This driver is for Compaq's SMART Array Controllers. Supported Cards: ---------------- This driver is known to work with the following cards: * SA 5300 * SA 5i * SA 532 * SA 5312 * SA 641 * SA 642 * SA 6400 * SA 6400 U320 Expansion Module * SA 6i * SA 6422 * SA P600 * SA P400 * SA P400i * SA E200 * SA E200i If nodes are not already created in the /dev/cciss directory # mkdev.cciss [ctlrs] Where ctlrs is the number of controllers you have (defaults to 1 if not specified). Device Naming: -------------- You need some entries in /dev for the cciss device. The mkdev.cciss script can make device nodes for you automatically. Currently the device setup is as follows: Major numbers: 104 cciss0 105 cciss1 106 cciss2 etc... Minor numbers: b7 b6 b5 b4 b3 b2 b1 b0 |----+----| |----+----| | | | +-------- Partition ID (0=wholedev, 1-15 partition) | +-------------------- Logical Volume number The suggested device naming scheme is: /dev/cciss/c0d0 Controller 0, disk 0, whole device /dev/cciss/c0d0p1 Controller 0, disk 0, partition 1 /dev/cciss/c0d0p2 Controller 0, disk 0, partition 2 /dev/cciss/c0d0p3 Controller 0, disk 0, partition 3 /dev/cciss/c1d1 Controller 1, disk 1, whole device /dev/cciss/c1d1p1 Controller 1, disk 1, partition 1 /dev/cciss/c1d1p2 Controller 1, disk 1, partition 2 /dev/cciss/c1d1p3 Controller 1, disk 1, partition 3 Support for more than 8 controllers ----------------------------------- Originally the driver only supports 8 controllers in the system, and this is due to the major numbers assigned to the driver (104 thru 111). The driver can now support up to 32 controllers in the system. For the ninth controller and beyond, the major numbers will be assigned dynamically by the kernel when it is discovered, and there is no guarantee what the major number you will get, but most likely it will start from 254 and goes down from there. You can check the assigned major numbers by typing cat /proc/devices And look for cciss controllers Once you have this, you need to create device nodes in /dev/cciss directory. The nodes for the first 8 controllers should already be created by mkdev.cciss script or /etc/makedev.d/cciss script. You can add the major number(s) in those scripts, or create the nodes manually by using the mknod command. You can also use mknod_dyn.cciss and rmnod_dyn.cciss scripts to create or remove nodes easily. These scripts can be found in the Documentation directory. Then you can mount the devices and create partitions (You also need to make nodes for these partitions). As for the minor number, the disk device will have a minor number divisible by 16 (e.g: 0, 16, 32 etc), and the partitions on those disk devices will have the minor number of the disk device plus the partition number (1-15). For example, disk d2 will have minor number 32, and its partitions 1 and 2 will have minor numbers 33 and 34. Look at the mkdev.cciss script for example. Note: In 2.4 kernel, partition names are hard coded in /usr/src/linux/fs/partitions/check.c and only for the first 8 cciss controllers. The rest will be reported as ccissXX. This should not affect I/O operation or performance. Please apply the cciss_2.4_partition_name.patch to address this. This will not be an issue under 2.5 kernel, since partition names will be handled by the driver. SCSI tape drive and medium changer support ------------------------------------------ SCSI sequential access devices and medium changer devices are supported and appropriate device nodes are automatically created. (e.g. /dev/st0, /dev/st1, etc. See the "st" man page for more details.) You must enable "SCSI tape drive support for Smart Array 5xxx" and "SCSI support" in your kernel configuration to be able to use SCSI tape drives with your Smart Array 5xxx controller. Additionally, note that the driver will not engage the SCSI core at init time. The driver must be directed to dynamically engage the SCSI core via the /proc filesystem entry which the "block" side of the driver creates as /proc/driver/cciss/cciss* at runtime. This is because at driver init time, the SCSI core may not yet be initialized (because the driver is a block driver) and attempting to register it with the SCSI core in such a case would cause a hang. This is best done via an initialization script (typically in /etc/init.d, but could vary depending on distibution). For example: for x in /proc/driver/cciss/cciss[0-9]* do echo "engage scsi" > $x done Once the SCSI core is engaged by the driver, it cannot be disengaged (except by unloading the driver, if it happens to be linked as a module.) Note also that if no sequential access devices or medium changers are detected, the SCSI core will not be engaged by the