Re: GeForce 4xx/Fermi to Quadro Modifying Quide

[Gordan Bobic <gordan@bobich.net>]

 On Mon, 2 Dec 2013 21:50:17 -0500, Konrad Rzeszutek Wilk 
 <konrad.wilk@oracle.com> wrote:
> On Wed, Nov 27, 2013 at 04:49:19PM +0000, Gordan Bobic wrote:
>> On Wed, 27 Nov 2013 11:43:03 -0500, Konrad Rzeszutek Wilk
>> <konrad.wilk@oracle.com> wrote:
>> >On Wed, Nov 27, 2013 at 04:39:26PM +0000, Gordan Bobic wrote:
>> >>On Wed, 27 Nov 2013 12:26:44 -0400, Konrad Rzeszutek Wilk
>> >><konrad@darnok.org> wrote:
>> >>>>> From what I can see there are several variants of the GTX460,
>> >>>>> some with the GF104 GPU, others with the GF114 GPU. There is 
>> no
>> >>>>> Quadro based on the GF114, but there are two mobile Quadro
>> >>>>> variants based on it: Q3000M and Q4000M.
>> >>>>>
>> >>>>> The GTX460 variants have the following device IDs:
>> >>>>> GeForce GTX 460		0x0E22
>> >>>>
>> >>>>That is the one!
>> >>>>
>> >>>>> GeForce GTX 460 SE	0x0E23
>> >>>>> GeForce GTX 460		0x0E24
>> >>>>>
>> >>>>> The corresponding Quadros have IDs:
>> >>>>>
>> >>>>> Quadro 3000M	0x0E3A
>> >>>>> Quadro 4000M	0x0E3B
>> >>>>>
>> >>>>> So you _should_ be able to modify your card into one of those,
>> >>>>> since the adjustment range is within the bottom 5 bits of the
>> >>>>> device ID and the GPU itself is the same.
>> >>>>>
>> >>>>> Q4000M looks like a closer match. I'd say it's certainly
>> >>>>> worth a try. :)
>> >>>
>> >>>Success! My desktop machine now sports an Q4000M. It went
>> >>>flawless - with most of the time spent on double-checking
>> >>>the computations.
>> >>
>> >>Splendid! Is this a Linux or Windows machine (or do you dual
>> >>boot it)?
>> >
>> >This was Windows 7.
>>
>> If the Nvidia driver detected it as a Quadro and installed
>> itself without problems, that is indeed a very good sign.
>
> I am quite happy to report that it works with Windows 7 (32-bit).

 Excellent! I have added it to the wiki page of tested working adapters. 
 :)

> What is interesting is that it works nicely with SeaBIOS (couldn't
> the old BOCHS BIOS one to do PCI passthrough?). SeaBIOS does not
> yet support PCI hotplug so I had to define the PCI entries in the
> guest config, which means it was:
>
> [root@phenom konrad]# cat /root/Win7.xm | grep -v \#
> builder='hvm'
> memory = 2048
> name = "Windows7"
> vcpus=2
> vif = [ 'type=ioemu, mac=00:0F:4B:00:00:65,bridge=switch' ]
> disk=[ 'phy:/dev/vg_private/Win7,hda,w']
> vnc=1
> videoram=8
> vnclisten="0.0.0.0"
> vncpasswd=''
> stdvga=0
> usb=1
> usb_add="host:045e:0039"
> usbdevice='tablet'
> pci=['05:00.0','05:00.1']


 I always define it in the config anyway.

> 05:00.0 VGA compatible controller: NVIDIA Corporation GF104GLM
> [Quadro 4000M] (rev a1)
> 05:00.1 Audio device: NVIDIA Corporation GF104 High Definition Audio
> Controller (rev a1)
>
>
> Linux on the other hand is more troublesome. This is with Fedora 20:
>
> (nouveau kernel driver looks OK)
>
> [    30.747] (EE)
> [    30.747] (EE) Backtrace:
> [    30.749] (EE) 0: /usr/bin/Xorg (OsLookupColor+0x129) [0x473899]
> [    30.749] (EE) 1: /lib64/libpthread.so.0 (__restore_rt+0x0) 
> [0x3504e0f74f]
> [    30.750] (EE) 2: /usr/bin/Xorg (xf86ProbeOutputModes+0x9c1) 
> [0x4c3511]
> [    30.750] (EE) 3: /usr/bin/Xorg (xf86PruneDuplicateModes+0x152c)
> [0x4cdbdc]
> [    30.750] (EE) 4: /usr/bin/Xorg (RRGetInfo+0x9c) [0x50f4dc]
> [    30.750] (EE) 5: /usr/bin/Xorg (ProcRRGetProviderProperty+0x1129)
> [0x517069]
> [    30.750] (EE) 6: /usr/bin/Xorg (SendErrorToClient+0x427) 
> [0x43a3d7]
> [    30.752] (EE) 7: /usr/bin/Xorg (_init+0x3b1a) [0x42c02a]
> [    30.752] (EE) 8: /lib64/libc.so.6 (__libc_start_main+0xf5) 
> [0x3504621d65]
> [    30.752] (EE) 9: /usr/bin/Xorg (_start+0x29) [0x428c45]
> [    30.753] (EE) 10: ? (?+0x29) [0x29]
> [    30.753] (EE)
> [    30.753] (EE) Segmentation fault at address 0x200000000000
> [    30.753] (EE)
>
> (so using nouveau)

 Hmm... What distro and version was David Techer using when he
 got his Linux domU working with GPU passthrough?

> And with Ubuntu 12.10 with Nvidia driver:  304.88-0ubuntu0.1
>
> [  859.790255] nvidia: module license 'NVIDIA' taints kernel.
> [  859.790259] Disabling lock debugging due to kernel taint
> [  859.813635] xen: --> pirq=20 -> irq=36 (gsi=36)
> [  859.813705] vgaarb: device changed decodes:
> PCI:0000:00:05.0,olddecodes=io+mem,decodes=none:owns=io+mem
> [  859.813707] vgaarb: transferring owner from PCI:0000:00:05.0 to
> PCI:0000:00:03.0
> [  859.813874] NVRM: loading NVIDIA UNIX x86_64 Kernel Module  304.88
> Wed Mar 27 14:26:46 PDT 2013
>
> it looks to be OK, but Xorg.0.log says:
>
> [  1028.381] (II) NVIDIA GLX Module  304.88  Wed Mar 27 14:46:57 PDT 
> 2013
> [  1028.381] Loading extension GLX
> [  1028.381] (II) LoadModule: "nvidia"
> [  1028.381] (II) Loading
> /usr/lib/x86_64-linux-gnu/xorg/extra-modules/nvidia_drv.so
> [  1028.382] (II) Module nvidia: vendor="NVIDIA Corporation"
> [  1028.382]    compiled for 4.0.2, module version = 1.0.0
> [  1028.382]    Module class: X.Org Video Driver
> [  1028.411] (II) NVIDIA dlloader X Driver  304.88  Wed Mar 27
> 14:28:14 PDT 2013
> [  1028.411] (II) NVIDIA Unified Driver for all Supported NVIDIA GPUs
> [  1028.411] (++) using VT number 7
>
> [  1028.421] (EE) No devices detected.

 304.88 driver is quite old, maybe it doesn't include support for
 the Q4000M. Maybe a more recent driver is worth trying?

 Gordan