Re: [CHECKER] a couple potential deadlocks in 2.4.5-ac8

[torvalds@transmeta.com (Linus Torvalds)]

In article <19317.992115181@redhat.com>,
David Woodhouse  <dwmw2@infradead.org> wrote:
>
>Obtaining a read lock twice can deadlock too, can't it?

If it does (with spinlocks), then that's an implementation bug (which
might well be there).  We depend on the read-lock being recursive in a
lot of places, notably the fact that we don't disable interrupts while
holding read-locks if we know that the interrupt routines only take a
read-lock. 

>	A		B
>	read_lock()
>			write_lock()
>			...sleeps...
>	read_lock()
>	...sleeps...
>
>Or do we not make new readers sleep if there's a writer waiting?

The writer-waiter should not be spinning with the write lock held.

Note that the blocking versions are different, and I explicitly meant
only the read-spinlocks, not read-semaphores. For the semaphores I think
your schenario is indeed correct.

		Linus