Re: [CHECKER] potential NFS deadlock in 2.6.8.1

[Nick Piggin <nickpiggin@yahoo.com.au>]

Dawson Engler wrote:
> Hi All,
> 
> below is a possible deadlock in the linux-2.6.8.1 NFS found by a static
> deadlock checker I'm writing.  Let me know if it looks valid and/or
> whether the output is too cryptic.
> 
> Thanks,
> Dawson
> 
> ------------------------------------------------------------------
> 
> ERROR:DEADLOCK: 2 thread cycle:
>      lock_kernel  <<===>>  client_sema
> 

Hi Dawson, your tool looks nice. I don't think the output is too
cryptic.

However, I think these deadlocks may be incorrect because
lock_kernel is not a normal lock. It gets dropped when the
thread blocks, so I *think* it is basically invisible to any
sort of blocking lock when it comes to deadlock cycles.

So if your threads interleave like this:
1			2
lock_kernel();
			down(semaphore);
			lock_kernel();
down(semaphore);

Then thread 2 will spin on the lock_kernel lock (bkl) until
thread 1 hits down(semaphore) and goes to sleep, releasing
the bkl. Thread 2 takes the bkl and proceeds.

I think?