On 03/12/2014 11:05 AM, Moritz Grimm wrote: >> On a Shuttle DS47, with a recent Samsung SSD 840, significant I/O churn >> (cvs checkout on top of existing repository, while removing another, and >> some more) appears to have caused and/or triggered intermittent I/O errors. >> >> In dmesg, the error shows up as this: >> ahci0: attempting to idle device >> ahci0: couldn't recover NCQ error, failing all outstanding commands. > [...] >> sd0 at scsibus0 targ 0 lun 0: <ATA, Samsung SSD 840, DXM0> SCSI3 >> 0/direct fixed naa.5002538550409483 >> sd0: 122104MB, 512 bytes/sector, 250069680 sectors, thin > > The exact model of this disk is MZ-7PD128 (SSD 840 Pro). Off-list, I was > asked to check, if the disk runs the latest firmware.
A new firmware came out just around the time of that email, and I upgraded (DXM06B0Q). No change in behaviour, and the NCQ/IO error is very easy for me to reproduce on the Shuttle. Next test I'll try is put the disk and the OpenBSD on it into an older laptop. If I can't reproduce it there, I assume that it remains probable that it is a Shuttle hardware and/or ahci driver issue. Otherwise, I assume the issue to be tied to the disk itself. Further testing ideas are very welcome. Moritz
