Archive for the ‘Windows Debugging’ Category

Windows 10 on ARM is looking good

May 26, 2018



  • WoW64


Transfer Learning sample from Microsoft AI framework – CNTK

March 27, 2018

nsfer learning is really a great way to save on resources by transfer of learning from an existing model. WMore info here:

Transfer Learning is a useful technique when, for instance, you know you need to classify incoming images into different categories, but you do not have enough data to train a Deep Neural Network (DNN) from scratch. Training DNNs takes a lot of data, all of it labeled, and often you will not have that kind of data on hand. If your problem is similar to one for which a network has already been trained, though, you can use Transfer Learning to modify that network to your problem with a fraction of the labeled images (we are talking tens instead of thousands).

What is Transfer Learning?

With Transfer Learning, we use an existing trained model and adapt it to our own problem. We are essentially building upon the features and concepts that were learned during the training of the base model. With a Convolutional DNN (ResNet_18 in this case), we are using the features learned from ImageNet data and cutting off the final classification layer, replacing it with a new dense layer that will predict the class labels of our new domain.

The input to the old and the new prediction layer is the same, we simply reuse the trained features. Then we train this modified network, either only the new weights of the new prediction layer or all weights of the entire network.

This can be used, for instance, when we have a small set of images that are in a similar domain to an existing trained model. Training a Deep Neural Network from scratch requires tens of thousands of images, but training one that has already learned features in the domain you are adapting it to requires far fewer.

In our case, this means adapting a network trained on ImageNet images (dogs, cats, birds, etc.) to flowers, or sheep/wolves. However, Transfer Learning has also been successfully used to adapt existing neural models for translation, speech synthesis, and many other domains – it is a convenient way to bootstrap your learning process.

Here is an example:



CNTK contains a transfer learning sample. I have a machine with Nvidia GPU GeForce GTX 960M

Here is what Happens when I start this sample with python:

Activate CNTK

C:\CNTK-Samples-2-4\Examples\Image\TransferLearning>conda activate cntk-py36

(cntk-py36) C:\CNTK-Samples-2-4\Examples\Image\TransferLearning>


(cntk-py36) C:\CNTK-Samples-2-4\Examples\Image\TransferLearning>python
Traceback (most recent call last):
File “”, line 9, in <module>
import cntk as C
ModuleNotFoundError: No module named ‘cntk’


This gets resolved by picking python from the anaconda.


cntk-py36) C:\Users\Dell\Downloads\ethereum-mining-windows>where python

(cntk-py36) C:\CNTK-Samples-2-4\Examples\Image\TransferLearning>C:\Users\Dell\Anaconda3\python.exe


Selected GPU[0] GeForce GTX 960M as the process wide default device.
Build info:

Built time: Jan 31 2018 14:57:35
 Last modified date: Wed Jan 31 01:10:27 2018
 Build type: Release
 Build target: GPU
 With 1bit-SGD: no
 With ASGD: yes
 Math lib: mkl
 CUDA version: 9.0.0
 CUDNN version: 7.0.5
 Build Branch: HEAD
 Build SHA1: a70455c7abe76596853f8e6a77a4d6de1e3ba76e
 MPI distribution: Microsoft MPI
 MPI version: 7.0.12437.6
Training transfer learning model for 20 epochs (epoch_size = 6149).
Training 15949478 parameters in 68 parameter tensors.
CUDA failure 2: out of memory ; GPU=0 ; hostname=DESKTOP-IA3HLGI ; expr=cudaMalloc((void**) &deviceBufferPtr, sizeof(AllocatedElemType) * AsMultipleOf(numElements, 2))
Traceback (most recent call last):
 File "", line 217, in <module>
 max_epochs, freeze=freeze_weights)
 File "", line 130, in train_model
 trainer.train_minibatch(data) # update model with it
 File "C:\Users\Dell\Anaconda3\lib\site-packages\cntk\train\", line 181, in train_minibatch
 arguments, device)
 File "C:\Users\Dell\Anaconda3\lib\site-packages\cntk\", line 2975, in train_minibatch_overload_for_minibatchdata
 return _cntk_py.Trainer_train_minibatch_overload_for_minibatchdata(self, *args)
RuntimeError: CUDA failure 2: out of memory ; GPU=0 ; hostname=DESKTOP-IA3HLGI ; expr=cudaMalloc((void**) &deviceBufferPtr, sizeof(AllocatedElemType) * AsMultipleOf(numElements, 2))

 > Microsoft::MSR::CNTK::CudaTimer:: Stop
 - Microsoft::MSR::CNTK::CudaTimer:: Stop (x2)
 - Microsoft::MSR::CNTK::GPUMatrix<float>:: Resize
 - Microsoft::MSR::CNTK::Matrix<float>:: Resize
 - std::enable_shared_from_this<Microsoft::MSR::CNTK::MatrixBase>::enable_shared_from_this<Microsoft::MSR::CNTK::MatrixBase>
 - std::enable_shared_from_this<Microsoft::MSR::CNTK::MatrixBase>:: shared_from_this (x3)
 - CNTK::Internal:: UseSparseGradientAggregationInDataParallelSGD
 - CNTK:: CreateTrainer
 - CNTK::Trainer:: TotalNumberOfUnitsSeen
 - CNTK::Trainer:: TrainMinibatch (x2)
 - PyInit__cntk_py (x2)

I run into a problem with memory. My graphics card has a dedicated memory of 2Gb  as you can see below:

Apparently that isn’t enough. Next question I have is does CUDA use dedicated GPU memory or shared system memory? As my graphics card has 3.9 GB of that:

So my total Graphics memory is 5.9 GB


It turns out that I am running out of Dedicated GPU memory not shared system memory.

I couldn’t find nany GPU setting but I tied reducing the mini batch size and that the trick. In the  source code on line 41 and changing the the variable as ‘mb_size = 50’ to ‘mb_size = 30’ did the trick. Your mileage may vary based on  your GPU so do experiment.

AI next conference Jan. 2018 Seattle WA

February 6, 2018

The  NextCon conference series  began  early last year. The conference was organized by Association of Technology and Innovation (ATI) ‘s and Bill Liu was the lead.  This was in succession to the last one in Seattle in March 2017. The conference had  loads of companies which sponsored and had tech. talks. Roughly around 400 people attended the conference.

Conference Schedule:





Quick summary of recently concluded AI conference Nextcon in Seattle, Jan 2018.   This summary has my key takeaways and my learnings, Keynote summaries,  some of the break-away sessions I attended summaries, my side discussions as well as links and resources to the tech talks as well as links to blogs and Videos. It also includes some of my follow-up blog reading to understand  concepts. Hope this is useful for you as  much I found it to be!!


The conference had 4 tracks:

  1. Computer Vision
  2. Speech /NLP
  3. Data Science/Analytics
  4. Machine Learning

With limited time I picked mainly the Data Science and Machine learning tracks to understand trends on how to handle large amount of data and how to make sense of large amounts of data.


  1. Key takeaways & learnings for us:

These are some of things I have distilled and filtered from the Conference as areas of interest.

