rtai 3.2: latency test hangs

[Jan Kiszka]

Jacques GANGLOFF wrote:
> Hi,
> 
> I'm struggling with rtai 3.2 for a while and here
> is the state where I am stuck :
> 
> * Kernel 2.4.29 + rtai 3.2 & Kernel 2.6.12 + rtai 3.2 :
>   the latency test works in console mode but hangs after
>   pressing CTRL-C in an xterm (needs a hard-reset).
>   The 2 LEDs "Caps-lock" and "Scroll lock" are blinking
>   together after the crash.
> 
> * Kernel 2.6.12 + fusion 0.8.3 :
>   The latency test works perfectly well
> 
> * Kernel 2.4.29 + latest magma cvs :
>   Unable to run the test : unresolved symbols problems ->
>   /usr/realtime/modules/rtai_hal.o: /usr/realtime/modules/rtai_hal.o: 
> unresolved symbol __adeos_std_irq_dtype
>   /usr/realtime/modules/rtai_hal.o: /usr/realtime/modules/rtai_hal.o: 
> unresolved symbol adeos_extern_irq_handler
> 
> My hardware :
> 
> Pentium IV (one CPU) + SiS chipset + NVidia GPU.
> 
> BTW : with adeos compiled in, the binary "nvidia" driver gives me stranges
> complaints :
> Jul 26 11:25:12 pc3gra kernel: NVRM: Trying to sleep during raised irql!!
> Jul 26 11:25:12 pc3gra kernel: NVRM:   are we holding a lock?
> Jul 26 11:25:12 pc3gra kernel: NVRM:   skipping os_delay

Interesting output ("raised irql"), looks like they ported the Windows 
driver. Sorry, that's off-topic. ;)

> 
> This is not the cause of the crash since it occurs also when
> the X server is launched with the open-source "nv" driver.
> 
> Some help would be greatly appreciated. I would like
> preferably to run on a 2.4 kernel.
> 

You will have to avoid the nvidia binary driver if you want to run RTAI. 
It likely contains code directly manipulating the interrupt flag, thus 
violates Adeos/RTAI requirements.

This is kind of a general rule: no binary-only kernel modules under RTAI 
- as long as you did not reviewed them. I did for the madwifi HAL 
(Atheros WLAN driver), and it seems that this one is safe (the numbers 
we gained also comply with it).

Jan