Bug: Edimax BLE Dongle Fails After Teardown and Re-Instantiation

This patch addresses an issue where the some RTK BLE dongles fail to perform
an HCI reset after the transport is torn down and re-instantiated. To address
that, we prevent crashing the background threads when invalid data comes in,
and time out if no response is received within a fixed amount of time. When
the timeout occurs, we retry the reset, and ultimately skip over reading the
local version information if that fails.

bumble/drivers/rtk.py

@@ -301,6 +301,8 @@ class Driver(common.Driver):
         fw_name: str = ""
         config_name: str = ""
 
+    POST_RESET_DELAY: float = 0.2
+
     DRIVER_INFOS = [
         # 8723A
         DriverInfo(
@@ -495,12 +497,22 @@ class Driver(common.Driver):
 
     @classmethod
     async def driver_info_for_host(cls, host):
-        await host.send_command(HCI_Reset_Command(), check_result=True)
+        try:
-        host.ready = True  # Needed to let the host know the controller is ready.
+            await host.send_command(
+                HCI_Reset_Command(), check_result=True, response_timeout=cls.POST_RESET_DELAY
+            )
+            host.ready = True  # Needed to let the host know the controller is ready.
+        except asyncio.exceptions.TimeoutError:
+            logger.warning("timeout waiting for hci reset, retrying")
+            await host.send_command(HCI_Reset_Command(), check_result=True)
+            host.ready = True
+
+        command = HCI_Read_Local_Version_Information_Command()
+        response = await host.send_command(command, check_result=True)
+        if response.command_opcode != command.op_code:
+            logger.error("failed to probe local version information")
+            return None
 
-        response = await host.send_command(
-            HCI_Read_Local_Version_Information_Command(), check_result=True
-        )
         local_version = response.return_parameters
 
         logger.debug(

bumble/host.py

@@ -514,7 +514,7 @@ class Host(AbortableEventEmitter):
         if self.hci_sink:
             self.hci_sink.on_packet(bytes(packet))
 
-    async def send_command(self, command, check_result=False):
+    async def send_command(self, command, check_result=False, response_timeout: Optional[int] = None):
         # Wait until we can send (only one pending command at a time)
         async with self.command_semaphore:
             assert self.pending_command is None
@@ -526,7 +526,8 @@ class Host(AbortableEventEmitter):
 
             try:
                 self.send_hci_packet(command)
-                response = await self.pending_response
+                await asyncio.wait_for(self.pending_response, timeout=response_timeout)
+                response = self.pending_response.result()
 
                 # Check the return parameters if required
                 if check_result:
@@ -625,14 +626,21 @@ class Host(AbortableEventEmitter):
 
     # Packet Sink protocol (packets coming from the controller via HCI)
     def on_packet(self, packet: bytes) -> None:
-        hci_packet = hci.HCI_Packet.from_bytes(packet)
+        try:
+            hci_packet = hci.HCI_Packet.from_bytes(packet)
+        except Exception as error:
+            logger.warning(f'!!! error parsing packet from bytes: {error}')
+            return
+
         if self.ready or (
             isinstance(hci_packet, hci.HCI_Command_Complete_Event)
             and hci_packet.command_opcode == hci.HCI_RESET_COMMAND
         ):
             self.on_hci_packet(hci_packet)
         else:
-            logger.debug('reset not done, ignoring packet from controller')
+            logger.debug(
+                f'reset not done, ignoring packet from controller: {hci_packet}'
+            )
 
     def on_transport_lost(self):
         # Called by the source when the transport has been lost.