  1. AI is a great tool to have in your tool box. It isn’t the end all of all tools at least for now. But this could change in the future. For eg: AI which can disambiguate and find flaws in the electronic welds cant tell you whether a Kid is holding a tooth brush or base ball bat.
  2. Reinforcement learning(RL) is making a come back and yielding great results thought at a slightly higher cost of latency etc., time and infrastructure costs. Martin  Gomer  from google showed that how he trained a Pong playing AI with just historic data and by making it play itself and generating lot of data and get better at it. Think of how a kid learns to bike or learns to walk …ai_nvidiaReinforcement  learning (RL) and Neural networks Neural networks are algorithms, RL is a problem type. You can approach RL with neural networks.What makes RL very different from the others is that you typically don’t have a lot of data to start with, but you can generate a lot of data by playing. You have to deal with the problem that you have to make decisions, but it is not clear what is good (delayed reward). For example, it might take several moves until you know in Go if a move was smart.3. OS for AI – The next frontier will be when people will use each others algorithms and models to come up with a sophisticated service which aggregates. For eg: John peck from Algortigmia shows how. Someone writes a fruit classifier and another persona vegetable classifier and then a third party could aggregate them into a fruit or vegetable classifier.composability

elastic scale

Algorithmia maybe a good resource for paying for ML algorithms. I suggested to them after the talk to support  offering data as well at a $$.  This is inline with Data Science as a service idea

4. Auto ML or off-the-shelf machine learning methods – Machine learning is evolving at a pretty strong pace. More and more it is possible to just feed the AI platform dataset and it tunes the hyperparameters and comes up with a trained model

5. In the large scale of things AI is currently pretty early in its evolution


aiishere6. Future of AI from Prof. Oren Otzioni  – When will Superintelligence Arrive? AI experts try and answer  the question. It still is far out!!

7. Another interesting talk was by Twitter on Online ML and why they didn’t use deep learning. Deep learning currently has some disadvantages especially in real time low latency scenarios. More details on this below

8. Deep learning is providing lot of value however it comes at a cost as it requires a large data set. It however does require a solid hardware infrastructure.  Unfortunately, in deep learning, people usually see very sublinear speedups from many GPUs. Top performance thus requires top-of-the-line GPUs.

9. Microsoft AI platform is super rich in terms of tools, services, 3rd party tools integrations etc.

Microsoft demoed Azure ML workbench which seems like a really cool tool for the time consuming activity  of data wrangling.


2. Conference KeyNote summary:

  1. Steve Guggenheimer from Microsoft

Steve talked about the Microsoft AI platform and applications already on a lot of features.


Microsoft AI platform-


Microsoft demoed Azure ML workbench which seems like a really cool tool for the time consuming activity  of data wrangling.



The platform is super rich in terms of tools, services, 3rd party tools integrations etc.

Ethics in AI

Microsoft realizes the potential of AI and how it can be misused and hence Steve shared the Microsoft AI ethics. Satya has talked about compassionate AI  as  the AI for the future

Microsoft has published a nice book on this subject  called “The Future Computed”

ai ethics msft

I also liked the live demo on how the Bing  team uses specialized FPGA’s. FPGA’s or Field Programmable Gate Arrays are programmable hardware devices sort of a CPU for specific task rather than general purpose which allows optimizations to be built in.


CPU vs FPGA performance within Bing team – FPGAs and ASIC derivatives just dedicated to a certain task perform really really well at the same time taking a fraction of the Power.


2. AI at DIdi Chuxing – Didi Chuxing is like the Uber of China and the scale they have to deal with is humongous.   I liked DIdi Chuxing’s Presentation on how they are using AI in the transportation sector.  Lot of it can be applied to other fields as well as the problems are similar in nature. They  presented the iterations  on how they solved their problems using various AI algorithms and have narrowed it down to deep learning and Reinforcement learning to look at forecast, ETA, dispute resolution etc.  They started with  regression models to Deep learning models. Deep learning has helped them solve more problems.


They have applied AI to multiple problems areas within transportation –


More details here:

 3. UW Prof. Oren Etzioni also presented a good deck on Future of AI which is more like Is AI the evil power it is made out to be rather than typical technical trends of AI?


Winograd schemas is an alternative to the Turing Test developed by Hector Levesque.

The Turing Test is intended to serve as a test of whether a machine has achieved human-level intelligence. In one of its best-known versions , a person attempts to determine whether he or she is conversing (via text) with a human or a machine. However, it has been criticized as being inadequate. At its core, the Turing Test measures a human’s ability to judge deception: Can a machine fool a human into thinking that it too is human? It also suggests that the Turing Test may not be an ideal way to judge a machine’s intelligence.  An alternative is the Winograd Schema Challenge.

Rather than base the test on the sort of short free-form conversation suggested by the Turing Test, the Winograd Schema Challenge (WSC) poses a set of multiple-choice questions that have a particular form.  The test is dedicated to furthering and promoting research in the field of formal commonsense reasoning. For eg:

4. “Tensorflow and deep reinforcement learning without a PhD“ by Martin Gomer  from google.

He briefly alluded to Auto ML which learns the model architecture directly on the dataset of interest:


Google deep mind teaching a virtual human to walk/jump etc I was an athlete and when I look at the image below – The stride and the arms are just what a good long jumper would use and would be proud off below 🙂


Demonstration of playing  pong without any specialized algorithms with deep reinforcement leaning and lots of data:

 Link to the Video of his talk: –

Google deep mind taught itself to walk –


5. Key Note – “Deep learning at amazon Alexa” by Nikko Strom from Amazon

This is very powerful as it shows how Alexa is using multi modality, device and personal context which Alexa is using. This is really very powerful and can really engage the user!!

3. Summary of Breakout sessions (I attended)

  1. ML track – Twitter – Parameter Server approach for online ML at Twitter

The talk basically discussed the evolution of parameters servers in Twitter which need to scale and have real time approaches to online ML. Their approach has been around load balancing,  filtering, centralized Parameter servers. They have tried deep learning but they found as of now Deep Learning is not working for them:

Some of the disadvantages of Deep learning:

  1. Latency for their usage is high
  2. Model quality not impacted much ROI
  3. New approaches in ML could remove displace deep learning


  1. ML track – Machine learning at scale by Amy Unruh from Google The talk showed that there is a gap in the Google ML offering and is addressed by Auto ML for vision. Also, compares the various techniques in terms of resources needed to solve an AI problem typically:

1)      Time

2)      Prediction Code

3)      Serving Infrastructure

4)      Model Code

5) Training data


Resources needed to solve an AI problem  per Google


ML  as an API  – Mainly time and prediction code


Custom code and model – More resource intense

custom build

Custom model with transfer of learning from another project  – It takes less time and can reuse model code and training data

transfer learning

Google has identified Gap in the continuum  from DIY ML to ML APIs


They are trying to address it with Auto ML. This is currently limited to Vision API only.

cloud automl

Auto ML – Currently Google has it  only  for Vision API but allows for deep Neural networks  to be auto generated.

It allows savings on Model code etc, infrastructure etc


Only need to provide training data. It trains, deploys, creates a Neural network automatically.


Under the hood is creating new Neural network layers automatically.

