Currently win32kfull bpl does not include most interesting parts of source code.

SMB2 fails due to first .src files and then ESP failure.

See email

@rahulku Unlike the cloud version, the local version terminates when any of the procedures timeout.

For repro:
%sdxroot%\drivers\tablet\platform\math\Input\common\re6_stand\ptls\ls\src>%SDV%\smv /null:local /debug

Needed to create CFGs for Azure and Chakra, as discussed. In other words, don't put body of procedures in header files in the CFG.

We have come to accept a certain failure rate from SMV Cloud due to coordination with Azure back end issues. At this moment we can handle this thru an external rerun mechanism.

However once in a while SMV Cloud repeatedly fails the same driver with a 404. Last week I ran two PTPs with a max rerun of 5. That was not enough. This weekend I did a max rerun of 20. One driver kept failing up to the 14th rerun / 36th hour where I then ran out of patience. This means rerun approach does not work. The fix should either be to eliminative any code that would reuse state/tokens/guids for same driver between reruns. Alternatively a reduction of likelihood that we hit 404s.

The failing folder is here: \redmond\osg\Threshold\Tools\CORE\SiGMa\DPlat-DQT\Share\JakobL\Public\2017-03-31-SMVCloud404times14\win8_rtm_drivers_input_mouser\

The svb.log file shows 14 times:

[FATAL ERROR] Microsoft.WindowsAzure.Storage.StorageException : The remote server returned an error : (404) Not Found. ---> System.Net.WebException: The remote server returned an error: (404) Not Found. [C:\slam1\WDK\win8_rtm\drivers\input\mouser\mydriver.vcxproj]

See attached screenshot from the task manager.

As discussed 2016/11/18:

1. BreakOnError should be default.

2. It is reasonable that essentially all levels of the XML file can override this default behavior. On a per configuration, per action and per command.

project target name has spaces e.g 'Win10 Release' it fails with -1.

Allow SMV to produce defect traces that can be inspected and shared without any further user action.

Here is an example

When reviewing the SMV log (generated with /debug) I see:

[INFO] Launching copy with arguments: "J:\SD\slam\5\sdv\product\osmodel\osmodel.c" "J:\SD\slam\5\WDK\win8_rtm\admin\wmi\ipmi\driver\lib\sdv\sdv-harness.c"
...
J:\SD\slam\5\WDK\win8_rtm\admin\wmi\ipmi\driver\lib>copy "J:\SD\slam\5\sdv\product\osmodel\osmodel.c" "J:\SD\slam\5\WDK\win8_rtm\admin\wmi\ipmi\driver\lib\sdv\sdv-harness.c"
...
The system cannot find the file specified.

Request: It would be great if additionally SMV could echo the error level that the Command exited with.

It would be good to have an option for li2bpl that translates li files to bpl files, that can allow li2bpl to avoid translating the subset of procedures whose translation crashes (but can be caught).

Sometimes avn just hangs there forever. Here is an example output,

E:\linux\drivers\media_v4l2core>avn media-v4l2-core-null.inst.bpl /cloud
[INFO] Running using cloud...
[INFO] SMVFastAVN initialized.
[INFO] Using cloud.
[INFO] Validating XML against schema: e:\smv-fastavn-11-7-2016\bin\Config.xsd
[INFO] Running local scheduler with 4 threads
[INFO] Validating XML against schema: e:\smv-fastavn-11-7-2016\bin\CloudConfig.xsd
[INFO] Scheduler Instance GUID: 2e5627ad-4368-4e35-b4df-4c538e466b20
[INFO] Using plugin FastAVN.SMVFastAVN for analysis.
15 of 862 jobs remaining. Avg(s): 89.13. Std.Dev(s): 240.58[ERROR] Failed to complete action: e4863264-bfd3-4c74-ab7b-4cea1985fe04 (CheckEntryPoints)
12 of 862 jobs remaining. Avg(s): 97.47. Std.Dev(s): 295.36[ERROR] Failed to complete action: ee3ed645-3c49-4c2b-8305-54d6016e9e48 (CheckEntryPoints)
6 of 862 jobs remaining. Avg(s): 111.08. Std.Dev(s): 347.75[ERROR] Failed to complete action: aea26428-6f5f-4349-b0e5-64a588a9622b (CheckEntryPoints)
[ERROR] Failed to complete action: 0069e28a-a7fb-4538-96c8-cce5d72a3ded (CheckEntryPoints)
0 of 862 jobs remaining. Avg(s): 125.97. Std.Dev(s): 411.19
-2 of 862 jobs remaining. Avg(s): 134.12. Std.Dev(s): 444.05

I have to kill the process and run fastavn to merge bugs manually.

