On Sun, 12 Mar 2006, Maxim Kozover wrote: > Congratulations! The kernel from scsi-rc-fixes git and your patch are > working. Actually Mike R. and James S. deserve the credit for the composite patch which consists of: 1) [PATCH] FC transport : Avoid device offline cases by stalling aborts until device unblocked http://marc.theaimsgroup.com/?l=linux-scsi&m=114225658724378&w=2 2) Serialize scan work during fc_remote_port_delete() so rport removal doesn't deadlock midlayer scans. The problem you were seeing. (Mike R.) 3) rport race fixes during removal (James S.). > By the way, could you, please, tell me how I get only scsi patches > from the git repository, cause I got the whole kernel by using > cg-clone http://kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6.git > > Now the process looks like following: > Mar 11 23:54:22 multipath kernel: qla2xxx 0000:03:01.0: LOOP DOWN detected (2). > Mar 11 23:54:28 multipath kernel: rport-4:0-0: blocked FC remote port time out: > removing target and saving binding > Mar 11 23:54:37 multipath kernel: qla2xxx 0000:03:01.0: LIP reset occured (f7f7). > Mar 11 23:54:37 multipath kernel: qla2xxx 0000:03:01.0: LOOP UP detected (2 Gbps). > Mar 11 23:54:59 multipath kernel: 4:0:0:0: timing out command, waited 22s > > And the disks appear. > Could you tell me, please, where this 22sec timeout came from? Essentially there's currently several issues with rport consumers making delete() calls during mid-layer scanning. I'm hoping at a minimum we can get Mike R's fixes into 2.6.16, and address the additional races going forward... James/Mike? Here's a minimal the serialize scan-work patch, could you check to see that this addresses your issue? Start with any latest linux-2.6.git tree. Thanks, Andrew --- diff --git a/drivers/scsi/scsi_transport_fc.c b/drivers/scsi/scsi_transport_fc.c index 929032e..3d09920 100644 --- a/drivers/scsi/scsi_transport_fc.c +++ b/drivers/scsi/scsi_transport_fc.c @@ -1649,6 +1649,8 @@ fc_remote_port_delete(struct fc_rport * return; } + /* flush any scan work */ /* which can sleep */ + scsi_flush_work(rport_to_shost(rport)); scsi_target_block(&rport->dev); /* cap the length the devices can be blocked until they are deleted */ - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