Content below came from Martin Gomer  Google keynote speakers with the talk  titled “Tensorflow and deep reinforcement learning without a PhD


He briefly alluded to Auto ML which learns the model architecture directly on the dataset of interest


More details on Auto ML  here:

 3. Deep multimodal intelligence by Xiaodong He from Microsoft –

Microsoft Research Xiaodong He described  the scene with natural language :

  1. Understanding the image’s content
  2. Reasoning relationships among objects & concepts
  3. Generate a story in natural language

However, true understanding of the world is much more challenging


There were quite a few other parallel talks  but time was limited 😦

4. Presentation Tid- Bits

Slight digression on a nice presentation tool I found some speakers use at the conference which compliments the laser pointer. It retails for  around $130


And is great for highlighting code etc.


5.  Links /References



Papers — Lots of good papers on KDD:







How to Kernel Debug Connected Standby/Modern Standby systems?

October 24, 2016


Debugging a Modern Standby (Connected Standby earlier) scenario can be challenging as there are some smaller subtle things to keep in mind. Most modern standby/connected Standby systems are newer systems with USB 3.0 xHCI controllers so this blog post only focuses on systems which support USB 3.0 debugging.

What you need:

  1. USB 3 cable –
  2. USB Type C to type A adapter – Needed only if the device doesn’t have a USB Type A port
  3. Windbg Bits – Many sources including the Kits —  WDK or ADK

Methodology to setup Kernel Mode debugging

  1. Setup the machine for USB 3.0 debug  as mentioned here:
  2. Make sure you Disable Secure Boot in the BIOS menu
  3. Hook up the cable as follows  setup
  4. Check the USB device hierarchy and turn  off all the components. You can do this from device manager, usb tools usb_hierarcy
  5. Disable Turning off  USB stack components – Hubs and controllers on target Disable Powersaving on USb controller For the uSB hub, Uncheck the box to allow the computer to turn off the debice to save power  usb_hub_power
  6. Disable Powersaving on USB HUB/s — For the uSB xHCXI controller, Uncheck the box to allow the computer to turn off the device to save power usb_hub_power
  7. If there are multiple controllers or Hubs make sure you pick the right one where you plan to debug . Also if there is another level of hub in between do the same for that as well.
  8. Debug away!!


Decrypting wdf01000.sys interrupts with WPA

June 13, 2014

If you are trying to figure out which WDF driver is the source of all interrupts there is a way out. Since wdf01000.sys fields all interrupts and then calls the actual driver, it is difficult to figure out which driver caused the interrupts. Fortunately, there is a way out: You need to use the trace flags in the kernel – WDF_INTERRUPT and WDF_DPC. You can find all Kernel trace flags by the following command: “xperf -providers KF” .

You can trace as follows: ”

xperf -on diageasy+WDF_DPC+WDF_INTERRUPT+0x48000000+PROC_THREAD+LOADER+INTERRUPT+DPC+CSWITCH+TIMER+CLOCKINT -stackwalk TimerSetPeriodic+TimerSetOneShot+CSwitch+readythread+profile -clocktype perfcounter -buffersize 1024 -minbuffers 1024

xperf -d test.etl

Ring 0 debugging and Windbg – part 2

September 28, 2009

As promised earlier I am including some more savory Kernel debugging topics. Some of these scenarios are more corner case one-off ones.

  1. Virtual Machine (VM) debugging

Debugging a virtual machine 

You will need to enable debugging on the VM just like a regular OS. Also, you will need to set one of the COM Port(COM1 or Com2) as a named pipe (=\.\pipe\<com1| com2 |yourstring>). On the VM HOST OS you can type this to attach to the debuggee.

windbg [-y SymbolPath] -k com:pipe,port=\.\pipe\PipeName[,resets=0][,reconnect] 
 Virtual to virtual Machine debugging

This is the scenario where you use one VM running on the host to debug another VM running on the same host. This is interesting scenario and is useful in any scenario where you may have to reimage the host OS maybe in a lab. Environment.  I had to set this up for showing kernel debugging for KMDF at WinHEC.

This is how I did it:

  • Created a named pipe on COM1  as following on the VM acting as host/debugger


  • Created a named pipe on COM2 on the debuggee


When I tried this with COM1 port on the debugee VM I couldn’t get it to work but Named pipe on Com2 port worked for me. If you had a different experience please share it for others benefit.

Invoke the debugger on host VM as:

“Windbg –k com:port=com2”.


  1. Debugging Local/single mode debugging

In some cases you may not have access to another machine or you may want to look at some device state or read a global variable for a driver. In the absence of other applications  or tools one can use the local debugger. You will need to enable debugging the local machine on Vista and forward:

bcdedit /debug on

followed by a reboot. You can look at the documentation for how to edit the boot.ini file to do this on pre-vista OS.

Then you can begin debugging with the following command:

C:\> kd –kl

However, it has limited use. You can’t set break points or check call stacks.

You can use it for checking state of global variables in your drivers etc. You could also use it for trying out commands/ approaches when you don’t have another machine handy if you are going to visit a customer for example.

  1. I am stuck now what. How do I get my machine back?

At some point in your life just by law of averages J you will come across a system which is crashing regularly maybe even at boot that you can’t get any work done or even log on to the system. These are a few techniques you could use to get your system back.

Safe mode

Try booting safe mode. The set of drivers loaded in safe mode are a subset of the remaining drivers. Majority of times these are well tested critical boot  drivers. If you enabled verifier for all drivers. This is especially  useful if one of the non-safe mode drivers is failing hence doesn’t get loaded in safe mode.

In some cases (more of a driver developer /test/verification scenario) the non-safe mode driver maybe failing a verifier check if you enable verifier for all drivers on your machine maybe to debug a corruption problem.


System restore

System restore helps you restore your system to a previous state. A thing to note is that driver binaries are not rolled back. It can be useful as it primarily reverts the registry to a prior well known state. If the state the system is in, involved the registry directly or indirectly, this is a good option to try.

The scenario I discussed earlier  of enabling driver verifier to safe mode  drivers and one such driver barfing could get your system to not boot even in the safe mode. It is unfortunate but you could find even safe mode drivers sometimes fail verifier checks.

Driver Verifier saves state in the registry so by reverting the registry (with system restore) you can disable the verifier and claim your machine back perhaps temporarily since the offending driver is still on your system. If you are lucky there is an update waiting for you from the IHV/ISV which fixes the issue flagged by driver verifier.

BIOS and disabling devices

You also find most BIOSes have the option of disabling devices. If the offending driver is for hardware and you can figure out the name of the driver binary (from the BSOD)which is crashing you can try disabling it from the BIOS. For e.g.: This is useful for devices you couldn’t care more for like the finger print reader (unless you actually use it) which has a buggy driver. Once the driver is fixed  you can re-enable the device.

Kdfiles and Windbg

Kdfiles is an excellent way to change a boot driver or another driver especially if you are debugging/developing a driver. Windbg acts as a conduit for passing bits of the driver you need to change on to the target machine. It can be used for boot drivers as well except for windows vista.

