I'm having this issue on several servers. It could be reproduced with high IO load on xfs md raid5 together with md data-check or resync and wait about 16 hours. Once message "blocked for more than 120 seconds." appears, system gets into weird state. Every access into process tscpd memory block the process (tscpd is IO load maker). For example reading /proc/9772/cmdline blocks, ps command stop working. Access into files opened by tscpd for writing is blocked until the end of weird state. Other files on md are readable without waiting. This state takes about 110 minutes. During this state there was not high IO load. At the end of weird state I started copying kernel source into md. Note that this is first time when hung processes wake up. I'm attached log with hung processes traceback. I do not see this behavior on 2.6.26. Using kernel from linux-image-2.6.32-bpo.5-amd64 2.6.32-23~bpo50+1 Please let me know if you need something else.
Attachment:
kern.log.gz
Description: GNU Zip compressed data