Re: [Xen-users] Trouble with PCI Passthrough

[David Escott <david.escott@xxxxxxxxxxxxxxxx>]

I only have two x16 slots, both of which are taken up by the two x16
cards, so I can't move anything around.

Since I'm using xm (xl is broken on fedora 17) I can't figure out how
to create logs. Based on the rather limited documentation in man lspci
I think the 00:01.[0-1] indicates that the two devices are behind the
same bridge and therefore would not work per the intel specification.

When I have a bit more time I'll try installing debian and seeing if I
can get things to work with xl.

On Sat, Oct 6, 2012 at 4:18 PM, Casey DeLorme <cdelorme@xxxxxxxxx> wrote:
>
> On Sat, Oct 6, 2012 at 3:31 PM, David Escott <david.escott@xxxxxxxxxxxxxxxx>
> wrote:
>>
>> Casey,
>>
>> Not sure how to read this. Is 01.0-[01] and 01.1-[02] indicative they
>> are behind the switch 01.*? That would make the most sense (and be the
>> most frustrating... how is a consumer to know what the PCI layout of
>> the motherboard is until they open the box?).
>>
>
> You don't.  Name of the game I'm afraid, short of emailing the manufacturer
> you have no idea as to the actual hardware layout.  Also they are not
> obligated to send you that information even if you request it.  They don't
> cater to us crazies so much as the average user who would never need to know
> that level of detail.
>
> I am not a hardware expert, in fact I can shamefully admit that almost all
> of my Xen related hardware knowledge was learned "by accident".
>
> I don't "read" lspci tree output so much as use it to get a basic
> understanding as to the layout, I don't really follow how the numbers
> correspond to their BDF format.
>
> I haven't had a chance to experiment with multiple discrete graphics cards,
> but I have had a motherboard where the card was behind the same switch as
> the onboard NIC and ended up passing both.  At the same time, my current
> board has devices behind a switch and I am able to pass individual
> components without a problem.  I am sure if you ran lspci -vvv there are
> flags or settings for different types of bridges that effect that specific
> situation, but I wouldn't know where to begin looking that information up.
>
>> -[0000:00]-+-00.0  Intel Corporation Ivy Bridge DRAM Controller
>>            +-01.0-[01]--+-00.0  ATI Technologies Inc RV630 [Radeon HD
>> 2600XT]
>>            |            \-00.1  ATI Technologies Inc RV630/M76 audio
>> device [Radeon HD 2600 Series]
>>            +-01.1-[02]--+-00.0  ATI Technologies Inc Device 6819
>>            |            \-00.1  ATI Technologies Inc Device aab0
>>
>>
>> The motherboard is an MSI Z77A-G45.
>>
>
> I did notice your motherboard has three PCI Express slots.  Have you tried
> swapping which card is connected to which slot?  It may sound like a
> slightly less stupid suggestion now that you've checked the tree output of
> lspci.
>
> I also still would recommend trying a fresh Windows 7 install, and comparing
> create log differences, since neither of us can prove that it is in fact the
> pci switch.
>
>>
>> On Sat, Oct 6, 2012 at 2:48 PM, Casey DeLorme <cdelorme@xxxxxxxxx> wrote:
>> > In your situation, here are the points I would investigate:
>> >
>> > What model is your motherboard?  Have you used lspci -tv to see whether
>> > or
>> > not your cards are behind a PCI Switch perhaps?  While NF200 chips are
>> > not
>> > compatible, PLX seems to be, but it'll attempt to pass all devices
>> > behind
>> > it.
>> >
>> > Is it possible when you installed drivers it was only setup for the
>> > 2600?
>> > Have you tried a fresh installation of Windows 7, and passing just the
>> > HD
>> > 7850 after the installation is completed?
>> >
>> > Do your logs have anything special when you pass the HD 7850 solo to the
>> > HVM?  If you run "xl -vvv create" with one then both cards, how does the
>> > output differ?
>> >
>> > On Sat, Oct 6, 2012 at 2:34 PM, David Escott
>> > <david.escott@xxxxxxxxxxxxxxxx>
>> > wrote:
>> >>
>> >> Passthrough seems to be the most common problem, but I don't know
>> >> enough
>> >> of
>> >> the voodoo magic to understand where to start investigating my issue,
>> >> which is that passthrough works but only when passing through multiple
>> >> devices:
>> >>
>> >> dom0: Fedora 17 with stock XEN (a later 4.1 version and a 3.5 kernel)
>> >> Hardware: VT-d capable MSI Z77 w/ i7-3770
>> >> Graphics cards:
>> >>       Intel HD 4000 IGD set as primary in the bios and used by dom0
>> >>       Radeon HD2600XT (from 2005ish) PCI-E slot 1, PCI 0000:01:00.[0,1]
>> >>       Radeon 7850 (brand new)              PCI-E slot 3, PCI
>> >> 0000:02:00.[0,1]
>> >>
>> >> What works:
>> >>     dom0 with the HD4000 IGD
>> >>     domU Windows 7 when both Radeon cards are passed through
>> >>               ie stubbing out 0000:01:00.[0-1] and 0000:02:00.[0-1]
>> >> and giving both to Win7
>> >>
>> >> What doesn't work (and what I dont understand):
>> >>     domU Win7 when only one of the two devices is passed to Windows
>> >> (in which case I get the error 43) [tried it both ways, passing only
>> >> the 2600 or only the 7850, and same error either way]
>> >>
>> >> Given that passthrough works, I dont understand why it would not work
>> >> when I only pass one of the devices. Is there something I should be
>> >> looking at (DMA ranges? IRQs?) to determine what linkage there is
>> >> between these two devices?
>> >>
>> >> Thanks,
>> >> David
>> >>
>> >> --
>> >> David Wright Escott
>> >> A comathematician turns cotheorems into phi.
>> >>
>> >> _______________________________________________
>> >> Xen-users mailing list
>> >> Xen-users@xxxxxxxxxxxxx
>> >> http://lists.xen.org/xen-users
>> >
>> >
>>
>>
>>
>> --
>> David Wright Escott
>> A comathematician turns cotheorems into phi.
>
>



-- 
David Wright Escott
A comathematician turns cotheorems into phi.

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