I did a quick and simple test now: 1.) switch on 1st disk (sdd) Jan 19 13:46:18 stein sd 88:0:0:0: [sdd] Attached SCSI disk Jan 19 13:46:18 stein sd 88:0:0:0: Attached scsi generic sg3 type 14 2.) start "dd if=/dev/sdd of=/dev/null" 3.) switch on 2nd disk (sde) Jan 19 13:48:11 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 19 13:48:11 stein sd 88:0:0:0: [sdd] Result: hostbyte=DID_BUS_BUSY driverbyte=DRIVER_OK,SUGGEST_OK Jan 19 13:48:11 stein end_request: I/O error, dev sdd, sector 7871464 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983933 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983934 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983935 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983936 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983937 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983938 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983939 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983940 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983941 Jan 19 13:48:11 stein Buffer I/O error on device sdd, logical block 983942 Jan 19 13:48:11 stein sd 88:0:0:0: [sdd] Result: hostbyte=DID_BUS_BUSY driverbyte=DRIVER_OK,SUGGEST_OK Jan 19 13:48:11 stein end_request: I/O error, dev sdd, sector 7871712 Jan 19 13:48:11 stein sd 88:0:0:0: [sdd] Result: hostbyte=DID_BUS_BUSY driverbyte=DRIVER_OK,SUGGEST_OK Jan 19 13:48:11 stein end_request: I/O error, dev sdd, sector 7871720 Jan 19 13:48:11 stein sd 88:0:0:0: [sdd] Result: hostbyte=DID_BUS_BUSY driverbyte=DRIVER_OK,SUGGEST_OK Jan 19 13:48:11 stein end_request: I/O error, dev sdd, sector 7871464 Jan 19 13:48:11 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:11 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:11 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:12 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:12 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:12 stein firewire_sbp2: failed to reconnect to fw3.0 Jan 19 13:48:12 stein firewire_sbp2: logged in to fw3.0 LUN 0000 (0 retries) Jan 19 13:48:26 stein firewire_sbp2: orb reply timed out, rcode=0x11 Jan 19 13:48:27 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:27 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:27 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:27 stein scsi89 : SBP-2 IEEE-1394 Jan 19 13:48:27 stein firewire_core: created new fw device fw4 (6 config rom retries, S800) Jan 19 13:48:27 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:27 stein firewire_sbp2: logged in to fw4.0 LUN 0000 (0 retries) Jan 19 13:48:27 stein scsi 89:0:0:0: Direct-Access-RBC HDS72404 0KLAT80 KFAO PQ: 0 ANSI: 4 Jan 19 13:48:27 stein sd 89:0:0:0: [sde] 1562845488 512-byte hardware sectors (800177 MB) Jan 19 13:48:27 stein sd 89:0:0:0: [sde] Write Protect is off Jan 19 13:48:27 stein sd 89:0:0:0: [sde] Mode Sense: 11 00 00 00 Jan 19 13:48:27 stein sd 89:0:0:0: [sde] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA Jan 19 13:48:27 stein sde: sde1 Jan 19 13:48:27 stein sd 89:0:0:0: [sde] Attached SCSI disk Jan 19 13:48:27 stein sd 89:0:0:0: Attached scsi generic sg4 type 14 Jan 19 13:48:27 stein firewire_sbp2: error status: 0:9 Jan 19 13:48:27 stein firewire_sbp2: failed to reconnect to fw3.0 Jan 19 13:48:28 stein firewire_sbp2: logged in to fw3.0 LUN 0000 (0 retries) Doing that, dd aborted: dd: reading `/dev/sdd': Input/output error 7871464+0 records in 7871464+0 records out 4030189568 bytes (4.0 GB) copied, 57.6538 s, 69.9 MB/s sdd was however accessible again after the login at 13:48:28. As you see, the SCSI stack did not take the disk offline after the I/O errors. However, the desired behavior is that dd is simply put into I/O wait state rather than failing with errors while the SBP-2 transport is busy. Both disks used in this test are based on the OXFW912 bridge. The "error status: 0:9" before reconnect is typical for them, at least as I have set them up right now (1394b card, 1394b hub, disks on the hub). The status means "request complete, function rejected". It has always been recoverable here when I switched on the 2nd disk with the 1st disk connected but idle. Perhaps we should do it in fw-sbp2 like in sbp2: Put the scsi_device (in sbp2: the Scsi_Host) into blocked state between bus reset event and successful reconnection. We took a while to get this right in sbp2 because blocking the host is prone to deadlocks. Therefore I was so far satisfied with fw-sbp2 not using that scheme. However, the first thing I shall try is to insert a "return SCSI_MLQUEUE_HOST_BUSY" early in sbp2_scsi_queuecommand, depending on a check of the logical unit's generation. A variation of this problem is if the bus reset -- reconnect phase happens while the SCSI stack is just in the middle of INQUIRY or READ CAPACITY. During this, the SCSI stack is very quick to offline a device. We need to do something there, because this situation is not too unusual (e.g. when several FireWire devices are being powered up together). I will proceed to experiment with this as spare time permits. Of course any advice on how to best interact with the SCSI core while the transport is busy would be appreciated. -- Stefan Richter -=====-==--- ---= =--== http://arcgraph.de/sr/ - To unsubscribe from this list: 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
