Bug#606968: DMA: Out of SW-IOMMU space with Xen kernel during heavy I/O

[Rik Theys <Rik.Theys@esat.kuleuven.be>]

On Tue, 14 Dec 2010, Ben Hutchings wrote:

> On Tue, 2010-12-14 at 13:24 +0100, Rik Theys wrote:
> > This system _has_
> > a RAID controller and the disk is in a RAID1. Why would the filesystem switch
> > to read-only mode? Is it because the RAID is too slow? If the RAID is too slow,
> > this should just give a high iowait, but not make the file system read-only?
>
> The DMA mapping failure will be treated as an I/O error.  Most
> filesystems do not attempt to retry after an error.  This has nothing to
> do with the speed of the disk.
>
> Since you use RAID, the failure may only put a single disk in degraded
> state, which will then be recoverable in the usual way.  The error would
> be hidden from the filesystem.

The RAID in this system is a PERC H200, which should be a hardware RAID card.
Can the mpt2sas driver see the physical disks that are part of the RAID?

It is my understanding that the mpt2sas driver only sees the RAID as a scsi disk
and not the individual drives? If that is the case, how would this I/O error
lead to a single disk in the RAID becoming faulty and the RAID degraded?

Will the H200 mark the disk as faulty when the mpt2sas driver triggers this?

Regards,

Rik