Latest proxmox (2.6.24-9), freebsd 7.2 amd64 vm and ad0: FAILURE WRITE_DMA48

[davidedamico]

Hi,
I am using Proxmox on a Dell R710 with 16gb ram, EE5530 and a perc6/i controller with 3 raid1 1tb sata disks (one per VM):
pxx1:~# uname -a
Linux pxx1 2.6.24-9-pve #1 SMP PREEMPT Tue Nov 17 09:34:41 CET 2009 x86_64 GNU/Linux

and I have a FreeBSD 7.2-p4 amd64.
Very often the VM stops responding (I received network error trying to attach to vnc console) and looking for errors in freebsd /var/log/messages I found:

Nov 29 18:09:26 vm2 kernel: ad0: TIMEOUT - WRITE_DMA48 retrying (1 retry left) LBA=1302967215
Nov 29 18:09:31 vm2 kernel: ad0: TIMEOUT - WRITE_DMA48 retrying (0 retries left) LBA=1302967215
Nov 29 18:09:31 vm2 kernel: ad0: FAILURE - WRITE_DMA48 status=41<READY,ERROR> error=4<ABORTED> LBA=1302967215

How could I solve this issue? Recreating a VM on a blank .qcow2 file?

Here is my freebsd conf file:

pxx1:/etc/qemu-server# more 102.conf
name: Elbalink_Web
sockets: 1
vlan0: rtl8139=2E:73:49:F8:99:14
bootdisk: ide0
ide0: vm-102-disk.qcow2
ostype: other
memory: 4096
onboot: 1
description: IP: 172.16.7.246
boot: c
freeze: 0
cpuunits: 1000
acpi: 1
kvm: 1
pxx1:/etc/qemu-server#


Thanks in advance,
d.

 

[ParanoiK]

Unfortunately fbsd is not best OS for virtualization. I know, that apic sometimes causes problems with clock and stability. First of all try adding hint.apic.0.disabled="1" to /boot/loader.conf and restarting the system.

BTW: what is the time of executing command "sleep 1"? ;-)

 

[davidedamico]

Time of sleep 1 command:

dave@vm2:~> time sleep 1
0.000u 0.000s 0:01.00 0.0% 0+0k 0+0io 0pf+0w


Thanks for the hint, I'll let you know if disabling apic will solve the problem (I will remove device apic from kernel config file, too)

Thanks,
d.

 

[davidedamico]

Disabling apic, freebsd refuses to start

d.

 

[ParanoiK]

That's strange, it souldn't happen. Are you sure the line in /boot/loader.conf is correct, without any special characters, etc?

 

[davidedamico]

Hi,
sure it happened.
It happened the same than an old post:

showthread.php?t=1995

but it wasn't resolved.

Is it something related to proxmox config or freebsd config?

Thanks in advance,
d.

 

[ParanoiK]

Any error msgs, logs?

I've got 4 fbsd (7.2, 8.0) machines running with no problems. Have you changed kernel config?

 

[davidedamico]

Yes, I have a customized kernel config.
Do you use the GENERIC one?

thanks,
d.

 

[ParanoiK]

In that moment yes, there was no point in using customized (or in fact I'm too lazy to make new one while everything works fine ;-)). Also I'm trying to find some info about anyone trying to implement VIRTIO drivers.

 

[davidedamico]

Using GENERIC without hint.apic.0.disabled="1" the VM remained up 4 hours (with the custom kernel it remained up 2hours or less).
Using hint.apic.0.disabled="1" it refused to boot.
What you config file?

It's strange because the VM stayed up from June without any problem, and in October/November it started to have problems.

Ideas?

Thanks in advance,
d.

 

[ParanoiK]

What config? VM config or fbsd kernel config?

Have you done any updates then?

 

[davidedamico]

4 hours is not my idea of server reliability

I intended your qemu/kvm configuration.

I am using freebsd 7.2-p4 amd64 with GENERIC kernel and with 2 jails inside.

Thanks,
d.

 

[ParanoiK]

Hmm... the configuration is straight-forward, nothing complex.

ostype: other
memory: 512
sockets: 1
name: SERWER
ide2: none,media=cdrom
vlan0: e1000=BE:D8:3A:62:EB:A4
bootdisk: scsi0
scsi0: local:104/vm-104-disk-1.raw
onboot: 1
cores: 1
scsi9: local:104/vm-104-disk-2.raw

One thing came in mind: all my guests are 32 bit - maybe that's the case.

 

[davidedamico]

There are two differences:
- my guest is 64bit
- i use ide disk (raw format).

I'll try to migrate it to 8.0 i386

 

[ParanoiK]

Ide disk shouldn't be a problem, I use it on FreeNAS

ostype: other
memory: 1024
sockets: 1
name: FreeNAS
bootdisk: scsi0
scsi0: local:101/vm-101-disk-1.raw
boot: c
freeze: 0
cpuunits: 8000
acpi: 1
kvm: 1
ide0: /dev/disk/by-id/scsi-SATA_SAMSUNG_HD753LJS13UJ1LS800949,snapshot=off,cache=none
ide2: /dev/disk/by-id/ata-WDC_WD1200JB-00CRA0_WD-WMA8C1879787,snapshot=off,cache=none
vlan0: e1000=5E:26:83:05:96:BB
onboot: 1
cores: 1
vlan2: e1000=6A:81:71:B0:F0:59

Actually I think that ide is faster on fbsd than scsi.

 

[degrootm]

I have had a similar issue where FreeBSD seems to run for some time, approximately 6 hours in my case and then it just stops.
After I found this thread I tried disabling ACPI. Just to eliminate the issue. That didn't change anything.
So for full disclosure purposes, I'm not the exact same configuration as above.
I'm running the 9.0 Release of FreeBSD, IDE disks. So nothing major in what I've written so far, but on a hunch I reduced the number of processors to 1 and the KVM has been rock solid for over 48 hours, after a week of up and down.
I'm now trying to sort out if this is something on the KVM(ProxMox) side or if it's a FreeBSD issue. I'm sure that will take some investigating.

If anyone has seen this before please chime in.

I had a similar number of processors issue with a container(not KVM) but in the opposite situation. In that case I needed to have a minimum of two processors to get the instance to even boot. That was moving from ProxMox 1.8 to 2.0. So logic would say I need a better understanding of how the processor virtualization is implemented in ProxMox before I turn to the FreeBSD implementation. They have also had multi-processor support in FreeBSD since 2005.