printks in the scheduler freeze during scripts

printks in the scheduler freeze during scripts

[anton wilson]

I'm running linux 2.4.17 and Redhat 7.2 with the preemptive and low latency 
patches, and whenever I stick printks in the scheduler(void) my system 
freezes somewhere after it tries to load the system font. Where it stops 
seems to be random. I can only run under single user mode without my system 
freezing. Does anyone have any clues why? Or any better ways to go about 
tracking the scheduling of processes in the scheduler?


Anton
-- 
Camotion
Software Development
678-471-0895

Re: printks in the scheduler freeze during scripts

[Andrew Morton]

anton wilson wrote:
> 
> I'm running linux 2.4.17 and Redhat 7.2 with the preemptive and low latency
> patches, and whenever I stick printks in the scheduler(void) my system
> freezes somewhere after it tries to load the system font. Where it stops
> seems to be random. I can only run under single user mode without my system
> freezing. Does anyone have any clues why? Or any better ways to go about
> tracking the scheduling of processes in the scheduler?
> 

printk() calls wake_up(), to give klogd a kick.  So a printk
from the scheduler tends to deadlock.  Just delete the
last two lines of kernel/printk.c:release_console_sem() and it
should work OK.

Re: printks in the scheduler freeze during scripts

[Michael S. Zick]

On Thursday 27 June 2002 09:03 am, anton wilson wrote:
> I'm running linux 2.4.17 and Redhat 7.2 with the preemptive and low latency
> patches, and whenever I stick printks in the scheduler(void) my system
> freezes somewhere after it tries to load the system font. Where it stops
> seems to be random. I can only run under single user mode without my system
> freezing. Does anyone have any clues why? Or any better ways to go about
> tracking the scheduling of processes in the scheduler?

Something I have used in my testing:

Add, in the task structure:

task_t  *sched_from;

In schedule, prior to context_switch:

next->sched_from = prior; /* at this point still == get_current ()*/

Somewhere else (so you don't inadvertantly side-effect to death schedule()):

Follow the back-link and printk whatever for the task that you are interested 
in, being careful not to reference task structures that no longer exist.

Mike
>
>
> Anton