다음을 통해 공유


Review common Device Fundamentals Reliability test failures

This topic describes common test failures that you can encounter when you run Windows Hardware Certification Kit (Windows HCK) Device Fundamentals Reliability tests.

In this topic:

  • Device Status Check task fails during setup

  • Device Path Exerciser fails with “Test thread exceeded timeout limit. Terminating thread error” error

  • Surprise Remove test fails with “Failed to receive IRP_MN_REMOVE_DEVICE after receiving IRP_MN_SURPRISE_REMOVAL” error message

  • SimpleIO plugins log failures

  • Testing I/O on a particular device permanently hangs, and eventually causes the test to fail because of a timeout

  • Tests do not resume from sleep

  • WirelessPlugin: ConnectToTestProfile() - Failed to connect to test profile. Reason string: "The specific network is not available." error message

  • WDTFSensorsPlugin: Open() - GPS Sensor did not go to ready state

Device Status Check task fails during setup

Device Status Check tasks often fail because the device is not properly set up with media or a connection before testing starts.

The Device Status Check task is included in the setup phase of every Device Fundamentals Reliability test job. It runs a script to verify that the device under test (DUT) is in working condition. If it fails, a log is created that indicates the problem with the device.

For example, for Bluetooth devices, you might get the following error:

PerformIO(Example ) Failed : Streaming error capturing audio HRESULT=0x8445001F Count 1

This error message can indicate that, prior to testing, you must connect to the Bluetooth device by using the Audio Control panel.

In the following example, the test device reports problem code A - CM_PROB_FAILED_START status. It should report problem code 0 (no problem).

WDTF_TARGETS          : INFO  : - Query("IsPhantom=False AND (DeviceID='USB\VID_0BDB&PID_1917&CDC_0D&MI_06\6&2A131B9E&1&0006')")
WDTF_TARGETS          : INFO  : Target: F5321 gw Mobile Broadband Network Adapter USB\VID_0BDB&PID_1917&CDC_0D&MI_06\6&2A131B9E&1&0006 
WDTF_TEST             : ERROR : Found a device that has a non-zero problem code or is phantom. Logging device info.
WDTF_TEST             : INFO  : DeviceID:     USB\VID_0BDB&PID_1917&CDC_0D&MI_06\6&2A131B9E&1&0006
WDTF_TEST             : INFO  : DisplayName:  F5321 gw Mobile Broadband Network Adapter
WDTF_TEST             : INFO  : Status:       Status Flags=0x1802400 (DN_HAS_PROBLEM DN_DISABLEABLE DN_NT_ENUMERATOR DN_NT_DRIVER) Problem Code=a (CM_PROB_FAILED_START)
WDTF_TEST             : INFO  : IsPhantom:    False

Device Path Exerciser fails with “Test thread exceeded timeout limit. Terminating thread error” error

When the test logs the Test thread exceeded timeout limit. Terminating thread error error during a device path exerciser test, the test also logs the last operation that it performed. Driver developers must determine why the last logged operation would cause the test to hang. For example:

WDTF_FUZZTEST : Test thread exceeded timeout limit. Terminating thread
WDTF_FUZZTEST : Last logged operation: ZwDeviceIoControlFile, CtrlCode=0x2b0020, InBuf=0xfffffc00, 0 OutBuf=0xfffffc00, 0

Surprise Remove test fails with “Failed to receive IRP_MN_REMOVE_DEVICE after receiving IRP_MN_SURPRISE_REMOVAL” error message

The DF - PNP Surprise Remove Device Test (Certification) might fail with the following error message if the PnP manager does not send the remove IRP to the test device stack after it sends the surprise remove IRP:

"Failed to receive IRP_MN_REMOVE_DEVICE after receiving IRP_MN_SURPRISE_REMOVAL. Ensure that there are no open handles or references to the test device (in user mode or in kernel mode) preventing IRP_MN_REMOVE_DEVICE from being sent. You may need to terminate any processes or services that may have open user mode handles to this device."

