMessage(
+					new StopWatchdogRequest(this.driver),
+				);
+
+				return true;
+			} catch (e) {
+				this.driver.controllerLog.print(
+					`Stopping the hardware watchdog failed: ${
+						getErrorMessage(e)
+					}`,
+					"error",
+				);
+			}
+		}
+		return false;
+	}
+
 	private _inclusionState: InclusionState = InclusionState.Idle;
 	public get inclusionState(): InclusionState {
 		return this._inclusionState;
@@ -4331,43 +4389,6 @@ supported CCs: ${
 		return false;
 	}
 
-	/**
-	 * Is called when the Serial API restart unexpectedly.
-	 */
-	private async handleSerialAPIStartedUnexpectedly(
-		msg: SerialAPIStartedRequest,
-	): Promise<boolean> {
-		// Normally, the soft reset command includes waiting for this message.
-		// If we end up here, it is unexpected.
-
-		switch (msg.wakeUpReason) {
-			// All wakeup reasons that indicate a reset of the Serial API
-			// need to be handled here, so we interpret node IDs correctly.
-			case SerialAPIWakeUpReason.Reset:
-			case SerialAPIWakeUpReason.WatchdogReset:
-			case SerialAPIWakeUpReason.SoftwareReset:
-			case SerialAPIWakeUpReason.EmergencyWatchdogReset:
-			case SerialAPIWakeUpReason.BrownoutCircuit: {
-				// The Serial API restarted unexpectedly
-				if (this._nodeIdType === NodeIDType.Long) {
-					this.driver.controllerLog.print(
-						`Serial API restarted unexpectedly.`,
-						"warn",
-					);
-
-					// We previously used 16 bit node IDs, but the controller was reset.
-					// Remember this and try to go back to 16 bit.
-					this._nodeIdType = NodeIDType.Short;
-					await this.trySetNodeIDType(NodeIDType.Long);
-				}
-
-				return true; // Don't invoke any more handlers
-			}
-		}
-
-		return false; // Not handled
-	}
-
 	private _rebuildRoutesProgress = new Map<number, RebuildRoutesStatus>();
 	/**
 	 * If routes are currently being rebuilt for the entire network, this returns the current progress.
@@ -6945,13 +6966,17 @@ ${associatedNodes.join(", ")}`,
 			);
 		}
 
+		// Disable watchdog to prevent resets during NVM access
+		await this.stopWatchdog();
+
 		let ret: Buffer;
 		try {
 			if (this.sdkVersionGte("7.0")) {
 				ret = await this.backupNVMRaw700(onProgress);
 				// All 7.xx versions so far seem to have a bug where the NVM is not properly closed after reading
 				// resulting in extremely strange controller behavior after a backup. To work around this, restart the stick if possible
 				await this.driver.trySoftReset();
+				// Soft-resetting will enable the watchdog again
 			} else {
 				ret = await this.backupNVMRaw500(onProgress);
 			}
@@ -7071,6 +7096,9 @@ ${associatedNodes.join(", ")}`,
 			);
 		}
 
+		// Disable watchdog to prevent resets during NVM access
+		await this.stopWatchdog();
+
 		// Restoring a potentially incompatible NVM happens in three steps:
 		// 1. the current NVM is read
 		// 2. the given NVM data is converted to match the current format