For e.g.:

Kd> .kdfiles –m \systemroot\system32\driver.sys \myshare\checked\driver.sys

Crash dump

If all else fails and you desperately need a crash dump to pass on to the IHV for debugging. On windows 7, kernel crash dump which is good enough normally for debugging  is enabled by default.

F10 trick to edit the boot parameters passed to the kernel at boot time

If you forgot to enable kernel debugging with bcdedit you could always do it at boot time. Pressing F8 can give you the boot menu which has the option of enable debugging. There is another way of doing it however you need to have precision timing. You will need to add the following lines to the boot debug options after pressing the F10 key. The debug option doesn’t persist across a boot.

  • Serial  — “/debug /debugport=comX /baudrate=115200”
  • 1394  –  “/debug /debugport=1394 /channel=[1-63]”
  • USB –  “/debug /debugport=usb /targetname=String “


Getting a crash dump for a process without enabling the kernel debugger

If the machine is not booted in debug mode (local kd cannot be used) and you can’t enable debugging and reboot it in fears of losing the repro, then you could try the following:

kdbgctrl –td <pid> <file>

This will capture a driver dump with the hang that should contain the IRP information to see if any  thread in the  process is hung on an IRP. Kdbgctrl.exe is a tool from the debuggers packages.


Debugging over USB-serial converter

When you set a target to use a COM port for debugging, the debugging engine
takes over the port and drives it itself. This is why a COM port will no
longer appear in device manager when being used for debugging.  Let’s say now you want to  
use a port on a USB device. When the system first boots, that port do
not exist according to the system hardware. Therefore there is no such thing
as COMX, so it just doesn’t work. Ports on USB-Serial converters do not get
created until later on when the PNP manager loads the driver for the USB

The kernel drives the debug ports so if a driver is driving the debug port that won’t work.

So debugging over USB-serial converters doesn’t work.

Accessing the registry from Windbg

Won’t it be great if you could access the registry values from windbg. This way you could disable driver verifier through the debugger or set/unset some driver registry value. There are numerous scenarios which this could be handy. You can actually do this over windbg. Let me warn you that this may require a little poking and digging around as I show below.

The verifier stores its settings in the registry at:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management\VerifyDrivers

For more information refer
If we could access this from the debugger we could actually turn off verifier using windbg.

Before we dig into the registry I would recommend you to read the following article from Mark Russinovich which explains the internals of the registry:

 What is the system registry?

The registry holds system information, such as configuration data, for both hardware and software.  The registry is broken up into “Hives” which are registry files for parts of the registry.  For example, the Software hive is where the HKEY_LOCAL_MACHINE\Software information is kept.  The Hives are broken up into “Bins” and the Bins are broken up into “Cells” which hold the registry key and value data.    

The diagram below which is borrowed from the above article shows the layout of the registry.



How is this information managed?

The registry subsystem maintains the registry on a Hive basis.  That is how the registry keeps track of open registry files.  The open Hives in the system can be displayed by using the debugger extension “!reg hivelist”.

Below is an example output:

lkd> !reg hivelist


| HiveAddr |Stable Length|Stable Map|Volatile Length|Volatile Map|MappedViews|PinnedViews|U(Cnt)| BaseBlock | FileName


| 8bc0c2e8 |       1000  | 8bc0c364 |       1000    |  8bc0c4a0  |        0  |        0  |     0| 8bc0e000  | <UNKNOWN>

| 8bc1a5a0 |     dbb000  | 8bc22000 |      44000    |  8bc1a758  |        0  |        0  |     0| 8bc1c000  | SYSTEM

| 8bc3f678 |      14000  | 8bc3f6f4 |       7000    |  8bc3f830  |        0  |        0  |     0| 8bc48000  | <UNKNOWN>

| 8bd6d9d0 |      8f000  | 8bd6da4c |       1000    |  8bd6db88  |        0  |        0  |     0| 901cb000  | temRoot\System32\Config\DEFAULT

| 8bd811f8 |    3c85000  | 901a3000 |       3000    |  8bd813b0  |        0  |        0  |     0| 8bcdd000  | emRoot\System32\Config\SOFTWARE

| 8bd85008 |       6000  | 8bd85084 |          0    |  00000000  |        0  |        0  |     0| 8bd05000  | <UNKNOWN>

| 8bd87008 |       8000  | 8bd87084 |       1000    |  8bd871c0  |        0  |        0  |     0| 8bd8e000  | <UNKNOWN>

| 911f53e8 |      3e000  | 911f5464 |       a000    |  911f55a0  |        0  |        0  |     0| 95661000  | files\NetworkService\NTUSER.DAT

| 956e2798 |      3d000  | 956e2814 |          0    |  00000000  |        0  |        0  |     0| 956e3000  | rofiles\LocalService\NTUSER.DAT

| 9ec258b8 |     406000  | 9ec3d000 |       d000    |  9ec25a70  |        0  |        0  |     0| 9ec27000  | \??\D:\Users\vmanan\ntuser.dat

| a0003650 |     3b6000  | a0073000 |          0    |  00000000  |        0  |        0  |     0| a007b000  | \Microsoft\Windows\UsrClass.dat

| a89d0008 |     861000  | a89d8000 |          0    |  00000000  |        0  |        0  |     0| a89d2000  | Volume Information\Syscache.hve

| d1d0c3b0 |       b000  | d1d0c42c |          0    |  00000000  |        0  |        0  |     0| b0bc3000  | Device\HarddiskVolume1\Boot\BCD


Before we can access the registry, we need to know where to begin.  So we need the root block:

In the above case I pick the root block for the System Hive. Below I try and dump a few values under the “HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet1\Control\Session Manager\Memory Management\” key namely “SystemPages” 

lkd> !reg baseblock 8bc1a5a0

FileName :  SYSTEM


Sequence1:  84a7

Sequence2:  84a7

TimeStamp:  1ca3f22 34d2fd1c

Major    :  1

Minor    :  5



RootCell :  20    -> Index of the root cell

Length   :  dbb000

Cluster  :  1

CheckSum :  4bd3452c

Since everything in the registry is represented as a cell we look at the cell index of the root block to calculate the cell address. The Cell address is broken down to the map directory offset,  map table offset , and then block offset to access registry data.  The debugger extension !reg cellindex does this for us automatically. !reg cellindex is then used to get the cell address. 


lkd> !reg cellindex  8bc1a5a0 20

Map = 8bc22000 Type = 0 Table = 0 Block = 0 Offset = 20

MapTable     = 8bc23000

BlockAddress = 937d3000

pcell:  937d3024


The !reg subkeylist command can do this for you.  This breaks down the Key Node and does all the work.


lkd> !reg subkeylist 8bc1a5a0 937d3024

Dumping SubkeyList of Key <CMI-CreateHive{3FE9C764-973F-45AC-B646-C7564C5B6CAD}> :

SubKeyCount[Stable  ]: 0x7

SubKeyLists[Stable  ]: 0xaf1f30

SubKeyCount[Volatile]: 0x1

SubKeyLists[Volatile]: 0x80000178

[   7] Stable SubKeys:

[Idx]   [SubKeyAddr]    [SubKeyName]

[0]     937d3164        ControlSet001

[1]     937d3164        ControlSet001

[2]     937d3164        ControlSet001

[3]     937d3164        ControlSet001

[4]     937d3164        ControlSet001

[5]     937d3164        ControlSet001

[6]     937d3164        ControlSet001

[   1] Volatile SubKeys:

[Idx]   [SubKeyAddr]    [SubKeyName]

[0]     8bc31024        CurrentControlSet

 Use ‘!reg knode <SubKeyAddr>’ to dump the key

What the List points to depends on the number of values.  The registry uses an index mechanism to access the sub key data.  The type of index is determined from the type code.  Above 0x686c indicates this is a CM_KEY_HASH_LEAF.  Where the cell contains a hash of the key and the cell index of the key.   Another example is a CM_KEY_FAST_LEAF where the first four characters of the key name are used instead of a hash.  What is used depends on the version of the hive.  Also when there are many sub keys, the index can be a CM_KEY_INDEX_ROOT, which contains cell indexes which point to leafs. 

