Bug in aic94xx driver in 2.6.25-rc3
A message that is not in the archive (sent off-list or not yet received): 1203987834.6909.17.camel@gnattop
A message that is not in the archive (sent off-list or not yet received): 1203992592.26232.79.camel@alexis
My original post was user error and not acutally a bug. I didn't realize that there was another patch I need to apply to rc3 to get the latest scsi drivers and error handler code. Alexis clued me in. Now the error handler appears to be working properly. I included a sample at the bottom of this email.
I am still seeing the disk go offline if I run i/o performance tests on sas disks connected to the aic94xx (sequencer version 32). It doesn't happen right away. The i/o tests will run for several hours before it fails. Eventually you see the filesystem abort and then be remounted as read only.
Peter Bogdanovic
IBM
sas: command 0xffff810142569200, task 0xffff81022ad27980, timed out:
EH_NOT_HANDLED^M
sas: Enter sas_scsi_recover_host^M
sas: trying to find task 0xffff81022ad27980^M
sas: sas_scsi_find_task: aborting task 0xffff81022ad27980^M
aic94xx: tmf tasklet complete^M
aic94xx: tmf resp tasklet^M
aic94xx: tmf came back^M
aic94xx: task not done, clearing nexus^M
aic94xx: asd_clear_nexus_tag: PRE^M
aic94xx: asd_clear_nexus_tag: POST^M
aic94xx: asd_clear_nexus_tag: clear nexus posted, waiting...^M
aic94xx: task 0xffff81022ad27980 done with opcode 0x23 resp 0x0 stat 0x8d but aborted by upper layer!^M
aic94xx: asd_clear_nexus_tasklet_complete: here^M
aic94xx: asd_clear_nexus_tasklet_complete: opcode: 0x0^M
aic94xx: came back from clear nexus^M
aic94xx: task 0xffff81022ad27980 aborted, res: 0x0^M
sas: sas_scsi_find_task: task 0xffff81022ad27980 is done^M
sas: sas_eh_handle_sas_errors: task 0xffff81022ad27980 is done^M sd 0:0:3:0: [sdd] Result: hostbyte=DID_ABORT driverbyte=DRIVER_OK,SUGGEST_OK^Mend_request: I/O error, dev sdd, sector 27106623^M
sas: --- Exit sas_scsi_recover_host^M
James Bottomley
maintainerQuoted 11 lines · peter
My original post was user error and not acutally a bug. I didn't realize that there was another patch I need to apply to rc3 to get the latest scsi drivers and error handler code. Alexis clued me in. Now the error handler appears to be working properly. I included a sample at the bottom of this email.
I am still seeing the disk go offline if I run i/o performance tests on sas disks connected to the aic94xx (sequencer version 32). It doesn't happen right away. The i/o tests will run for several hours before it fails. Eventually you see the filesystem abort and then be remounted as read only.
Yes, I've seen this one too. in my case it's caused by error handling tripping a flutter on the disk phy, so the HBA actually sees an unplug/replug event, but that causes the disk to go offline (even though it actually reappears almost immediately). I have that on my list of things to fix ... probably by stealing the devloss tmo from fc.
James