Re: [patch] updates XFS mutex patch

[Ingo Molnar <mingo@elte.hu>]

* Christoph Hellwig <hch@infradead.org> wrote:

> > > It's say just switch XFS to the one-arg mutex_init variant.
> > > 
> > > And ingo. please add the mutex_t typedef, analogue to spinlock_t it's 
> > > a totally opaqueue to the users type, so it really should be a 
> > > typedef.  After that the XFS mutex.h can just go away.
> > 
> > that's not possible, due to DEFINE_MUTEX() and due to struct mutex being 
> > embedded in other structures. I dont think we want to lose that property 
> > of struct semaphore, and only restrict mutex usage to pointers.
> 
> Sorry, but I don't get this sentence at all.  Can you try to rephrase 
> it? What does DEFINE_MUTEX have to do with declaring either a typedef 
> or structure?

i think i misunderstood you. I thought you wanted a mutex_t a'la 
kmem_cache_t (which is only fully defined in mm/slab.c) - for the 
purpose of hiding the implementation of mutex_t. If the implementation 
of mutex_t is still present in mutex.h, i dont see what the advantage 
is. What's the difference between 'struct mutex' and 'mutex_t', besides 
that first one being clearer that here we have a kernel object? (we 
generally use the _t types for type-safe variations of integer types, 
e.g. pte_t.)

	Ingo