The PnP manager does not send the IRP_MN_REMOVE_DEVICE request until all outstanding file handles to the device are closed. That is, the PnP manager does not send the IRP_MN_REMOVE_DEVICE request until reference count of the PDO reaches zero. See Handling an IRP_MN_SURPRISE_REMOVAL Request for information on how to properly handle IRP_MN_SURPRISE_REMOVAL request.

To help debug this test failure, you should determine how the reference count of the physical device object (PDO) changes. Identify the process that is changing the reference count and examine what the call stack looks like when the reference count is changed. The following steps can be used for debugging this failure:

  1. If you have not already done so, connect a kernel debugger to the test computer. See Configuring a Computer for Driver Deployment, Testing, and Debugging.

  2. Set a ba (Break on Access) breakpoint at the location where the reference count of the PDO of the test device is stored. See Processor Breakpoints (ba Breakpoints) for more information about access breakpoints. In the following example, the kernel debugger !devnode command obtains information about the devnode for the USBvideo driver. The address of the PDO for this devnode is 0x849e9648.

    0: kd> !devnode 0 1 usbvideo
    Dumping IopRootDeviceNode (= 0x848fadd8)
    DevNode 0x849e9448 for PDO 0x849e9648
      InstancePath is "USB\VID_045E&PID_076D&MI_00\7&1243e0b7&0&0000"
      ServiceName is "usbvideo"
      State = DeviceNodeStarted (0x308)
      Previous State = DeviceNodeEnumerateCompletion (0x30d)
    
  3. Use the !devobj command on the PDO to display information about the reference count (RefCount) of the PDO.

    0: kd> !devobj 0x849e9648
    Device object (849e9648) is for:
     0000004e \Driver\usbccgp DriverObject 8727e120
    Current Irp 00000000 RefCount 0 Type 00000022 Flags 00003040
    Dacl 82910320 DevExt 849e9700 DevObjExt 849e99e0 DevNode 849e9448 
    ExtensionFlags (0x00000800)  DOE_DEFAULT_SD_PRESENT
    Characteristics (0x00000180)  FILE_AUTOGENERATED_DEVICE_NAME, FILE_DEVICE_SECURE_OPEN
    AttachedDevice (Upper) 88310588 \Driver\usbvideo
    Device queue is not busy
    
  4. Examine the PDO device object by using the dt (Display Type) kernel debugger command. The ReferenceCount shows the number of open handles for the device that are associated with the device object.

    0: kd> dt nt!_DEVICE_OBJECT 849e9648  
    …
       +0x002 Size             : 0x398
       +0x004 ReferenceCount   : 0n0
       +0x008 DriverObject     : 0x8727e120 _DRIVER_OBJECT
    ..
    …
    
  5. If the reference count is greater than 0 before starting the test:

    • Set a breakpoint where the PDO gets created.

    • After the PDO is created, set the break on access (ba) breakpoint at the location where the reference count of the PDO is stored.

      For example, the following command sets a ba (Break on Access) breakpoint on the device object (0x849e9648). The breakpoint is set on write access to the ReferenceCount (+4 offset) with a size of 4 bytes (the size of ReferenceCount).

      0: kd> ba w 4 849e9648+4 
      
    • If the reference count of the PDO is equal to 0 before starting the test, it is likely that running the test is what is causing the reference count of the PDO to be greater than zero at the time the test performs the surprise remove of the device. This usually indicates a handle leak. Run the PNP Surprise Remove Device test from a Command Prompt window or from Visual Studio to reproduce the failure and capture the information that is needed to troubleshoot the problem.

    Note  

    If you set the DoConcurrentIO parameter to TRUE, the test opens hundreds of file handles to the PDO. We recommend that you set this parameter to FALSE when you reproduce this failure.

     

  6. When the break on access (ba) breakpoint occurs, you can use the !thread and k (Display Stack Backtrace) kernel debugger commands to debug the failure. Because the reference count can change multiple times during the course of running the test, as an option, you can use the commandString parameter of the ba (Break on Access) debugger command to get the information that you need on each change to the reference count, and still continue to test.

