Re: Question on Schedule and Preemption

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, 2006-04-19 at 01:21 +0800, Liu haixiang wrote:
> 2006/4/18, Steven Rostedt <rostedt@xxxxxx.com>:

Damn! I used my work email account to send that. I try to keep the spam
down on that account, so I don't publish it as much. Darn Evolution
keeps picking the wrong account.

> > On Tue, 2006-04-18 at 20:55 +0800, Liu haixiang wrote:
> >
> > Well according to your output, it was the task "TURNER0" with the pid
> > 611.  After you recompile the kernel with CONFIG_FRAME_POINTERS you will
> > need to follow the stack trace and see what function turned off
> > preemption.  This can happen calling a spin_lock and not unlocking it
> > before calling something that would schedule.
> >
> 
> If it's true, it is a great help for me.
> 
> > If you also turn on in "Kernel Hacking" -> "Compile the kernel with
> > debug info", you can then do a "gdb vmlinux" in the root directory of
> > the compile, and pass in the stack address with the command
> > "li *0xc042382e" to show where that function would return. (replace the
> > c042382e with the location you are looking for).
> >
> > Also, what arch are you using to get a 0x04000000 in the preempt_count
> > (the 0x1 is the depth). Of course I'm looking at 2.6.17-rc1 and not the
> > one you are using.  Hmm, your arch may not even support frame pointers.
> >
> > -- Steve
> >
> 
> The arch is SH4.

Well I just tried a "make ARCH=sh menuconfig" and the
CONFIG_FRAME_POINTERS option is there.  Whether it makes a difference or
not, I haven't the foggiest clue.

-- Steve