You’ll see that some hives are volatile and don’t have associated files. The system creates and manages these hives entirely in memory; the hives are therefore temporary in nature. The system creates volatile hives every time the system boots. An example of a volatile hive is the HKEY_LOCAL_MACHINE \HARDWARE hive, which stores information regarding physical devices and the devices’ assigned resources. Resource assignment and hardware detection occur every time the system boots, so not storing this data on disk is logical.

You can use the !reg subkeylist recursively till you hit a roadblock where you don’t see all the child keys showing up in which case we resort to manual means. Since we have CurrentControlSet1, lets try and o down to its child key “Control”.

lkd> !reg subkeylist 8bc1a5a0 937d3164

Dumping SubkeyList of Key <ControlSet001> :

SubKeyCount[Stable  ]: 0x5

SubKeyLists[Stable  ]: 0x2ccbb0

SubKeyCount[Volatile]: 0x0

SubKeyLists[Volatile]: 0xffffffff

[   5] Stable SubKeys:

[Idx]   [SubKeyAddr]    [SubKeyName]

[0]     937d3634        Control

[1]     937d3634        Control

[2]     937d3634        Control

[3]     937d3634        Control

[4]     937d3634        Control

[   0] Volatile SubKeys:

 Use ‘!reg knode <SubKeyAddr>’ to dump the key


Trying again recursively to dump the subkeys didn’t work as the debug extension dumped I only ACPI out of the 84 subkeys.


lkd> !reg subkeylist  8bc1a5a0 937d3634

Dumping SubkeyList of Key <Control> :

SubKeyCount[Stable  ]: 0x54

SubKeyLists[Stable  ]: 0x12a968

SubKeyCount[Volatile]: 0x3

SubKeyLists[Volatile]: 0x8001d930

[  84] Stable SubKeys:

[Idx]   [SubKeyAddr]    [SubKeyName]

[0]     93732884        ACPI

[1]     93732884        ACPI

[2]     93732884        ACPI

[3]     93732884        ACPI

[4]     93732884        ACPI

[5]     93732884        ACPI

[6]     93732884        ACPI

[7]     93732884        ACPI

[8]     93732884        ACPI

[9]     93732884        ACPI

[10]    93732884        ACPI

[11]    93732884        ACPI

[12]    93732884        ACPI

[13]    93732884        ACPI

[79]    93732884        ACPI

[80]    93732884        ACPI

[81]    93732884        ACPI

[82]    93732884        ACPI

[83]    93732884        ACPI

[   3] Volatile SubKeys:

[Idx]   [SubKeyAddr]    [SubKeyName]

[0]     8bc31484        hivelist

[1]     8bc31484        hivelist

[2]     8bc31484        hivelist

 Use ‘!reg knode <SubKeyAddr>’ to dump the key


Since I was interested in getting to the address of “Session Manager”, I decided to this manually. On my particular system I looked at the offset of the Session Manger and it was the 60th key in alphabetical order.

I took a guess that the keys are arranged alphabetically so if you have a rough idea you could use that information to get the key you are looking for you could calculate the offset. This alphabetical order offset as it is displayed in the registry doesn’t correlate to the offset in the key list but I guess I have been lucky most times in getting close.

Let me warn you that your mileage may vary but instead of going brute force and looking at every key this approach can be fast if you get lucky. So I looked at the offset 60. Each sub-key takes 8 bytes so it was simple math

Offset of Session Manger = Control key base address + offset to subkeys + (60+1) *8 bytes

Offset to subkeys with some hit and trial is 4.


lkd> dc 936a996c+0x4+0x1e8

936a9b58  00018e30 d9dc4f27 00101220 0041d336  0…’O.. …6.A.

936a9b68  0000c8e0 5f6aaac2 0012acd0 0001c805  ……j_……..

936a9b78  0012ad80 fc209d70 00281b80 bb4d8ce3  ….p. …(…M.

936a9b88  00281950 f964a859 000a4020 dfc3fe3c  P.(.Y.d. @..<…

936a9b98  00281a60 95e28f5b 000367d0 c0743deb  `.(.[….g…=t.

936a9ba8  000fdba8 3d165065 002fd5f0 70574717  ….eP.=../..GWp

936a9bb8  0013ebd0 25abb786 000988a8 0001d599  …….%……..

936a9bc8  000c4ba0 09d542e7 0003d048 2ed17f8e  .K…B..H…….


lkd> !reg cellindex  8bc1a5a0 00018e30

Map = 8bc22000 Type = 0 Table = 0 Block = 18 Offset = e30

MapTable     = 8bc23000

BlockAddress = 937bb000

pcell:  937bbe34


lkd> !reg knode 937bbe34


Name                 : Session Manager

ParentCell           : 0x630

Security             : 0xfc10 [cell index]

Class                : 0xffffffff [cell index]

Flags                : 0x20

MaxNameLen           : 0x2a

MaxClassLen          : 0x0

MaxValueNameLen      : 0x3c

MaxValueDataLen      : 0x2e

LastWriteTime        : 0x 1ca319c:0x2470026e

SubKeyCount[Stable  ]: 0xf

SubKeyLists[Stable  ]: 0xaf8c0

SubKeyCount[Volatile]: 0x0

SubKeyLists[Volatile]: 0xffffffff

ValueList.Count      : 0xe

ValueList.List       : 0x3b8db8


After I got to the Session Manager, to get to the “Memory Management” sub-key the same dilemma presented itself where only the first value was visible and the other sub-keys weren’t so I had to sub-key hunting again J.

lkd> !reg subkeylist 8bc1a5a0 937bbe34

Dumping SubkeyList of Key <Session Manager> :

SubKeyCount[Stable  ]: 0xf

SubKeyLists[Stable  ]: 0xaf8c0

SubKeyCount[Volatile]: 0x0

SubKeyLists[Volatile]: 0xffffffff

[  15] Stable SubKeys:

[Idx]   [SubKeyAddr]    [SubKeyName]