    For example, in the following break on access command, the commandString consists of a !thread command that will identify the process causing the reference count change, and the .reload ; k 100 commands that will identify the call stack, a !devobj command to print the reference count on each change, and g command to continue after the breakpoint.

    0: kd> ba w 4 849e9648+4 "!thread; .thread /p /r; .reload; k 100; !devobj 849e9648; g"
    

Example:

In the following example, the CreateFile function call from a thread that is running in cscript.exe causes an increment to the reference count. Capturing all the instances where the reference count is changed while running the test and analyzing these call stacks can help identify the handle leaks.

THREAD 87eb3d40  Cid 1094.1490  Teb: 7f5a8000 Win32Thread: 82da2210 RUNNING on processor 3
Not impersonating
DeviceMap                 a71b3228
Owning Process            88199cc0       Image:         cscript.exe
Attached Process          N/A            Image:         N/A
Wait Start TickCount      1232688        Ticks: 0
Context Switch Count      18             IdealProcessor: 2             
UserTime                  00:00:00.000
KernelTime                00:00:00.000
Win32 Start Address ntdll!TppWorkerThread (0x7710704d)
Stack Init a6ebfde0 Current a6ebfa6c Base a6ec0000 Limit a6ebd000 Call 0
Priority 9 BasePriority 8 UnusualBoost 0 ForegroundBoost 0 IoPriority 2 PagePriority 5
ChildEBP RetAddr  Args to Child              
a6ebfa50 814a73fe f81771f8 814a72e5 8281000e nt!IopCheckDeviceAndDriver+0x61 (FPO: [Non-Fpo]) (CONV: stdcall) [d:\w8rtm\minkernel\ntos\io\iomgr\parse.c @ 182]
a6ebfb70 8149fb76 849e9648 848f9200 87164008 nt!IopParseDevice+0x11d (FPO: [Non-Fpo]) (CONV: stdcall) [d:\w8rtm\minkernel\ntos\io\iomgr\parse.c @ 1634]
…
…
0236f874 7710689d ffffffff 77195ae2 00000000 ntdll!__RtlUserThreadStart+0x4a (FPO: [SEH]) (CONV: stdcall) [d:\w8rtm\minkernel\ntdll\rtlstrt.c @ 1021]
0236f884 00000000 7710704d 0031c540 00000000 ntdll!_RtlUserThreadStart+0x1c (FPO: [Non-Fpo]) (CONV: stdcall) [d:\w8rtm\minkernel\ntdll\rtlstrt.c @ 939]

Implicit thread is now 87eb3d40
Connected to Windows 8 9200 x86 compatible target at (Wed Sep 19 21:04:27.601 2012 (UTC - 7:00)), ptr64 FALSE
Loading Kernel Symbols
...............................................................
................................................................
...............
Loading User Symbols
................................................................
...........................
Loading unloaded module list
.....................
ChildEBP RetAddr  
a6ebfa50 814a73fe nt!IopCheckDeviceAndDriver+0x61 [d:\w8rtm\minkernel\ntos\io\iomgr\parse.c @ 182]
a6ebfb70 8149fb76 nt!IopParseDevice+0x11d [d:\w8rtm\minkernel\ntos\io\iomgr\parse.c @ 1634]
…
…
0236f2d4 6970274e KERNELBASE!CreateFileW+0x61 [d:\w8rtm\minkernel\kernelbase\fileopcr.c @ 1194]
0236f31c 6b6ce0e1 deviceaccess!CDeviceBroker::OpenDeviceFromInterfacePath+0x178 [d:\w8rtm\base\devices\broker\dll\broker.cpp @ 177]
0236f34c 6b6cc5c0 MFCORE!CDevProxy::CreateKsFilter+0x46 [d:\w8rtm\avcore\mf\core\transforms\devproxy\devproxy.cpp @ 2263]
…
…
0236f874 7710689d ntdll!__RtlUserThreadStart+0x4a [d:\w8rtm\minkernel\ntdll\rtlstrt.c @ 1021]
0236f884 00000000 ntdll!_RtlUserThreadStart+0x1c [d:\w8rtm\minkernel\ntdll\rtlstrt.c @ 939]

