2.5.8: Preempt problems

2.5.8: Preempt problems

[Russell King]

Hi,

I'm trying to debug a flood of:

error: sh[81] exited with preempt_count 1

messages on ARM with kernel 2.5.8, and its looking like a generic kernel
bug is causing it.

The point at which preempt_count gets set to '1' for each task is at the
lock_kernel() in start_kernel().  Why?

We start booting the kernel.  One of the very first things we do is call
lock_kernel(), which increments the preempt_count for PID0.  We initialise
stuff, and then clone the thread for PID1, init.

When we clone a thread, dup_task_struct() creates a new kernel stack and
thread_info structure.  The childs thread_info structure inherits from
the parent, which includes the preempt_count of 1.

We start the user space init(8) program, and run various other programs,
forking as we do.  Each time we fork(), we call dup_task_struct() which
causes each task to inherit the non-zero preempt_count.

Thus, we never reach a preempt_count of 0, and never actually preempt.

Comments?  Is there some fiddle in the x86 code somewhere that we've
missed in the ARM tree to get this to work?

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html

Re: 2.5.8: Preempt problems

[Russell King]

On Wed, Apr 17, 2002 at 11:43:31AM +0100, Russell King wrote:
> Comments?  Is there some fiddle in the x86 code somewhere that we've
> missed in the ARM tree to get this to work?

Appologies - just found the problem.  We were missing the call to
schedule_tail() after a fork.

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html