HP 16505A Operating System Upgrade

[DanJP]

I am in the process of trying to update the operating system for the HP 16505A to use HPUX 10.20 or later versus the HPUX 9.05 it was supplied with and have run into an issue with the original HPUX 9.05 vs 10.20 SCSI Passthrough Driver.

In particular the HP 16505A is a dedicated workstation (same as a HP 712/100) that was to hooked up to a HP 16500B/C Logic Analyzer via the external SCSI interface to provide the analysis tools for the analyzer.

The original software supplied in the HP 16505A was HPUX 9.05 and was tailored to only run the analysis program (vp) for the analyzer. It was very easy to gain access to the OS via the IPL and setup a root account. So easy to see how it was setup.

What I found was that the analysis program used the SCSI passthrough driver SCTL to provide the I/O to the analyzer. Running an ioscan on the HPUX 9.05 OS showed the analyzer as a scanner. Not sure why at this point but it was at the correct address. In addition the device nodes used are /dev/scsi/0 - 6 and used the major device 105 which was the scsi passthrough driver. Since the dedicated software to control the analyzer is easy to partition out of the base system, I wanted to see if I could upgrade to 10.20 or even 11 and get the analysis software to work on the upgraded OS. So I did a cold install of 10.20 with no issues. I was also able to transplant the analysis software without issue. I also recreated the /dev/scsi/0 - 6 using the new major device of 203 via mknod which is for the passthrough driver in 10.20.

The problem I am having is that when I hookup the analyzer and do a ioscan the OS hangs with ibolt errors recorded in the syslog. ioscan is hanging up on the analyzer address. Same happens during a cold boot. The OS hangs for about 5 min and then proceeds. The results from ioscan on 10.20 do not show the analyzer at all.

Does anyone know what the difference is between the SCSI drivers between 9.05 and 10.20 that might give me some hope of getting this working as it seems that something changed in the SCSI drivers between 9.05 and 10.20.

[Bill Hassell]

The workstation and instrument connections were heavily dependent on the version of OS for which they were released. 

9.05 was released in 1993, more than 25 years ago. And back in those days, the passthrough driver was very problematic. I'm afraid there is no hope to port this instrument to a newer OS. Support for instruments on HP-UX is long gone with the Agilent spinoff in 2000 and HP workstations such as the 712 are completely unsupported.

Your workstation cannot run the only supported version of HP-UX which is 11.31.
Version 10.20 was released in 1996 with support ending in 2003.

Is there a reason you have to try to upgrade?

[DanJP]

Thanks for the reply.

There is no real compelling reason to upgrade to a newer (although unsupported) OS. The original works OK but I thought it would be fun to try and extend the life of the equipment by bringing it slightly into the future.

What surprises me though is the area I am having issues with to me seems like it would have been the most stable that being the SCSI interface. I had read other posts where people were saying the passthrough driver did not work properly on 10.20 vs 9.05 resulting in standard items such as scanners not working after an upgrade. The symptoms that they reported such as bus hangs being the same as what I am seeing. I would have thought that this would not have been the case and that moving to a newer OS would not cause a SCSI issue.

So I was asking if there were any patches specifically like PHKL_5406 for the S800 that can restore the old 9.05 driver into the 10.20 kernel or is there any other way to port the old 9.05 driver into 10.20.

[Bill Hassell]

Since the intended purpose of the OS is to run the 16505A logic analyzer, a newer OS won't improve the operation. The differences between 9.05 and 10.xx or 11.xx are enormous at the kernel level. It isn't possible to load the SCSI driver from 9.05 into a newer version due to numerous changes within the OS. A while SCSI might seem to be a generic label, there are numerous versions of the driver to accomodate not only interfaces but many variations in the protocol. Standard SCSI? The problem with a standard is that there are so many to choose from... And the passthrough module is essentially a sidedoor (trapdoor?) to bypass the primary design of the driver and provide a custom handler for non-disk and non-tape devices. 

The biggest issue with specialized instrument systems is that they are very badly out of date with security issues. Since there have been no new patches for 9.05 in more than 20 years, the network for these types of instruments must be totally isolated, much in the same way that Java-based web pages from special hardware such as consoles must be protected. These networks must not allow router connections beyond the lab environment. When your company performs independent security audits, they will find numerous vulnerabilities that cannot be fixed.

[DanJP]

All good reasons. You are just confirming my suspicsion that there is no way to make this work.

As for security, I am not worried as this is for private use and not coporate use so no issue with isolation.

What this does seem point to though, is that the pass through driver does not work past 9.05 or works rather poorly. It would be interesting to see if anything else that used the passthrough driver worked past 9.05. My suspicsion is that things that did work on 9.05 no longer work after 10.20.

I could probably make this work if it were possible to find the source for the passthrough driver from both OS's to see what changed but most likely that is not going to happen.

It is a shame though that for something as simple as a passthrough driver to block for what could have been an easy upgrade path.

[DanJP]

I forgot to mention, if it makes any difference, is that when I used mc (the media changer) program to test the device nodes, what I saw on 10.20 was an I/O error being reported on the newly created /dev/scsi/0-6 device nodes with or without the analyzer hooked up. Not sure is this is pointing to an issue with the device node itself or just a symptom of the driver iteself. The mc helpfile does not list an I/O error as something it reports.

[Bill Hassell]

The symptom you are seeing with mc also affected optical disc and tape changers. And zero diagnostic messages for troubleshooting. In most cases, the hardware is fine but the handshake did not work. Starting at 11.00 and beyond, the tape changers now work well but they have their own changer driver (separate from the tape) rather than the passthrough kludge.