action of the above script. Hot plug support for SCSI tape drives ------------------------------------- Hot plugging of SCSI tape drives is supported, with some caveats. The cciss driver must be informed that changes to the SCSI bus have been made, in addition to and prior to informing the SCSI mid layer. This may be done via the /proc filesystem. For example: echo "rescan" > /proc/scsi/cciss0/1 This causes the adapter to query the adapter about changes to the physical SCSI buses and/or fibre channel arbitrated loop and the driver to make note of any new or removed sequential access devices or medium changers. The driver will output messages indicating what devices have been added or removed and the controller, bus, target and lun used to address the device. Once this is done, the SCSI mid layer can be informed of changes to the virtual SCSI bus which the driver presents to it in the usual way. For example: echo scsi add-single-device 3 2 1 0 > /proc/scsi/scsi to add a device on controller 3, bus 2, target 1, lun 0. Note that the driver makes an effort to preserve the devices positions in the virtual SCSI bus, so if you are only moving tape drives around on the same adapter and not adding or removing tape drives from the adapter, informing the SCSI mid layer may not be necessary. Note that the naming convention of the /proc filesystem entries contains a number in addition to the driver name. (E.g. "cciss0" instead of just "cciss" which you might expect.) This is because of changes to the 2.4 kernel PCI interface related to PCI hot plug that imply the driver must register with the SCSI mid layer once per adapter instance rather than once per driver. Note: ONLY sequential access devices and medium changers are presented as SCSI devices to the SCSI mid layer by the cciss driver. Specifically, physical SCSI disk drives are NOT presented to the SCSI mid layer. The physical SCSI disk drives are controlled directly by the array controller hardware and it is important to prevent the OS from attempting to directly access these devices too, as if the array controller were merely a SCSI controller in the same way that we are allowing it to access SCSI tape drives. Monitor Threads --------------- For multipath configurations (acheived via a higher level driver, such as the "md" driver) it is important that failure of a controller is detected. Ordinarily, the driver is entirely interrupt driven. If a failure occurs in such a way that the processor cannot receive interrupts from an adapter, the driver will wait forever for i/o's to complete. In a multipath configuration this is undesirable, as the md driver relies on i/o's being reported as failed by the low level driver to trigger failing over to an alternate controller. The monitor threads allow the driver to detect such situations and report outstanding i/o's as having failed so that recovery actions such switching to an alternate controller can occur. The monitor threads periodically sends a trivial "no-operation" command down to the controllers and expect them to complete within a a reasonable (short) time period. The firmware on the adapter is designed such that no matter how busy the adapter is serving i/o, it can respond quickly to a "no-operation" command. In the event that a deadline elapses before a no operation command completes, all outstanding commands on that controller are reported back to the upper layers as having failed, and any new commands sent to the controller are immediately reported back as failed. To enable the monitor threads, the compile time option must be enabled (via the usual linux kernel configuration) and the monitor thread must be enabled at runtime as well. A system may have many adapters, but perhaps only a single pair operating in a multipath configuration. In this way, it is possible to run monitoring threads only for those adapters which require it. To start a monitoring thread on the first cciss adapter, "cciss0" with a polling interval of 30 seconds, execute the following command: echo "monitor 30" > /proc/driver/cciss/cciss0 To change the polling interval, to say, 60 seconds: echo "monitor 60" > /proc/driver/cciss/cciss0 (Note, the change will not take effect until the previous polling interval elapses.) To disable the monitoring thread, set the polling interval to 0 seconds: echo "monitor 0" > /proc/driver/cciss/cciss0 (Again, the monitoring thread will not exit until the previous polling interval elapses.) The minimum monitoring period is 10 seconds, and the maximum monitoring period is 3600 seconds (1 hour). The no-operation command must complete with 5 seconds of submission in all cases or the controller will be presumed failed.