[0]     937bbe94        AppCompatCache

[1]     937bbe94        AppCompatCache

[2]     937bbe94        AppCompatCache

[3]     937bbe94        AppCompatCache

[4]     937bbe94        AppCompatCache

[5]     937bbe94        AppCompatCache

[6]     937bbe94        AppCompatCache

[7]     937bbe94        AppCompatCache

[8]     937bbe94        AppCompatCache

[9]     937bbe94        AppCompatCache

[10]    937bbe94        AppCompatCache

[11]    937bbe94        AppCompatCache

[12]    937bbe94        AppCompatCache

[13]    937bbe94        AppCompatCache

[14]    937bbe94        AppCompatCache

[   0] Volatile SubKeys:

 Use ‘!reg knode <SubKeyAddr>’ to dump the key


In this case at least the challenge was a little less since I was dealing with only 15 sub-keys instead of 84 earlier. So just like earlier first get the cell address of the subkey list.


lkd> !reg cellindex  8bc1a5a0 0xaf8c0

Map = 8bc22000 Type = 0 Table = 0 Block = af Offset = 8c0

MapTable     = 8bc23000

BlockAddress = 93724000

pcell:  937248c4


Now lets go through this list and use the offset of “Memory Management” sub-key which happens to be 11 on my particular system. By hit and trial I found that the correct offset was 0x54 bytes for “Memory management”.

lkd>  dc 937248c4+0x54

93724918  00098068 b76d431e 00040910 092eb60d  h….Cm………

93724928  00304c98 122bab7f 000af950 381a2f7e  .L0…+.P…~/.8

93724938  000af178 0001dd10 00000000 00000000  x……………

93724948  00000000 00000000 ffffffa0 00206b6e  …………nk .

93724958  03a3c1c6 01ca3ef7 00000000 00018e30  …..>……0…

93724968  00000000 00000001 ffffffff 80035830  …………0X..

93724978  00000007 000b0340 0000fc10 ffffffff  ….@………..

93724988  0000000a 00000000 00000010 00000246  …………F…


To confirm it lets calculate the cell address and check the knode.

lkd> !reg cellindex  8bc1a5a0 00098068

Map = 8bc22000 Type = 0 Table = 0 Block = 98 Offset = 68

MapTable     = 8bc23000

BlockAddress = 9373b000

pcell:  9373b06c


lkd> !reg knode 9373b06c


Name                 : Memory Management

ParentCell           : 0x18e30

Security             : 0x3a68c8 [cell index]

Class                : 0xffffffff [cell index]

Flags                : 0x20

MaxNameLen           : 0x24

MaxClassLen          : 0x0

MaxValueNameLen      : 0x30

MaxValueDataLen      : 0x2a

LastWriteTime        : 0x 1ca3ef7:0x 3a88487

SubKeyCount[Stable  ]: 0x2

SubKeyLists[Stable  ]: 0x2a0c48

SubKeyCount[Volatile]: 0x0

SubKeyLists[Volatile]: 0xffffffff

ValueList.Count      : 0xf

ValueList.List       : 0x303f70


What about Values?

Keynodes have values as well.  Now that we have got the actual sub-key we need to get the Value from the sub-key. This task is similar to sub-key hunting as each sub-key maintains a sub-key list and a value list.

Just like with subkeylist there is also a valuelist debugger extension

lkd> !reg valuelist  8bc1a5a0 9373b06c

Dumping ValueList of Key <Memory Management> :

[Idx]   [ValAddr]       [ValueName]

[   0]  9373b1a4        ClearPageFileAtShutdown

[   1]  9373b204        DisablePagingExecutive

[   2]  9373b374        LargeSystemCache

[   3]  9373b39c        NonPagedPoolQuota

[   4]  9373b3cc        NonPagedPoolSize

[   5]  9373b3f4        PagedPoolQuota

[   6]  9373b43c        PagedPoolSize

[   7]  9373b4e4        SecondLevelDataCache

[   8]  9373b53c        SessionPoolSize

[   9]  9373b564        SessionViewSize

[   a]  9373b514        SystemPages

[   b]  93724f9c        PagingFiles

[   c]  935077ec        PhysicalAddressExtension

[   d]  9cd4ffcc        IOPageLockLimit

[   e]  9cd0d434        ExistingPageFiles

 Use ‘!reg kvalue <ValAddr>’ to dump the value


lkd> !reg  kvalue 9373b514


Name      : SystemPages {compressed}

DataLength: 80000004

Data      : c3000  [cell index]

Type      : 4

In our case the value of SystemPages is 0xc300. You can dump the node address 9373b514 directly and change the value with “ed” command. The actual value is at offset 0x8.

lkd> dc 9373b514

9373b514  000b6b76 80000004 000c3000 00000004  vk…….0……

9373b524  00090001 74737953 61506d65 00736567  ….SystemPages.

9373b534  00098538 ffffffd8 000f6b76 80000004  8…….vk……

9373b544  00000004 00000004 00000001 73736553  …………Sess

9373b554  506e6f69 536c6f6f 00657a69 ffffffd8  ionPoolSize…..

9373b564  000f6b76 80000004 00000030 00000004  vk……0…….

9373b574  00000001 73736553 566e6f69 53776569  ….SessionViewS

9373b584  00657a69 ffffffe0 00086b76 00000042  ize…..vk..B…


Now you can change the value of SystemPages directly using “ed”.

lkd> ed 9373b514+0x8  <new value>

Just like with sub-keys sometimes the debugger extension may not show all the values in which case you could do it manually just like earlier. First lets get the cell address of the value list.

lkd> !reg cellindex  8bc1a5a0 0x303f70

Map = 8bc22000 Type = 0 Table = 1 Block = 103 Offset = f70

MapTable     = 8bc25000

BlockAddress = 934d0000

pcell:  934d0f74


Since our Value list had 15 values lets dump out all of them. Dumping all the Value offsets in the value list.

lkd> dc 934d0f74 l 0xf

934d0f74  000981a0 00098200 00098370 00098398  ……..p…….

934d0f84  000983c8 000983f0 00098438 000984e0  ……..8…….

934d0f94  00098538 00098560 00098510 000aff98  8…`………..

934d0fa4  002cc7e8 00c72fc8 00cb3430           ..,../..04..


The Value “SystemPages” is the last one of the 15 but with hit and trial I found that “SystemPages” in the Value list was not  close to offset 15 so it was pretty much manually poking around at each value offset. I found it at offset 11.

lkd> !reg  kvalue 9373b514


Name      : SystemPages {compressed}

DataLength: 80000004

Data      : c3000  [cell index]

Type      : 4


Ring 0 debugging and Windbg – part 1

August 29, 2009

In this article I talk about debugging in ring 0 in the world of Windows(TM). I discuss some tricks which should help you with debugging. Kernel Debugging has become easier but a lot is not very clear and may require tricks to get it to work. Sigh!!!

 I start off illustrating how the basic transports used by Windbg work and then troubleshooting them. I discuss one topic on which I had to scrounge for information which has to debugging the boot environment.

