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

Re: [Xen-devel] XEN Panic with VT-d support enabled on Xeon E55xxplatform



Francois,

Thank you for the log.

(XEN) [VT-D]iommu.c:1237:d32767 domain_context_mapping:PCI: bdf = 0:1d.0

Did you see 1 second delay after above line was printed? Or the system got panic immediately following it?

(XEN) DMAR_IQA_REG = 7f79d000
(XEN) DMAR_IQH_REG = 0
(XEN) DMAR_IQT_REG = 20
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 0:
(XEN) queue invalidate wait descriptor was not executed
(XEN) ****************************************
(XEN)
(XEN) Reboot in five seconds...


Isabelle, Francois wrote:
When you only use iommu=no-intremap and see the panic, can you attached >>the 
xen log?
And can you also try iommu=passthrough,no-intremap and attach the log?

See attached files.

VT-d Context Entry is used, the DMA requests related to the Context Entry can 
bypass DMA remapping. This can be faster than DMA remapping.
The BIOS does not have an option for this on my board. But I did mistook it for 
DMA remapping.


And thank you for answering my questions.

François Isabelle




-----Message d'origine-----
De : Cui, Dexuan [mailto:dexuan.cui@xxxxxxxxx]
Envoyé : 7 avril 2009 10:22
À : Isabelle, Francois; Zhao, Yu
Cc : xen-devel@xxxxxxxxxxxxxxxxxxx
Objet : RE: [Xen-devel] XEN Panic with VT-d support enabled on Xeon 
E55xxplatform

iommu=xxx
Dexuan, 'iommu=no-intremap' is not enough to make it run, ' 
iommu=no-qinval,no-intremap' allows booting.
When you only use iommu=no-intremap and see the panic, can you attached the xen 
log?
And can you also try iommu=passthrough,no-intremap and attach the log?

(XEN) [VT-D]dmar.c:485: Host address width 39
Shouldn't this be 38 ?
The Host Address Width (HAW) of the platform is computed as (N+1), where N is 
the value reported in the Host Address Width field of DMA Remapping Reporting 
Structure.
Please see acpi_parse_dmar() or VT-d spec for details.

(XEN) Intel VT-d Snoop Control supported.
(XEN) Intel VT-d DMA Passthrough not supported.
Shouldn't that be enabled to support DMA from domU?
If the "DMA Passthrough" (please don't mistake it for DMA remapping:-) of VT-d 
Context Entry is used, the DMA requests related to the Context Entry can bypass DMA 
remapping. This can be faster than DMA remapping. If you host supports DMA passthrough 
(you could check your BIOS setup menu), in Xen we can enable DMA passthrough by 
iommu=passthrough (Note, even with this parameter, we only enable passthrough for Dom0 
considering secrity)


(XEN) HPET broadcast init failed, turn to PIT broadcast.
Does that mean the HPET is not functional?
From hpet_fsb_cap_lookup(), looks your HPET timers may work, but they don't 
support FSB routing, so won't be used in power management code.

Can we expect a good level of performance with interrupt remapping disabled? 
Will a domU be able to load a 82576 driver and use the NIC at all?
I think we should have almost the same performance with Interrupt Remapping 
enabled or disabled.
82576 should work in DomU.

Thanks,
-- Dexuan



-----Original Message-----
From: Isabelle, Francois [mailto:Francois.Isabelle@xxxxxxxxxxxxxx]
Sent: Tuesday, April 07, 2009 9:12 PM
To: Cui, Dexuan; Zhao, Yu
Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: [Xen-devel] XEN Panic with VT-d support enabled on Xeon 
E55xxplatform

Hi.

Here is what you both asked for:

Do you have this problem all the times? Can you please grab some logs
and send them to me?

Yes Yu, all the times. I will package the logs as attachment to this post.

iommu=xxx
Dexuan, 'iommu=no-intremap' is not enough to make it run, ' 
iommu=no-qinval,no-intremap' allows booting.

Here are some parts of the log I don't understand...

(XEN) [VT-D]dmar.c:485: Host address width 39
 Shouldn't this be 38 ?

(XEN) Intel VT-d Snoop Control supported.
(XEN) Intel VT-d DMA Passthrough not supported.
 Shouldn't that be enabled to support DMA from domU?

(XEN) Intel VT-d Queued Invalidation supported.
(XEN) Intel VT-d Interrupt Remapping supported.

(XEN) HPET broadcast init failed, turn to PIT broadcast.
 Does that mean the HPET is not functional?

Can we expect a good level of performance with interrupt remapping disabled? 
Will a domU be able to load a 82576 driver and use the NIC at all?

Thank you both for your replies.