Device object (849e9648) is for:
 0000004e \Driver\usbccgp DriverObject 8727e120
Current Irp 00000000 RefCount 1 Type 00000022 Flags 00003040
Dacl 82910320 DevExt 849e9700 DevObjExt 849e99e0 DevNode 849e9448 
ExtensionFlags (0x00000800)  DOE_DEFAULT_SD_PRESENT
Characteristics (0x00000180)  FILE_AUTOGENERATED_DEVICE_NAME, FILE_DEVICE_SECURE_OPEN
AttachedDevice (Upper) 88310588 \Driver\usbvideo
Device queue is not busy.

SimpleIO plugins log failures

Device Fundamentals Reliability tests use Provided WDTF Simple I/O plug-ins to test I/O on devices. SimpleIO plugins are WDTF extensions that test generic device-specific I/O functionality. If a WDTF plug-in exists for the type of device that is being tested, then the test uses IWDTFSimpleIOStressAction2 interface to test I/O on the device.

Errors that are logged by WDTF SimpleIO plugins use the WDTF_SIMPLE_IO tag in the TestTextLog.log file (see WDTF Object Name tags. The error message always identifies the device under test and the specific reason for failure.

Example:

In this example, the Wireless SimpleIO plug-in logged a failure that I/O failures occurred during the test of an 802.11n USB Wireless LAN Card device. Specifically, the SimpleIO plugin pinged the gateway address by using an IcmpSendEcho function, which returned an error 11010. This error translates to Error due to lack of resources.

WDTF_SIMPLE_IO            : ERROR :  - PerformIO(802.11n USB Wireless LAN Card USB\VID_XXXX&PID_XXXX\X&XXXXXXX&X&X ) Failed : WirelessPlugin: TestPingGateway() - IcmpSendEcho() call failed several times. The error reported is for the last failure instance Win32=11010 - Error due to lack of resources.

Testing I/O on a particular device permanently hangs, and eventually causes the test to fail because of a timeout

Device Fundamentals Reliability tests are scenario based tests and combine I/O testing with PNP & power test scenarios. The tests typically will test I/O for two minutes before and after a scenario. For instance, the DF - Sleep with IO Before and After (Basic) test does the following:

For each sleep state supported on the system (CS, S1, S2, S3, S4)

Test I/O on devices with I/O plugins in parallel (one thread per device) for 2 minutes

Enter sleep state & exit after 2 minutes

Next

The test ends up testing I/O on devices several times (one time for each sleep state) when it runs. See Review Log Files for information about how to see this in the log files.

One of the common failures when testing I/O is that testing I/O on a particular device can permanently hang. This results in the test to eventually fail after a test timeout period (which is typically hours). See Troubleshooting Windows HCK Test Failures for information about how to identify failures caused by timeout. 

Note  

Windows HCK will terminate the hung process after the timeout period. Instead of waiting for the test to eventually fail because of a permanent hang, we recommend that you investigate the hang while the hung process is still running on the system. See the Test Hangs section of the Troubleshooting Device Fundamentals Reliability Testing by using the Windows HCK topic for information about how to troubleshoot test hangs.

 

Depending on how many devices the I/O is being tested on, the hung device can be identified as follows:

  1. If the number of devices the test is testing I/O on is one, you will see no progress in the command window for more than ten minutes. The last log entry in the command window will have a WDTF_SIMPLE_IO or WDTF_SIMPLEIO_STRESS tag, and it will identify the hung device. See Review Log Files for more information about how to read the test log files.

  2. If the number of devices on which the test is testing I/O is greater than one, you will see a constant repetition of PerformIO(<Device Name>) Count … messages for longer than ten minutes in the command window. The test tries to stop testing I/O on one device at a time after two minutes of testing I/O on them. If the stop operation is successful for a particular device, you should see a “Stop” message followed by a “Close” message for the device in the logs. If the “Stop” message is seen, but the corresponding “Close” message is not seen for a device, then it implies that testing I/O to this device is hung.