Shaobo

Without fix for Bug VSO 9798617 (SMV: ITP: Regression: win8_rtm\net\netio\directaccess\danlb\sys) we have the intercepted build failing.

The evidence is in sdv\smvcl.log:

J:\SD\slam\5\WDK\win8_rtm\net\netio\directaccess\danlb\sys\sdv\headers\ndis.h(5): fatal error C1083: Cannot open include file: 'J:\SD\slam\5\sdv\product\osmodel\wdm\sdv_ndis.h': No such file or directory

Expected: SMV catch this msbuild error and stops verification.

Actual: SMV does not catch this build error. Verification proceed with scan step.

This is a large feature request. It would however be a strong foundational improvement to SMV.

What:

1. Wrap all processes in a job.
2. By default prevent processes "escaping".
3. Provide knobs to control limits on various resources: wall clock time, space, user time. Knobs would be accessible thru xml tags I would imagine.
4. Provide resource statistics upon job completion.

When:
For build, intercepted build and verification. Really any calls to "exec".

How:
I would be open to expose a wlimit API as a managed dll. This dll would essentially provide a replacement of "exec". Next step would be to call this from SMV and additionally expose the most useful knobs as XML.

Hi Rahul.

I ran two PTPs on two separate machines. On both machines I see sporadic examples of errors from the Azure layer yet SMV returns 0 indicating a succesful run.

Fix suggested is trivial: Change regular error to a "Fatal error". Or at least make sure an error results in a non-zero exit code from SMV.

Here is a specific example:
+++
robocopy /mir /nfl /ndl C:\slam1\WDK\src_5239\fail_drivers\kmdf\fail_driver3\driver \redmond\osg\Threshold\Tools\CORE\SiGMa\DPlat-DQT\Share\JakobL\Public\2017-04-10-smv_fatalerror_as_nonfatal
+++

Notice in svb.log you see:
+++
[INFO] 23 of 347 jobs remaining. Avg(s): 8.65. Std.Dev(s): 13.76[ERROR] Failed to complete action: 0a88c33f-6e33-4c7f-bd5c-c0debd6a2b1b (CheckRule)
...
jrerun final errorlevel 0 after 1 attempt(s).
+++

How often does this happen? In average 9.5 times per PTP.
+++
From jakobl1:
C:\slam1\WDK>findstr /m /s /c:"[ERROR] Failed to complete action" svb.log
src_5239\fail_drivers\kmdf\fail_driver3\driver\svb.log
src_5239\kmdf\fakemodem\svb.log
src_6001\input\mouser\svb.log
src_7600\storage\class\cdrom\svb.log
src_ndis\sdvmp_fail_09\svb.log
win8_rtm\drivers\wdm\1394\protocol\avc\svb.log
win8_rtm\drivers\wdm\usb\usb3\ucx\sys\driver\svb.log
win8_rtm\minkernel\storage\sbp2port\svb.log
win8_rtm\net\rras\ndis\ndproxy\svb.log
+++
From jakobl3:
K:\SD\slam\2\WDK>findstr /m /s /c:"[ERROR] Failed to complete action" svb.log
src_5239\kmdf\AMCC5933\sys\AMCC5933ISA\svb.log
src_5239\kmdf\nonpnp_fail4\svb.log
src_7600\storage\class\cdrom\svb.log
src_ndis\sdvmp_fail_30c\svb.log
src_storport\StorportMiniports\storahci\svb.log
win8_rtm\drivers\wdm\1394\protocol\avc\svb.log
win8_rtm\drivers\wdm\ir\circlass\svb.log
win8_rtm\minkernel\storage\sbp2port\svb.log
win8_rtm\net\unimodem\src\sys\modem\svb.log
win8_rtm\printscan\imaging\kernel\usbscan\winnt\svb.log
+++

There are important files that Microsoft projects should all have that are not present in this repository. A pull request has been opened to add the missing file(s). When the pr is merged this issue will be closed automatically.

Microsoft teams can learn more about this effort and share feedback within the open source guidance available internally.

Merge this pull request

The LI generation creates two copies of a procedure invoked within NT_VERIFY (entrypoint BdpReserveScratchVa). See email titled "Understanding the BPL" on 3/31/2017. Some context below

You can see there are two calls to BlMmFreeVirualPages (on blocks D_24 and D_30). You can also see a self loop on D_29, which I believe is SMV’s way of encoding an “assume false”. This is definitely the case that SDV (before slam.exe is run) is redefining NT_VERIFY to something else and we need to change this. But I cannot figure out where this happens. It must be in slamcl, I would guess...

Make a new check /useafterfree for SMV