ata3.00: failed command: WRITE DMA

[0badc0de]

Hi all.
I am running a few VMs under proxmox 5.2-2.
I find from time to time a very puzzling set of log messages inside the VMs:

Jun 20 05:07:09 tfg kernel: [497807.524612] ata3.00: exception Emask 0x0 SAct 0x0 SErr 0x0 action 0x6 frozen 
Jun 20 05:07:09 tfg kernel: [497807.524651] ata3.00: failed command: WRITE DMA 
Jun 20 05:07:09 tfg kernel: [497807.524677] ata3.00: cmd ca/00:08:38:b8:f9/00:00:00:00:00/ec tag 27 dma 4096 out 
Jun 20 05:07:09 tfg kernel: [497807.524677]          res 40/00:00:00:00:00/00:00:00:00:00/00 Emask 0x4 (timeout) 
Jun 20 05:07:09 tfg kernel: [497807.524746] ata3.00: status: { DRDY } 
Jun 20 05:07:09 tfg kernel: [497807.524771] ata3: hard resetting link 
Jun 20 05:07:29 tfg kernel: [497827.720611] ata3: SATA link up 1.5 Gbps (SStatus 113 SControl 300) 
Jun 20 05:07:29 tfg kernel: [497827.720854] ata3.00: configured for UDMA/100 
Jun 20 05:07:29 tfg kernel: [497827.720859] ata3.00: device reported invalid CHS sector 0 
Jun 20 05:07:29 tfg kernel: [497827.720865] ata3: EH complete

These lines can appear from a few times a day to a few times a hour.
I am also encountering some virtual disk image inconsistencies where the initial part of the image gets corrupted and made un-bootable.
If I restore the partition table I can mount the filesystem, but cannot boot it.
I cannot prove the two things are related, but I "feel" it's so.
As the disk is virtual, I am really puzzled by such events.

What I'd like to know is:

1. Should I worry about those events? ( I would say so).
2. As the disk is virtual, is this a bug in qemu-kvm?
3. Can this type of event be the cause of the disk inconsistencies?
4. Anyone here with the same issues?
5. Can this thing be "fixed" by me?

If you need more details, please ask.

 

[wolfgang]

Hi,

please send you VM Config and tell us what storage type do you use?

 

[0badc0de]

I hope I've got it right:

agent: 1
bootdisk: sata0
cores: 4
ide2: none,media=cdrom
memory: 8196
name: TFG-GATE
net0: e1000=6c:3b:e5:b2:fd:e8,bridge=vmbr0,link_down=1
net1: e1000=4A:B1:11:D3:BA:E6,bridge=vmbr0
numa: 0
onboot: 1
ostype: l26
sata0: SANMSA:vm-154-disk-1,discard=on,size=200G
scsihw: virtio-scsi-pci
smbios1: uuid=06208f32-a9fc-4659-9d14-230ae2595793
sockets: 1

The VM disk storage is a 200 GB logical volume (of course) and it's divided into a small swap partition and a single XFS file system.
The LV itself is hosted inside an HP MSA SAN.
There's no hardware warning or error into the host logs.
The errors pop up in the guests only.
I don't know whether these infos can be enough.

 

[0badc0de]

I have found something similar to my issue on
https://forum.proxmox.com/threads/error-on-ubuntu-server-16-04.34317/

But there's somehting that's puzzling me.

It looks like the kernel from Ubuntu Server 16.04 doesn't understand well the SATA interface as virtualized by ProxMox/QEmu, which would be rather weird. Wouldn't it?

Of course, if I remove the one and only SATA interface and replace it with SCSI, I won't get any error any more from the SATA interface. Thus I would hide the problem under the carpet, especially if the issue isn't actually within the Ubuntu kernel.

 

[wolfgang]

Sata and IDE are only for compatibility reasons available
and should only used if virtio or SCSI with virtio bus is not supported.
All modern OS has support for virtio.

 

[0badc0de]

Thanks for the answer.
I see it was shown in the original configuration, though.
So it's not an issue with Ubuntu, but rather with the qemu implementation of SATA and IDE.
Isn't it?

 

[wolfgang]

Yes, there are known problems with IDE and SATA in qemu.

 

[0badc0de]

Thanks a lot! That's good to know and should be clearly stated somewhere in the wiki.
Vielen Dank!

 

[0badc0de]

HI.
It's not clear to me whether this configuration line:

> scsihw: virtio-scsi-pci

is about the SATA0 device or not.
If so, the SATA0 should be already using the virtio device class.
If not, why it's there?

 

[wolfgang]

No this is the SCSI controller.

As a real HW controller card, you can have different kinds of SCSI controllers.
If you do not use SCSI as disk bus this setting has no effect.

 

[0badc0de]

Oh, I see. Thanks.
So that line is pretty useless as I have a single drive and it's SATA0. Isn't it?

 

[wolfgang]

Yes you can ignore it.