[DRBD-user] Invalid SGL issues with mpt driver

[Kilian CAVALOTTI]

Hi all,

I know that's perhaps not the right place to ask, but the weird problem I 
encounter really seems to be DRBD related.

I use an external disk array, attached to a SCSI HBA (LSI Logic U320, using 
mpt 2.6.16 kernel driver). This enclosure exports a large SCSI volume 
(/dev/sda), used as a DRBD device. The problem is that, after some time 
synchronizing, I get the following errors :

mptbase: ioc0: IOCStatus(0x0003): Invalid SGL
mptbase: ioc0: LogInfo(0x11070000): F/W: DMA Error
mptbase: ioc0: LogInfo(0x11070000): F/W: DMA Error
mptbase: ioc0: IOCStatus(0x004b): SCSI IOC Terminated

At first sight, it seems to be a SCSI problem, due to the controller, or the 
device. But the fact is that these messages only appear when the device is 
used as /dev/drbd0. I tried to directly mount /dev/sda and perform intensive 
I/O on it (bonnie++, dd), and I also tried to create software RAID1 volumes 
(using mdadm) with individual disks of the array (to be sure that all sectors 
will be written), and I never got any error. So I think I could say that the 
hardware is not in cause.

When the problem occur, I get no DRBD error, only this "Invalid SGL", and the 
syncing continue. 

Here my drbd.conf relevant part (the problematic host is 'achab'):
---------------------------------------------------------------------
resource r0 {
  protocol C;
  incon-degr-cmd "echo '!DRBD! pri on incon-degr' | wall; sleep 60; halt -f";
  startup {
  }
  disk {
    on-io-error   detach;
  }
  net {
  }
  syncer {
    rate 500M;
    group 1;
    al-extents 257;
  }
  on achab {
    device     /dev/drbd0;
    disk       /dev/sda;
    address    10.0.0.1:7788;
    meta-disk  internal;
  }
  on moby {
    device    /dev/drbd0;
    disk      /dev/sda;
    address   10.0.0.2:7788;
    meta-disk internal;
  }
}
---------------------------------------------------------------------
The DBRD module is version v0.7.18, provided by debian source package.

Is this a know problem? Does somebody else observe the same behavior? Is there 
a way to avoid these errors?

Thanks in advance for any hint,
Regards,
-- 
Kilian CAVALOTTI                      Administrateur réseaux et systèmes
UPMC / CNRS - LIP6 (C870)
8, rue du Capitaine Scott                          Tel. : 01 44 27 88 54
75015 Paris - France                               Fax. : 01 44 27 70 00