Example:

In the following case, the mobile broadband device is the problem device because there is a “Stop” message, but there is no corresponding “Close” message. On the other hand, the I2C HID Device has both a “Stop” message and a “Close” message, which implies that the test was able to stop I/O on the device without any problems. The test never had a chance to stop testing I/O on the Microsoft Basic Render Driver and Microsoft ACPI-Compliant System devices; therefore, “PerformIO” messages are continuously seen for these devices.

WDTF_SIMPLEIO_STRESS      : INFO  :  - Stop(I2C HID Device ACPI\STMQ7017\2&DABA3FF&3 )
WDTF_SIMPLE_IO            : INFO  :  - Close(I2C HID Device ACPI\STMQ7017\2&DABA3FF&3 )
WDTF_SIMPLEIO_STRESS      : INFO  :  - Stop(XYZ Mobile Broadband Device USB\VEN_XXX&PID_XXX\X&XXXXXX&X&X)
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft Basic Render Driver ROOT\BASICRENDER\0000 ) Count 119
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft ACPI-Compliant System ACPI_HAL\PNP0C08\0 ) Count 119
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft Basic Render Driver ROOT\BASICRENDER\0000 ) Count 119
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft ACPI-Compliant System ACPI_HAL\PNP0C08\0 ) Count 119
…
…
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft Basic Render Driver ROOT\BASICRENDER\0000 ) Count 119
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft ACPI-Compliant System ACPI_HAL\PNP0C08\0 ) Count 119
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft Basic Render Driver ROOT\BASICRENDER\0000 ) Count 119
WDTF_SIMPLE_IO            : INFO  :  - PerformIO(Microsoft ACPI-Compliant System ACPI_HAL\PNP0C08\0 ) Count 119
…
…

The next step is to inspect the stack traces of the threads in the test process to determine why testing I/O to the mobile broadband device is hung. You will find that one of the threads in the test process is used to specifically test I/O on the mobile broadband device. See Use kernel debugging to debug Device Fundamentals Reliability test failures for additional troubleshooting information.

Tests do not resume from sleep

Device Fundamentals Reliability tests rely on system wake timers to wake up the test system from sleep states during power management testing. Faulty wake timers can prevent the test system from automatically waking up during the test runs. If you the test system is not automatically waking up from sleep, you might need to contact your BIOS vendor to have them release a BIOS fix to address the wake timer issues, or run tests on a different system where the wake timers work as expected.

The system can also permanently hang during power up or power down because of driver bugs. In this case, you should re-run the test by using the test system connected to a kernel debugger, and debug the system hang from the kernel debugger.

See Setting Up Kernel-Mode Debugging Manually for information about how to setup a kernel debugger. See Troubleshooting Windows HCK Test Failures for general guidance on how to troubleshoot system hangs during Windows HCK test runs.

WirelessPlugin: ConnectToTestProfile() - Failed to connect to test profile. Reason string: "The specific network is not available." error message

Device Fundamentals tests will fail with this error message if correct values for Wpa2PskAesSsid and Wpa2PskPassword parameters are not supplied to the test at test schedule time. The tests require you to provide connection information (SSID and password) of a test wireless network if one of the devices under test is a WiFi adapter. See parameters section of the failing test's documentation page for more information about these test parameters.

WDTFSensorsPlugin: Open() - GPS Sensor did not go to ready state

Device Fundamentals Reliability tests require systems with a GPS sensor to be tested in an environment where there is a strong GPS signal (in order for the tests to be able to test I/O on the GPS sensor device). This error indicates that the GPS sensor on the test system cannot get a GPS fix. Please consider running the tests in a location where the test system can get a strong GPS signal.

Troubleshooting Device Fundamentals Reliability Testing by using the Windows HCK

 

 

Send comments about this topic to Microsoft