It is not my intention to show you the basics of kernel debugging. Please refer to the Windbg documentation for that. Also, I am not looking at NET debugging (debugging over the network) which is probably going to be available in next version of the OS (Windows 8). There are few more interesting scenarios like accessing and changing registry values through the debugger which I shall deal with in the next article in this series and try and make a last ditch effort to save your machine from crashing components. I will also show how to debug Virtual machines and one of the lesser known facts of Virtual machine to virtual machine debugging. I had to set this up for a self paced class room where people could had access to VM’s only so that once the student was done both the VM’s for rolled back to the original state for the next one. And finally single machine debugging and what it can be used for.

You can use bcdedit (command line) to enable kernel debugging (vista and forward) or msconfig (GUI) Windows XP and forward to enable debugging. The following link shows how to enable debugging for various transports. I want to emphasize however there maybe other ways of doing the same things and please don’t hesitate to add that to comments if you know of other simpler ways.

Vista and forward —


Pre-vista  —

  1. Serial

 target host


If you can’t get serial debugging to work look for the COM port under device manager. If that doesn’t show up then either check your BIOS to see if the COM port is enabled. Sometimes legacy USB support in the BIOS causes this. Disable it in the BIOS.

In case you can’t  still get Windbg to work, check if you set the debugger machine as a debuggee in the past. Also, if you set the debugee as the debugger that can interfere with the claim to the COM Port. My test machines were being used to repro a problem so I used my second dev machine as the test machine and enabled debugging on it. But once I was done I wanted to use the dev machine as the host again.

I just couldn’t get Windbg to work. The solution in this case turned out to be to turn off debugging.  On Vista and forward you could do that by using the following command:

bcdedit /debug off or use msconfig to disable debugging. This requires a reboot to take effect.

2. Debugging with 1394

Lets start of by discussion some of the common questions one has to deal with once starting with 1394 debugging. Hardware!!

1394 Cards you can use – If you don’t have a 1394 port on your laptop, you can buy a 1394 Express card or a cardbus card which cost around $40. On a desktop you could buy a PCI 1394 card too which maybe cheaper. 1394 a/both b work fine. Express cards and some PCI cards come with multiple ports/slots which you could use to connect to multiple debuggee’s using different channels.



When you buy these cards make sure you order the right cable.

4-4 cable

 Ideal when you are debugging laptop to laptop with built in 1394 ports







6-4 cable

Especially if you have a 1394 express card or card bus card on one of the end.


6-6 cable

 Ideal if you have express /card bus cards on both ends


Multiple 1394 cards

If you have multiple 1394 cards on your target how do you pick which 1 to use for kernel debugging. It  is and maybe if the failure on the system as a target is intermittent

Steps to get dual 1394 controllers working on a single PC, where one controller is used to target the PC with KD, and the other controller is left free to target other systems, or to perform other non-KD 1394 tasks.

These steps explain how to dedicate one 1394 controller to be used to target the main PC. Once these steps are taken, any other 1394 controllers should be available to allow the main PC to target other systems for KD debugging.


1.  Open Device Manager.

2 .  Locate the IEEE 1394 node, and expand it. You should see two 1394 controllers. They should both be functioning correctly without any errors. If there are errors (i.e., yellow caution in devmgr or other errors), troubleshoot those first before proceeding.

  1. For each controller, obtain its PCI location information as follows:

 a) Right-click and choose properties,

 b) Click on the Details node,

 c) Select “Location information”,

 d) Write down or note the information.

For example, if you see “PCI bus 21, device 0, function 1” write down 21.0.1. For this example, 21.0.1 will be used for one controller, and 49.0.0 will be used for the other. 21.0.1will be used to allow the main PC to be targeted by another host, and 49.0.0 will be used to allow the system with the two 1394 controllers to act as host, targeting other 1394 targets.

Note: The numbers for location should be in decimal, not hexadecimal.


2. Determine which controller you want the local system to use for its KD (as target), and issue something close to the following bcdedit commands:
a) bcdedit /dbgsettings 1394 channel:25
b) bcdedit /set {current} loadoptions busparams=21.0.1
This effectively makes it so the main PC with the two 1394 controllers capable of being targeted by a host, where that host will specify 1394 channel 25. The main PC controller used to handle this is 21.0.1.
After making the above bcdedit changes, reboot so they take effect.

3. You should be able to target other 1394 “targets” from the same PC.  In this case, 1394 controller 21.0.1 is used to target the PC within which it is installed, while 1394 controller 49.0.0 is available to target other 1394 targets.

Failure to open the 1394 channel on the  * host* error

Using 1394 for debugging

Failed to open 1394 channel 1.  Error0x80070002.

The 1394 debug driver may not be installed.

Attempting 1394 debug driver installation now.

One of the main cause is that 1394 debugging is already enabled on the box you are trying to use as a host i.e. it already acts as a target.


  • Run an elevated cmd prompt and type “bcdedit -debug off”. This should disable the machine from acting as a target and allow the host 1394 debug driver to take over control.
  • Reboot and you should be good to go.

 Disabling the 1394 Host controller

For kernel debugging to work properly with a IEEE 1394 (FireWire) connection, you might have to disable the 1394 host controller on the target computer or disable the 1394 network adapter on the host computer. The required adjustments vary depending on the operating system on the target computer.

Role OS Action to make it work Reason Caution
Target Srvr 2k3 (no ServicePack)XP Sp1 Disable the 1394 host controller If the 1394 host controller remains enabled on computers that are running these operating systems, the core 1394 stack might compete with the 1394 debugging connection for resources. This competition might cause the debugging connection to fail or to be lost shortly after boot time On Windows Vista (any version), Windows Server 2003 with Service Pack 1 (SP1), and Windows XP with Service Pack 2 (SP2), do not disable the host controller on the target computer. In fact, disabling the host controller while running these operating systems can prevent you from debugging the system during some power state changes.Do not disable the host controller on the host computer under any circumstances, regardless of the operating system. 
Host Pre-Windows Vista Disable the 1394 network adapter Disabling the 1394 network adapter can resolve issues that result from bus resets that are caused by loading the 1394 network adapter driver. Do not disable the 1394 network adapter if you are using IP1394. This method does not apply to Windows Vista, because it does not support 1394 network adapters.


To disable the 1394 host controller on the target computer

  1. On the target computer, open Device Manager (that is, click Start, click Run, and then enter Devmgmt.msc) and isable the 1394 contrller.
  2. Restart the target computer.

NOTE: Be careful not to accidentally disable the 1394 network adapter.

To disable the 1394 network adapter on the host computer

  1. On the target computer, open Device Manager (that is, click Start, click Run, and then enter Devmgmt.msc).
  2. Expand Network Adapters, and locate 1394 Net Adapter.
  3. Right-click 1394 Net Adapter, and then click Disable.
  4. Restart the target computer.

NOTE: Before performing kernel debugging over a 1394 cable, you must also install the 1394 Virtual Driver on the host computer.

If you see Windbg returning errors about connection problems (Ctrl-Alt-D for verbose output), you can plug out and plug in the cable or the card. Even on a crashed target this does seem to resolve the problem a lot of the times.

Some myths about 1394 debugging.

MYTH: Only channel 44 can be used for debugging.

Not really!! This comes from the example shown in Windbg. Any port from channel 1 to channel 63 works including 13 J. Are you lizening David? J

MYTH: I can use Virtual COM over USB on a target machine for kernel debugging if it doesn’t have 1394 or serial ports

No it doesn’t work. Try USB debugging.