François Isabelle

-----Message d'origine-----
De : Cui, Dexuan [mailto:dexuan.cui@xxxxxxxxx]
Envoyé : 7 avril 2009 03:05
À : Zhao, Yu; Isabelle, Francois
Cc : xen-devel@xxxxxxxxxxxxxxxxxxx
Objet : RE: [Xen-devel] XEN Panic with VT-d support enabled on Xeon 
E55xxplatform

(XEN) DMAR_IQA_REG = 7f79d000
(XEN) DMAR_IQH_REG = 0
(XEN) DMAR_IQT_REG = 20
Here  the IQT is such a big number 0x20... this seems pretty strange.
Oh, actually this 0x20 is normal -- actually QT = IQT_REG >> 4 accoriding the 
format of IQT_REG in VT-d spec. So QT = 0x2 -- this is ok.
So looks the panic somehow happens on the first invalidation: enable_intremap() 
-> iommu_flush_iec_global(). I expect your host can boot fine with 
iommu=no-intremap -- if this still doesn't work, could you please try 
iommu=no-qinval,no-intremap ?

Thanks,
-- Dexuan



-----Original Message-----
From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx 
[mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Cui, Dexuan
Sent: Tuesday, April 07, 2009 12:40 PM
To: Zhao, Yu; Isabelle, Francois
Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: [Xen-devel] XEN Panic with VT-d support enabled on Xeon E55xx 
platform

(XEN) DMAR_IQA_REG = 7f79d000
(XEN) DMAR_IQH_REG = 0
(XEN) DMAR_IQT_REG = 20
Here  the IQT is such a big number 0x20... this seems pretty strange.

Hi François,
And, could you please try the xen parameter "iommu=no-intremap" ? -- If you still see the 
panic, please  also add the parameter "nosmp" to see if  there would any change.
We don't have the same host at hand. :-(  So we need try these on your host.
BTW, can you boot the same host using the latest native linux-2.6.29-rc8 when 
you iommu_intel=on and interrupt remapping is enabled?

PS, what's your BIOS version and date -- could you try to find the latest BIOS 
for the host? I personally tend to think this is a BIOS issue.

Thanks,
-- Dexuan



-----Original Message-----
From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx 
[mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Zhao, Yu
Sent: Tuesday, April 07, 2009 10:31 AM
To: Isabelle, Francois
Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-devel] XEN Panic with VT-d support enabled on Xeon E55xx 
platform

Hi Francois,

Do you have this problem all the times? Can you please grab some logs
and send them to me? The dmesg from a native kernel if you can't boot up
any version of Xen, and the whole Xen hypervisor log would help us to
figure out the problem.

Thanks,
Yu


Isabelle, Francois wrote:
Hi all,

I'm working with a board on which XEN panics with the following message (see 
below). From what I can see on the lists, most of the problems on these 
platforms seem to be related to broken DMAR tables and RMRR, but this platform 
passes the VT-d firmware toolkit (http://edc.intel.com/Platforms/Xeon-5500/) 
successfully so I'm tented to blame XEN for it... should I file a bug entry?

>From what I can see in the 'queue invalidation' code, a timeout is reached. As 
anyone seen this on a similar platform?


(XEN) Xen version 3.4-unstable (root@localdomain) (gcc version 4.1.2 20080704 
(Red Hat 4.1.2-44)) Fri Apr  3 14:20:00 EDT 2009
(XEN) Latest ChangeSet: Mon Mar 30 16:48:26 2009 +0100 19433:d5ddc782bc49
(XEN) Command line: console=com1 com1=115200,8n1  iommu=1 loglvl=all 
loglvl_guest=all ro

..
(XEN) Enabling APIC mode:  Flat.  Using 1 I/O APICs
(XEN) ACPI: HPET id: 0xffffffff base: 0xfed00000
(XEN) [VT-D]dmar.c:485: Host address width 39
(XEN) [VT-D]dmar.c:494: found ACPI_DMAR_DRHD
(XEN) [VT-D]dmar.c:349: dmaru->address = fbffe000
(XEN) [VT-D]dmar.c:306: found IOAPIC: bdf = f0:1f.7
(XEN) [VT-D]dmar.c:358: found INCLUDE_ALL
(XEN) [VT-D]dmar.c:498: found ACPI_DMAR_RMRR
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.0
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.1
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.2
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.7
(XEN) [VT-D]dmar.c:498: found ACPI_DMAR_RMRR
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.0
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.1
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.2
(XEN) [VT-D]dmar.c:300: found endpoint: bdf = 0:1d.7
(XEN) [VT-D]dmar.c:502: found ACPI_DMAR_ATSR
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:3.0  sec = 5  sub = 5
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:4.0  sec = 6  sub = e
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:5.0  sec = 3  sub = 3
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:7.0  sec = 18  sub = 18
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:8.0  sec = f  sub = 17
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:9.0  sec = 4  sub = 4
(XEN) [VT-D]dmar.c:287: found bridge: bdf = 0:a.0  sec = 2  sub = 2
..
(XEN) Intel VT-d Snoop Control supported.
(XEN) Intel VT-d DMA Passthrough not supported.
(XEN) Intel VT-d Queued Invalidation supported.
(XEN) Intel VT-d Interrupt Remapping supported.
(XEN) DMAR_IQA_REG = 7f79d000
(XEN) DMAR_IQH_REG = 0
(XEN) DMAR_IQT_REG = 20
(XEN)
(XEN) ****************************************
(XEN) Panic on CPU 0:
(XEN) queue invalidate wait descriptor was not executed
(XEN) ****************************************
(XEN)
(XEN) Reboot in five seconds...

François Isabelle

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

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