[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] ATI/AMDs atikmpag.sys BSOD while vga passthrough



On Tue, Feb 12, 2013 at 4:20 AM, Matthias <matthias.kannenberg@xxxxxxxxxxxxxx> wrote:
Hi and thank you both for your input.

>From what I can tell, the things you mentioned are pretty similar to
what i wrote in my summary in the initial mail. But it's interesting
that you both got different cards running with CCC. So the CCC might
be an actual issue of either vga- and mainboard combo or vendor
specific. May I ask what setup's you got, especially the vendors /
exact models of your HD6870. ?


My card is an XFX Radeom HD 6870, the exact details on the box:

Model No: HD-687A-ZD
Part No: HD-687A-ZDFC
HD 6870 900M 1GB DDR5 DUAL MINIDP HDMI DUAL DVI PCI-E


My configuration is as follows:

Debian Wheezy Dom0 x64 - Control OS & Base System
Debian Wheezy DomU x64 - Web Development & DNS Server
IPFire (Debian Squeeze) - Firewall/Router/Proxy
Windows 7/8 x64 - Multimedia & Gaming VM

I pass my graphics card to Windows, and with my latest configuration onboard wireless to my IPFire for in-box WAP. ÂDom0 sits behind the IPFire HVM and is protected by the same Firewall as the other machines. ÂAs this is a home setup I am not as concerned with internal security, but I do plenty of backups since the only real danger behind the firewall is myself. ÂI use an uEFI boot loader, so at certain points I did modify Xen source since it did not see all my RAM otherwise. ÂHowever I did not do anything with VGA BIOS.


I have tested my card and configuration successfully on two sets of hardware.

Set 1:

MoBo: ASRock Z68 Extreme4 Gen3
CPU: Intel Core i7 2600
RAM: 12GB 1333Mhz Corsair XMS (2x2G 2x4G)
SSD: 240GB OCZ Vertex 3


Set 2:

MoBo: ASRock Z77 Extreme9
CPU: Intel Core i7 3770
RAM: 32GB 1600Mhz Corsair XMS (4x8GB)


I have tested an ASRock Z68 Extreme7 Gen3, which failed to provide for my needs on several levels. ÂIt had an NF200 PCIe Bridge which did not support passthrough and sat behind most of the PCI Express slots leaving me with only one 8x for a Graphics Card. ÂIt also had a (multiple?) PLX PCI Bridge, which treated several entirely different devices as "functions" in BDF reference, and passing one of them to pciback hid four out 7 behind it, including my dual lan controller which created lots of confusion and trouble.
Â
For the well-known fix for the atikmpaq.sys BSOD, at least in my
opinion you can ignore all the solutions posted which are not xen
related. I think the BSOD here is clearly xen related since at least
for my it only happens with the xl toolstack and not the xm one, with
the same domU. But if someone really got lucky with such a non-xen
fix, it would be interesting to see, too..

I started my experiments on Xen after xm had been deprecated and since I was building new from source I never bothered trying it. ÂFrom what I read the success stories were just as mixed as those using xl.

I am not a specialist, just an enthusiast, so while I have performed tests they were not exactly professionally documented. ÂAfter getting a successful install by accident once out of 5 attempts, I decided to run tests to narrow down the possible causes of my failed attempts. ÂI freshly rebuilt Xen and installed Windows 80 times, changing the process each time to attempt to pin-point the cause of failure. ÂMy conclusion was as stated, anytime I attempted to install or upgrade the drivers and the physical system had not been freshly rebooted OR I had not ejected the card then one of several things would happen (these appeared to be entirely random consequences). ÂI would encounter a BSoD during installation, sometimes with and sometimes without CCC (it appeared to occur less often if I de-selected CCC). ÂI sometimes encountered a BSoD rebooting and every subsequent. ÂOthertimes it almost worked but would BSoD over time, and this got worse (happened faster) as I continued to use the system. ÂAnother result was severe horizontal tearing on video playback, but the system would work normally otherwise even on reboots.

