Configuring the watchdog timer under Linux/X

I am using a GT240 under X, and while experimenting with CUDA, sometimes (due to bugs of my own) the CUDA kernel takes too long to finish. While that happens, my PC is frozen. I know that under Windows XP, there is a watchdog timer that “kills” the kernel if enough time has passed. My question:

A. Does this watchdog exist under Linux, when a GT240 is also being used to run the X server?
B. If so, how can I configure this timer? Is there an environment variable or something?

Thanks in advance,

There is a timer, and it is automatically enabled and will kick in after about 5 seconds. There is no configuration either available or required to make it work. If you are experiencing freezes, it might be a symptom of something else, either hardware or software.


I am using a configuration where the CUDA kernel is writing inside an OpenGL PBO - to accelerate displaying of the results (i.e. show the picture formed by the CUDA kernel without passing it back and forth over the bus).

I used cuda-memcheck, and it verified that my code is OK BUT only when I disabled the writing to the PBO - if I let it in, cuda-memcheck things these are invalid writes… (they are not).

Any specific suggestions on what else to check?


Does this timer also work when X is not used? How to disable it entirely? My application would initially need that I think…

No, it is a feature of the display manager model. No display manager on a card (so no display manager on a given card or no X11 at all) = no watchdog timer.