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

[Rusty Russell <rusty@rustcorp.com.au>]

In message <19317.992115181@redhat.com> you write:
> 
> torvalds@transmeta.com said:
> >  Good point. Spinlocks (with the exception of read-read locks, of
> > course) and semaphores will deadlock on recursive use, while the BKL
> > has this "process usage counter" recursion protection.
> 
> Obtaining a read lock twice can deadlock too, can't it?
> 
> 	A		B
> 	read_lock()
> 			write_lock()
> 			...sleeps...
> 	read_lock()
> 	...sleeps...
> 
> Or do we not make new readers sleep if there's a writer waiting?

We can never[1] make new readers sleep if there's a writer waiting, as
Linus guaranteed that an IRQ handler which only ever grabs a read lock
means the rest of the code doesn't need to block interrupts on its
read locks (see Documentation/spinlock.txt IIRC).

Also, netfilter will break (brlocks inherit this property from
their spinlocks constituents).

Rusty.
[1] Well, we could, but we'd have to do a special "same CPU?" check,
    which would suck badly.
--
Premature optmztion is rt of all evl. --DK