My conclusion was pretty simple, and it has worked for me successfully on over 30 Windows installations (25 being test runs). ÂI shared details about queer FLR issues in previous email chains, but many xen-users have reported entirely different problems or that my solution did not work for their configuration. ÂSo though I would like to imagine otherwise it is very possible that every card's make, model and vendor could behave differently.
Â
Last but not least, can you say something about your setup? I would be
especially curious about if you do some vga bios stuff, like
extracting and including it into the xen build, and what kernel and
domU parameters you use, for example if iommu=1 or iomu=soft @ kernel
and stuff like pci_msitranslate=1 and pci_powermanagement in domU
config (and if the viridian=1 option does actually do anything)..


I researched and tested setting the iommu kernel flag, but during all of my tests it never made any difference in the results so eventually I started omitting it. ÂThis is very different from the results others have gotten, and I haven't the foggiest why.

If you google Xen Man Pages you'll find complete documentation of xl.cfg for VM Configurations, which includes descriptions for all the flags for virtual machines.

The additional pci flags (global and otherwise) I tried but saw no visible differences in use or they caused brand new errors instead of solving anything so I also omitted them.

I have the viridian flag on for Windows, as the Xen Man Pages describes it as potentially boosting Windows Vista and newer performance by enabling Hyper-V extensions. ÂI do not know if this has had any real impact on my performance, as I also installed GPLPV drivers, if I have time I may perform some tests later.
Â
There are a lot of vga passthrough tutorials lately, but from what
it's worth, in my opinion they all lack some detail because everyone
simply describes the setup he got working and there is no
comprehensive summary about the benefits of the different options.


By the definition of the word detail I would disagree that many of the guides lack it, more that they are too detailed. ÂI created a guide and just like you said I was very specific about my hardware. ÂHowever that is because success with Xen is very hardware dependent. ÂIt would be extremely difficult to write a "catch-all" guide on passthrough that retained any usefulness if every card make/model/vendor not to mention every other part in a computer could behave even slightly differently.

I also don't know anyone who has both the time and money to purchase and test all available combinations of similar equipment, but if vendors are interested in lending equipment I'd happily volunteer to test it.

Not sure what you mean by different options, is that hardware options or the flags mentioned above?
Â
It would be great if you can help with your experience.


I hope this information is useful, and look forward to following further replies.
Â



