Hi Johannes,
your chipset needs to support IOMMU. As far as I've seen this in different
papers and on xen-devel, an additional amd_enable_iommu=1 entry in the grub
line should enable this for xen.
I don't know where to find a list of chipsets supporting this feature.
ht2000/ht1000 does not, ht1100 should (at least Wei Wang posted a patch for
that chipset in February, don't know if this made it to the mainline).
I suggest you posting this question with a description of your hardware
on xen-devel.
good luck!
Stephan
Johannes t. schrieb:
Hello Daniel, hello Stefan, hello xen-users,
just read this thread with high interest, read through what the Wiki has
on VT-D. I wonder if or when the AMD IOMMU will be usable to do the
same. I have an AMD 2400BE here (Dualcore, 2,3GHz, SVM, x64) and I
happen to have some funny PCI card that will only work with Windows.
I did not find anything production ready on AMD IOMMU, I do see the AMD
IOMMU switchoff line when Xen is booting.
Could someone please give a short status notice on what happens with HVM
PCI passthrough and AMD?
Thanks in advance :)
Regards,
Johannes
Am 28. Juli 2008 09:17 schrieb Stephan Seitz <s.seitz@xxxxxxxxxxxx
<mailto:s.seitz@xxxxxxxxxxxx>>:
Hi Daniel,
it's speculative, but have you additionally tried to use only one cpu
and pin that one to core?
domU:
# no SMP
vcpus = '1'
# second core
cpu = '1'
Daniel Kao schrieb:
Hi Stephan,
Thanks for the response.
I actually tried leaving default settings and irqbalance was
installed by default in my CentOS 5.2 install. Maybe I should
try turning irqbalance off? It definitely seems to be
irq-related. :(
03:02pm root@vps-cluster-01:/root% service irqbalance status
irqbalance (pid 3585) is running...
03:04pm root@vps-cluster-01:/root%
Cheers,
Daniel
Stephan Seitz wrote:
Hi,
I noticed similar behaviour of missing interrupts in domU's
(well, PV/pci-
passthrough) if the dom0 is under load (sometimes only
slightly above
idle has been enough).
Leaving dom0 all cpu's (just the default settings) and
installing irqbalance
in dom0 solved this issue.
I don't know if it's of much use to you, but it's worth to
try ;)
cheers,
Stephan
Daniel Kao schrieb:
Hi All,
So I've got a DVB-S PCI Express card ($30 USD Twinhan
AD-SE200 off eBay) that's being passed via VT-d/pciback
in Xen 3.2.1 under CentOS 5.2 to a Windows XP SP3 which
works! ... except for one small issue...
Motherboard is an Intel DQ35JO that much people has had
success with and an Intel E8400 CPU.
The issue is that when dealing with FTA MPEG-2 transport
streams (DVB-S based @ Galaxy 25 @ 97.0W), all is well
until I start putting a slight I/O (whether disk or
otherwise) load under dom0. The issue is very similar
and indicative of TV tuner cards and other PCI devices
which have poorly matched PCI latency timers. Basically
when dom0 isn't under load, the DVB-S card and Windows
XP SP3 works great. Once there's a slight load (I/O
load; not CPU) in dom0, I get continuity errors in the
MPEG-2 packet transport stream. Normally, this is an
issue with signal strength of the receiving satellite,
but this is not the issue. These continuity errors are
easily reproducible. All I have to do is log into my
CentOS dom0's gnome desktop via vnc (thru vncserver),
launch the "Virtual Machine Manager", and instantly the
MPEG-2 packet stream will start seeing continuity and
stream errors. The instant I close the "Virtual Machine
Manager" and log out of X, everything returns to normal.
There's also a Dell PERC 5/i card (flashed to an LSI
MegaRAID 8408E firmware) that's running 8-750GB drives
in RAID-5 which is visible in dom0 as a file-server
using samba. And that's about it.
I'm not sure where to start looking in resolving this
issue but issues of this type when searching other
forums and answers seems to be PCI latency related and
resetting them on certain PCI bus devices to fix
continuity errors (even in a non-VM environment and
straight up Windows-based OS). Doing an lspci -vvv on
dom0 under a Xen kernel, I noticed every single device
on the PCI and PCIe has a latency set to 0 which I find
very odd (or is that normal for native PCIe based
chipsets?). Also, I'm not even sure PCI latency affects
PCIe devices as they do PCI devices (especially since
the Q35 chipset is natively PCI-Express and all legacy
PCI devices are being a PCIe-to-PCI bridge). Or... am I
just barking up the wrong tree? ;)
I can post a lspci -vvv if that would help. Since the
domU is HVM and it's Windows XP, I've tried using the
"PCI Latency Tool 3.1.2" but the PCIe DVB-S card would
not show in the list of PCI devices. Other testing I've
done, I started off using "ACPI Multiprocessor" as the
HAL, then dropped to the more stable "Standard PC" HAL
(so going back to the ACPI HAL means a re-install of the
OS), and then disabling ACPI and APIC and turning off
PAE... etc. It has no affect.
Anyone have any ideas? Thanks in advance!
--
Daniel Kao
Übermind, Inc.
Seattle, WA, U.S.A.
------------------------------------------------------------------------
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
<mailto:Xen-users@xxxxxxxxxxxxxxxxxxx>
http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
<mailto:Xen-users@xxxxxxxxxxxxxxxxxxx>
http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx <mailto:Xen-users@xxxxxxxxxxxxxxxxxxx>
http://lists.xensource.com/xen-users
--
Stephan Seitz
Senior System Administrator
*netz-haut* e.K.
multimediale kommunikation
zweierweg 22
97074 würzburg
fon: +49 931 2876247
fax: +49 931 2876248
web: www.netz-haut.de <http://www.netz-haut.de>
<http://www.netz-haut.de/>
registriergericht: amtsgericht würzburg, hra 5054
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx <mailto:Xen-users@xxxxxxxxxxxxxxxxxxx>
http://lists.xensource.com/xen-users
------------------------------------------------------------------------
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
--
Stephan Seitz
Senior System Administrator
*netz-haut* e.K.
multimediale kommunikation
zweierweg 22
97074 würzburg
fon: +49 931 2876247
fax: +49 931 2876248
web: www.netz-haut.de <http://www.netz-haut.de/>
registriergericht: amtsgericht würzburg, hra 5054
s_seitz.vcf
Description: Vcard
signature.asc
Description: OpenPGP digital signature
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home