RTLinux problem on Geode

Have you tried to run non RTLinux (that is, pristine Linux kernel compiled for Geode) on this very same board? I presume the posts you mention above regard latency which is a real Geode SMM managed pseudo hardware problem. Still it's been fairly simple to override as long one can live with latencys of around 100us. Couldn't your problem be related to Linux BIOS not initializing everything Linux/RTLinux assumes will be ready when it boots and maybe some SMM management that is missing and that shoudn't be? Have you tried it with another BIOS?

Regards.

Elder.

There is no guaranteed latency in standard Linux. It's good for soft realtime, but unusable for hard realtime.

-Michael

You're right. I was thinking of RTAI (and maybe RTLinux) when I wrote that part but did not make it clear.

Regards.

Elder.

What BIOS are you using? I have had numerous problems - mostly intermittent - with some VSA code. I've spent many hours hand-crafting hacked BIOS modules with different VSA code variants. You may be able to solve your problems by experimenting in this way.

But in general if you have hard RT requirements, I'd advise against Geode. You can mitigate problems by using XPressBIOS and *native* drivers for *everything* (i.e. no VSA), but it's a hell of a lot of work.

Thanks all for your suggestions. We had considered that it might be a LinuxBios problem so we reverted our National Semi Pompano development board back to the original Bios (XpressROM BIOS) and booted Linux via Etherboot and experienced the same issues. We also ran the on a Compulab Geode SC2200 development board with a General Systems BIOS with power management disabled with the same result. We are using the Adeos patch instead of RThal but I believe we had the same problem under RThal.

To simplify the environment we came up with a test case that has 2 simple hard real time tasks, one running at 10Hz and another running at 100Hz. Both do nothing more than increment a variable and go back to sleep.. Even this simple test case crashes intermittently on Geode based boards but runs fine on our Pentium development boards. Problem manifests as a memory access or invalid op-code exception. Starting to wonder if its some problem with the Geode MMU under Linux.

Have not tried developing any test case for Linux alone but we have been using Telnet routinely without problems so in general Linux appears to be OK. We really need the real time response provided by RTLinux, not sure we’ll learn much running under Linux alone but we might try some experiments if all else fails.

Hi Bill,

Look at the CMOS setup options in your BIOS. If there is anything in there related to, for instance, audio support at "normal" I/O addresses, then your BIOS contains VSA code (My understanding was that XPressBIOS doesn't, which is why I suggested it, but I haven't actually used it). This is the Root Of All Evil on Geode systems. Power management is a red herring. VSA code runs all the time, and is constantly stealing cycles from your application.

I think it's some kind of double fault condition, when the system is already in SMM and something happens that can't be serviced in SMM. It's not clear to me why these problems only manifest under Linux, but I surmise that the whole contraption is inherently rickety and it just hasn't been tested thoroughly in Linux.

Just AAMOI - Try changing the RAM configuration. Add more RAM. This fixes our problems, though it doesn't improve the realtime performance (or lack thereof) which Geode exhibits.