2013/2/7 Casey DeLorme <cdelorme@xxxxxxxxx>:
> I have an HD 6870 and working CCC, tested on both Windows 7 and windows 8.
> Don't have any BSoD's either.
>
> I only had problems with CCC and drivers when I attempted to install or
> update drivers without having freshly rebooted or ejected the card.
>
> Unfortunately Windows 8 attempts to auto-install the drivers at boot time
> and this caused some headaches during my first go at it.
>
> I experience performance problems after rebooting the HVM, to fix it I just
> eject the card from the "Safely Remove Hardware and Eject Media" icon.
>
>
> On Thu, Feb 7, 2013 at 4:56 PM, David TECHER <davidtecher@xxxxxxxx> wrote:
>>
>> Issues for CCC depends on your Video Card too.
>>
>> - Got a HD 79070. It worked for me to add CCC while installing the 13.2
>> beta driver.
>>
>> Â Â Â Unfortunately there was ÂBSOD issue when rebooting the domU.
>>
>> Â Â ÂI had to install .NET 4 Framework and CCC now works as expected :)
>>
>> - CCC is not supposed to work for HD 6870 but it worked for HD 7970.
>>
>> FYI to fix atikmap.sys issue there is a well-known workaround with
>> something like
>>
>>
>> http://www.computertipsfree.com/fix-43029-atikmdag-sys-blue-screen-bsod-in-windows-7/
>>
>> expand -r atikmdag.sy_ atikmdag.sys
>>
>> But it doesn't work for me :(
>>
>> The good news is that I've just installed Crysis 3 beta demo. It works
>> like a charm!
>>
>>
>> ________________________________
>> De : Matthias <matthias.kannenberg@xxxxxxxxxxxxxx>
>> Ã : xen-users@xxxxxxxxxxxxx
>> Envoyà le : Jeudi 7 fÃvrier 2013 20h14
>> Objet : [Xen-users] ATI/AMDs atikmpag.sys BSOD while vga passthrough
>>
>> Hi everyone
>>
>> since I'm currently migrating from xm to xl and now are also a victim
>> of the said error, I started researching on it and found that many
>> people had the issue but not many found solution and behind this BSOD
>> (=Bluescreen of Death), there are actually multiple errors in hiding.
>>
>> What I would like to do is gather as much info as possible in this
>> thread to form a kind of comprehensive guide to fixing this issue.
>>
>> From what i gathered so far, there are at least 3 different problems
>> with the same symptopms.
>>
>> 1)
>> Situation: After booting the dom0 and booting a windows domU with an
>> ATI/AMD Card, Windows generally works fine, but after rebooting the
>> system, or shutting down / restarting, the Bluescreen occurs or the
>> performance massively below normal.
>> Error: Most likely this is cause the Card is not properly resetted (no
>> FLR [function level reset] )
>> Solution: If the domU reboots into BSOD, you are screwed and have to
>> reboot the dom0. If it only degrades performance, eject the card and
>> reinitialize to perform a manual reset. Or use the xm toolkit.
>> Status: Don't know if it is fixed yet
>> Source: http://lists.xen.org/archives/html/xen-users/2012-02/msg00443.html
>> , http://wiki.xen.org/wiki/Talk:Xen_VGA_Passthrough_Tested_Adapters
>>
>> 2)
>> Situation: Booting a windows domU results in a BSOD when the CCC
>> (=Catalyst Control Center) installed.
>> Error: The CCC tries to set some specific settings (most likely
>> performance tweaks) within the driver which the driver can't pass to
>> the vga through passthrough mode. Result is a BSOD.
>> Solution: Deinstall the CCC in windows safe mode. The driver itself
>> works without it pretty fine.
>> Status: Don't know if it is fixed yet
>> Source: http://lists.xen.org/archives/html/xen-users/2012-03/msg00617.html
>> Note: I don't really know if this holds for the Vision Engine Control
>> center, too. the VECC is basically the same as the CCC, so i assume it
>> does. I simply haven't found any things referencing it.
>>
>> 3)
>> Situation: This is actually the issue I have. Things seem a lot more
>> random then the other ones. A BSOD happens basically in 80% of booting
>> the Windows domU. Which means, after rebooting dom0, it mostly works.
>> Rebooting or shutting down / starting the domU works sometimes (Once i
>> achieved 6 consecutive reboots or shutdowns / starts before the BSOD
>> happened) but after I get one Bluescreen, almost never (I can't
>> exactly remember if really 'never') comes up again. Of course I tried
>> the above solutions but this didn't helped me. I also tried different
>> different versions of the Catalyst driver (from 12/1 to 13/1) and
>> current releases of xen-4.2-testing and xen-unsable. As I said, this
>> is when using the xl toolkit. I never had this issues using xm.
>> I thought this might belinked to me passing other PCI devices in
>> addition (I'm passing some USB controller to the domU) but even
>> deactivating them didn't help. What I actually found helping boosting
>> the success rate (but not solving the overall issue) is deinstalling
>> all VGA-Related AMD Driver. Windows then runs on it's own VGA-Driver
>> (the graphic card is currectly identified and this is an actual ati
>> driver, so I'm not talking about some generic standard vga driver. I
>> don't know where is driver is coming from).
>> I've attached an analysis file i've gethered from the bluescreen crash
>> dump. I created it using BluescreenView. This is more for currectly
>> identifieng the different BSODs, I guess.
>>
>>
>> Some Infos:
>> Kernel: 3.7.4 kernel.org kernel without patches
>> Xen: latest xen-unstable (but also tested with xen-4.2-testing)
>> Processor: AMD Phenom 1090T
>> Mainboard: Asus Crossfire IV
>> VGA: Radeon HD5700 and Radeon HD5450
>> Driver: Currently running Catalyst 13/1 but tested nearly everything
>> between 12/1 and 13/1
>>
>> Error: Don't know yet
>> Solution: Don't know yet
>> Status: unsolved
>> Source:
>>
>>
>> Okay, so basically that's the story. It would be great if some people
>> can complete and extend the list. Especially some who already fixed
>> the issue on their side. (And of course it would be great if someone
>> has an idea for my problem).
>>
>> Thanks
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@xxxxxxxxxxxxx
>> http://lists.xen.org/xen-users
>>
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@xxxxxxxxxxxxx
>> http://lists.xen.org/xen-users
>
>

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.