Hide Forgot
Description of problem: The two IO tests in WHQL that test the QXL driver in the XP SPICE client, kills the VM from running. The VM has to be powered on again to resume WHQL testing. The WHQL tests are: 991 Sleep Stress With IO 995 Common Scenario Stress With IO Version-Release number of selected component (if applicable): RHEL 6.1 Server with hybrid repo spice-client-0.8.0-2.el6.x86_64 spice-server-0.8.0-1.el6.x86_64 spice-server-debuginfo-0.8.0-1.el6.x86_64 spice-vdagent-0.6.3-8.el6.x86_64 spice-client-debuginfo-0.8.0-2.el6.x86_64 qemu-kvm-0.12.1.2-2.168.el6.x86_64 kernel-2.6.32-131.4.1.el6.x86_64 How reproducible: 100% Steps to Reproduce: 1. 2. 3. Actual results: Tests kill VM Expected results: Test should pass Additional info:
I am also adding the Windows7x86 client that has the same behavior when being tested with the "991 Sleep Stress With IO" WHQL test.
For future WHQL bugs - it helps if you can reproduce using the executable that the test actually runs. Since setting up a full DTM controller+client is both a lot of work and time consuming to rerun the test.
ok, I've been trying to find 991 - all I have here is "State Blocks (32-bit) under "WDK Tests\Display\32-bit". So I take back the "Future" part of the previous comment. Bill - can you provide me with detailed failed tests so I can find the equivalent in the DTM instance I have over here and run them to reproduce? p.s. As you know I have rerun the WHQL and failed a few tests, but not the sleep stress, and I want to make sure it is the same, since the numbers don't match. I am running Microsoft DTM Studio 1.6.7367.000 (2011/02/03-17:00). What are you running? Alon
I am running Microsoft DTM Studio 1.6.8367.000.
The child test that fails in the WHQL 991 test are the 900 "Sleep_Stress_With_IO. The tasks that are part of Job 900 are: Test 899: Un-install WDTF: $\WDK Tests\Library\WDTF\Un-install WDTF Check if test was completed: cmd /c if exist SleepTestNotFinished.log (exit /b 1) ELSE (exit /b 0) Copy Log Execute Sleep_Stress_With_IO Script. Execute Device_Status_CHeck Script.
Bill Hi, These two tests in my DTM pass with all windows versions I tested (XP 32, Win 7 32, Win 7 64). The below scratch build with the latest qemu-kvm (I think it's only a qxl-win fix, but better to do the testing with the latest qemu-kvm): http://brewweb.devel.redhat.com/brew/taskinfo?taskID=3682780 Under my DTM the tests and versions are: Sleep Stress With IO - 672 Common Scenario Stress With IO - 676 The windows 7 64 bit version of the test submissions is: operating system: Windows 7 Client x64 process architecture: AMD64 qualification program: Unclassified qualification level: Unclassified Signature Is this the same way you do it? I'm also running with a single QXL device, the default memory, and using virtio-net networking device. The later is very important, although I understand that using e1000 with the latest Intel drivers also works. The non working (as in, Common Scenario Stress With IO will fail because of them) options are: the default rtl8139 device with any driver. the e1000 device with the default Microsoft driver. I'm using Red Hat VirtIO Ethernet Adapter Driver Version 6.0.209.605 and Driver Date 2/21/2011, I expect/hope anything newer would serve as well. Alon
See https://bugzilla.redhat.com/show_bug.cgi?id=688883#c30 for note on skipped versions.
Just an additional note on how to set the machine under test (the vm): * use 64MB of vram (recent virt-manager fixed this, older ones reused the cirrus logic default of 9MB, that was bad). * use virtio-net-pci, do *not* use rtl8139, you can also use e1000 if you are using the newest intel drivers (*not* the microsoft default ones on win7) * do not install any guest agent / virtios-serial is verboten. * generally avoid adding any hardware that isn't needed for the test. * use the qemu-kvm version in the "Fixed in" or newer (I guess if you are verifying the bug you want to use the lowest version possible, to make it a tight verification)