On Saturday 17 May 2008 14:09:17 Manish Katiyar wrote: > On Sat, May 17, 2008 at 5:34 PM, Indraneel Mukherjee > > <[EMAIL PROTECTED]> wrote: > > Hi, > > I've put the following print in context_switch() in kernel/sched.c > > printk( " %d ", next->pid); > > > > Can that serve the purpose as well? > > It might, but as Johannes pointed out, it might be too many context > switches per sec. to track so you might want to restrict it in some > way.
What if you (and I'm just thinking loudly now, this might be a bad idea) a) create a buffer in kernelmode, say 1MB b) for every context-switch, add PID (and whatever extra information you need) to this buffer. Use it like a ring-buffer, so if you overflow, you just overwrite the oldest entries c) create a device-driver (block-type would be most appropriate I guess) that read (and empty) the buffer. d) Then, you only need to create a small user-land script that reads the block-driver often enought to extract the info. Does this make any sense? I am actually quite keen to try this out, so I might have something working soon (where soon is defined to contain all integers from 1 to infinity) :-) > But the basic Q is why do you want that output ? How are you going to > interpret it ? What are you trying to do ? I cant speak for Mukhherjee, but personally, it would be nice to know *how* often the processes are being scheduled, regardles of tick or not, and also for how long they have been running etc. -- mvh Henrik Austad
signature.asc
Description: This is a digitally signed message part.
