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

[Xen-devel] Problems with pci/vga passthrough


  • To: xen-devel@xxxxxxxxxxxxx
  • From: Diederik de Haas <didi.debian@xxxxxxxxx>
  • Date: Wed, 18 Jan 2017 02:37:16 +0100
  • Delivery-date: Wed, 18 Jan 2017 04:54:03 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xen.org>

Hi!

I reported/discussed this earlier on IRC, but was asked to report it here too.

The goal I tried to accomplish was getting vga passthrough working to get 
hardware acceleration and OpenGL 2+ working for a vm containing 
KDE Plasma 5 (Debian Stretch). 
The graphics card I tried it with was an (old) NVidia GeForce 6200 TurboCache 
and even though I already had a suspicion that it would be challenging, 
which was confirmed on IRC, I wanted to try/figure out what I should be doing 
as I'm (very) new to Xen. 

But as soon as I enabled 'gfx_passthru = 1` and `pci = [ '02:00.0' ]`, the 
`xl create <vm-config>` process kept crashing and the only way I was able to 
stop it was by doing `kill <xl-create-process>` and the <xl-create-process> was 
"/usr/lib/xen-4.8/bin/xl create -c /etc/xen/tradestation.home.cknow.org.cfg" 
(attached as tradestation.xen.cfg)

Here's some console output illustrating the crashing:
root@cknowsvr01:/home/diederik# xl create -c 
/etc/xen/tradestation.home.cknow.org.cfg 
Parsing config from /etc/xen/tradestation.home.cknow.org.cfg
libxl: notice: libxl_numa.c:518:libxl__get_numa_candidate: NUMA placement 
failed, performance might be affected
libxl: error: libxl_qmp.c:287:qmp_handle_error_response: received an error 
message from QMP server: Could not set password
root@cknowsvr01:/home/diederik# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0 112773    32     r-----      80.0
tradestation.home.cknow.org                  5 16383     1     --psc-       0.0
root@cknowsvr01:/home/diederik# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0 112773    32     r-----      86.2
tradestation.home.cknow.org                  6 16383     1     ---sc-       0.0
root@cknowsvr01:/home/diederik# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0 112773    32     r-----      86.3
tradestation.home.cknow.org                  6 16383     1     ---sc-       0.0
root@cknowsvr01:/home/diederik# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0 112773    32     r-----      90.4
root@cknowsvr01:/home/diederik# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0 112773    32     r-----      92.8
tradestation.home.cknow.org                  7 16383     1     ---sc-       0.0
root@cknowsvr01:/home/diederik# xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0 112773    32     r-----     132.1
tradestation.home.cknow.org                 13 16383     1     ---sc-       0.0

Trying with `xl destroy <id>` didn't help, `xl pause <id>` paused the 
crashing, but as soon as I tried to activate it again, the crash-loop 
continued and as said before `kill <xl-create-process>` was the only way out.

I've also attached the output of `xl info` and `xl dmesg` as that may 
provide some info as well. 
In the `xl dmesg` you'll notice various crashes as well and that is very likely 
due to hardware-wise failing of the vga card which disappeared when I took 
the vga card out of the system.

Furthermore, I have also attached a description of the process I tried to get 
it working when trying to use `xl pci-assignable-add` which resulted in a 
complete system hang. In retrospect it may have been caused by hardware 
failure, but still, getting a complete system hang when executing a `xl` 
command isn't nice. 
But as I said, I'm a n00b wrt Xen, so I figured I better provide too much 
info then too little.

No matter the hardware failure, I see the continuous loop of the  failed vm 
creation attempt as a real problem as I could only stop that by using 
`kill <pid>` and that was before the hardware (really) died.

As it looks like the hardware failure of the vga card even prevented booting 
of the whole system at some point, I have removed it and have no plans to 
put it back in, even for testing purposes. 
But otherwise I'll try to answer any questions to the best of my abilities.
I have ordered an XFX Radeon RX 460 - 4GB GDDR5 (passively cooled) and 
when that arrives I can try to see whether I can reproduce it with that too, 
but it may take a couple of days and it is a completely different card.

Lastly, I subscribed to this list but due to the huge volume, I unsubscribed 
again, so a CC of any response would be preferable.

Cheers,
  Diederik

Attachment: tradestation.xen.cfg
Description: Text document

Attachment: xl_info+xl_dmesg-with-nvidia-card.txt
Description: Text document

Attachment: xen-xl-pci-assign-system-hang.txt
Description: Text document

Attachment: signature.asc
Description: This is a digitally signed message part.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel

 


Rackspace

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