NOTE: Remember you need to start debugging on 1394 host with elevated privileges since it does involve installing the 1394 debug driver in the Windbg package on the host machine.

3.  Debugging Usb

Kernel debugging over USB works well if the target computer is running Windows Vista or later and the host is running Windows (XP+ preferred). However, it has some additional requirements on hardware.

Hardware/BIOS requirements

  1. Cable

A USB 2.0 debug cable. This cable is not a standard USB 2.0 cable, because it has an extra hardware component that makes it compatible with the USB2.0 Debug Device Functional Specification. You can find these cables with an Internet search for “USB 2.0 debug cable”. These are some of the common ones:

2. USB controller

  • The host computer must have a USB 2.0 controller that is compatible with the Enhanced Host Controller Interface (EHCI) specification.
  • The target computer must have a USB 2.0 controller that is compatible with the EHCI specification, and which supports kernel debugging. Not all EHCI-compatible controllers have this support, and there is no programmatic method to determine whether a given computer does have this support. The USB EHCI Specification states that this support was optional.

    Brands that I know off that support it include:

  •  Intel ICH4 (and later)
  •  VIA USB2.
  • Etc.
  • The USB debugging port must be exposed to the outside of the target computer. A few computers (mostly laptops) have a USB 2.0 controller that does support debugging through a certain physical port, but this port is not accessible from the outside of the computer. The tool usbview.exe which ships with Windbg package can help with this. Sigh!!


     The BIOS must support hi-speed and talk to EHCI

Note : USB debugging does not work over a hub or docking station.

Now that we have looked at the hardware requirements let us look at the software support to enable USB debugging. USB debugging.

Software Setup

If your target computer is running Windows Vista, and your host computer is running Windows 2000 or later, you can perform kernel debugging with a USB 2.0 debugging cable.

To enable debugging with a USB cable in Windows Vista, use bcdedit and set the debugtype parameter to USB. You can set the global debug settings using the /dbgsettings option or set the parameters for specific boot entries using the bcdedit /set option.

To use bcdedit, open a Command Prompt window with elevated privileges (right click Command Prompt and click Run as administrator from the shortcut menu) since it does involve installing the USB  debug driver in the Windbg package on the host machine.

To set the debug settings for USB globally, use the following syntax:

bcdedit /dbgsettings usb [targetname:targetname] 

The following example shows how to specify USB as the global debug setting.

bcdedit /dbgsettings usb targetname:myusb

The following example shows how to specify the USB debug settings for a specific boot entry, and how to use the /debug command to enable kernel debugging for that boot entry.

bcdedit /set {18b123cd-2bf6-11db-bfae-00e018e2b8db } debugtype usb
bcdedit /set { 18b123cd-2bf6-11db-bfae-00e018e2b8db} targetname u2
bcdedit /debug {18b123cd-2bf6-11db-bfae-00e018e2b8db} on

If no {ID} is specified, the settings apply to the current({current}) boot entry. You can use the

bcdedit /enum command to view the current boot entries and their settings.

Please note that in windows 7 you can’t use USB for debugging boot applications (winload.exe/bootmgr.exe)


When multiple USB controllers exist which is mostly the case just like 1394 you need to specify the busparams as follows.

USB controllers

Bcdedit /set {current} loadoptions busparams=0.26.7

NOTE: For USB 2.0 debugging ,the bus parameters must be specified in hexadecimal on Windows Vista and Server 2008 and in decimal on Windows 7. Thanks to Doug Howe for this information. For 1394 debugging, the bus parameters must be specified in decimal. Device manager shows the busparams  in Decimal.

pci device location







NOTE: USB debugging does not work over a hub or docking station.

 4. Boot debugging

Now let us look at a topic which requires a lot of spelunking when it comes to finding information. Bootdebuging referers to debugging components /applications which are loaded much before ntoskrnl.exe. I am going to focus on Windows Vista and forward since the design of boot components has had a major overhaul from Vista. . If you are looking at boot debugging for pre-Vista you could also refer to this well written blog here;:

 The boot components/applications include bootmgr, winload, winresume etc

The way of enabling boot debugging is using the bootdebug parameter. This enables or disables the boot debugger for a specified boot entry. Although this command works for any boot entry, it is effective only for boot applications. It is not sufficient to turn it on using “dbgsettings”.  

The bcdedit /set command sets a boot entry option value in the boot configuration data store. 

You can use this command to configure specific boot entry elements such as kernel debugger settings which is what we will be using it for. In addition to that it can be  data execution protection (DEP) , processor address extension (PAE) options, and to load alternate hardware abstraction layer (HAL) and kernel files. For more details refer to

You can look at the bcdedit elements you specified also from the registry also. This values used to store the  bootoptions specified in boot.ini pre-vista.

Path — “HKLM\System\CurrentControlSet\Control”

Value – “SystemStartOptions”

The following commands allow you to debug the bootmgr and break in very early pretty much as early as the hand off from the BIOS is done to the bootmgr.

For serial the settings should look like:

bcdedit /bootdebug {bootmgr} on

bcdedit /debug {bootmgr} on //this maybe optional

bcdedit /set {bootmgr} debugtype serial
bcdedit /set {bootmgr} baudrate <baudrate, typically 115200>
bcdedit /set {bootmgr} debugport <1>

For 1394 the settings look like

bcdedit /bootdebug {bootmgr} on

bcdedit /debug {bootmgr} on //this maybe optional

bcdedit /set {bootmgr} debugtype 1394

bcdedit /set {bootmgr} channel <channel #>
If you have multiple 1394 controllers then you can also specify:

bcdedit /set {bootmgr} loadoptions busparams= Bus.Device.Function

After you enable the above settings in BCDEdit and reboot the machine, and the appropriate boot binary(in our case bootmgr) should break into the debugger.

 There is one very important step to remember and that is to make sure you have set initial breakpoint using the Windbg before you restart

Kd> sxe ibp

I have also found it helps to do this in case it is the bootmgr you are interested in:

Kd> sxe ld:bootmgr


For example – Here’s how the bootmgr will break into the debugger upon load. Now you can go ahead step through code, set your own breakpoints etc.


Break instruction exception – code 80000003 (first chance)
00443100 cc              int     3
kd> k
ChildEBP RetAddr 
bootmgr!RtlpBreakWithStatusInstruction 00061e94  bootmgr!vDbgPrintExWithPrefixInternal+0xfe 00061ea4 0043e728 bootmgr!DbgPrint+0x11 00061ec0 0043ea30 bootmgr!BlBdStart+0xc3 00061f58 bootmgr!BlBdInitialize+0x17d 00061f68 0041e6a0 bootmgr!ReinitializeLibrary+0x24 00061f74 00401174 bootmgr!BlInitializeLibrary+0x10 00061ff0 00020a9a bootmgr!BmMain+0x174

After you are done you can clear the settings as follows:

Kd> Sxi ibp

Kd> Sxe ld:””

Thanks to David Rocca for proof reading this. Till next time, Happy debugging.