Description of problem: When running a RHEL4 install on a machine with a HostRAID configuration + adp94xx driver, the install fails because the adp94xx driver needs a module parameter to allow the HostRAID setup. Version-Release number of selected component (if applicable): RHEL4 U4 (and probably earlier) How reproducible: 100% Steps to Reproduce: 1. Run RHEL4.4 installation on system described above 2. 3. Actual results: device probe aborts because it requires a module parameter to be set. Expected results: adp94xx should load and installation should succeed. Additional info: The adp94xx module seems to work fine for Stratus in the HostRAID config. We are not sure why there is a module parameter (asd_attach_Host_RAID), which when not set causes the device probe to abort. Also, the RHEL5 replacement to this driver (aic94xx) does not have a similar parameter. The best answer may be to remove the parameter from the driver.
Created attachment 141777 [details] Remove asd_attach_Host_RAID parameter from driver
Did not want to create a duplicate of this bug...but this was tested on CentOS 4.3 and Fedora Core 6. Hardware is a Tyan add-on card, the m9000-10, which uses the AIC-9410W chip: http://www.tyan.com/products/html/m9000-10.html The install phase properly detects and loads the adp94xx driver. However, drives attached to this are not visible and the system is unable to use any drive attached. This is true for both SAS and SATA drives. No parameter is able to make the drive visible, including attach_HostRAID, or asd_attach_HOST_RAID, with or without a ":1" argument (via the adaptec version of the driver). Various BIOS settings were tested, with never any ability to see the drives. "The lights are on, but nobody is home" syndrome. This same setup works with Tyan's outdated driver (binary only) for CentOS 4.1, which at least guarantees that the hardware functions and is installed correctly. The outdated version of CentOS 4.1 was able to load without error using Tyan's binary driver, which happens to load with the name adp94xx. This same setup also works correctly if SATA drives are used via the motherboard's integrated SATA (which sadly does not support SAS...this is what the m9000-10 is for). Note that the m9000-10 does not appear to have the ability to completely disable RAID, that the host motherboard RAID can and was tested disabled/enabled. When using the outdated Tyan-supplied driver, there was no effect when setting up software RAID, regardless of RAID configuration (it's always a software RAID, which ignores the m9000 RAID setting). I can't help but feel that if the drives were known to be attached, that the existing drivers would work (detection appears to be the only issue).
I think you are saying that you also tried removing the asd_attach_Host_RAID parameter (and/or setting it to 1) and that after doing that, the driver still failed to work? If so, then that is a different bug. I would still like to remove this parameter, as it serves no real purpose. On our hardware setup, the driver does appear to work with the parameter removed.
It is correct, I tried with and without asd_attach_Host_RAID, as well as similar arguments provided by the Adaptec public documentation. The controller is recognized just fine, but it never believes there is a drive attached. The same is not true of the binary driver distributed by Tyan (but which only works with a far out of date version of RHEL or SuSE)...the binary version detects drives just fine. My guess is that the driver is very close to working right.
We've obtained some info from Adaptec. Apparently the adp94xx driver runs in non-HostRAID mode. But that's OK for Stratus because we use it that way. Given that, we'd like to go forward and apply this patch so we can use this driver at install time. Applying this patch could surprise somebody that wanted to drive this particular hardware in the HostRAID configuration. The would end up with this driver being loaded and running in non-HostRAID config, but there doesn't appear to be a driver for this hardware in HostRAID config mode anyway.
Odd, I appended a comment here before but it's gone. Oh well. Here goes again: Stratus would like to declare this "not-a-bug". After a few e-mail exchanges with Adaptec, we learned that a BIOS utility can change the behavior of the card so that it is detected in non-HostRAID mode. Once that change is made, the driver should function correctly. We failed to discover this on our own because apparently our BIOS is broken and it doesn't actually change anything. Once our BIOS is fixed, we're sure this problem will go away. Can someone set the not-a-bug flag? Not sure I'm allowed to do that.
Kimball, assuming the patch no longer needs to be included, setting to CLOSED.
Although I can't verify this due to lack of test system (the original is now in production), the behavior from the Tyan m9000-10 would tend to agree with the BIOS conclusion...during bootup the BIOS utility has the option to set to non-HostRAID mode, but the setting change is ignored by the card. I have no ability to test out if a new BIOS would solve this, but also consider it closed, and believe the Tyan m9000-10 has the same defect as described above for other similar hardware, probably repairable with a BIOS update.