Re: [PATCH] Keys: Improve usage of memory barriers and remove IRQ disablement

[David Howells <dhowells@redhat.com>]

Nick Piggin <nickpiggin@yahoo.com.au> wrote:

> Shouldn't be needed: Documentation/atomic_ops.txt specifies that any atomic_
> which both modifies its atomic operand and returns something is to be a full
> barrier before and after the operation.

Hmmm... It's possible that I've misunderstood what atomic_ops.txt actually
says.  For instance:

| 	int atomic_inc_and_test(atomic_t *v);
| 	int atomic_dec_and_test(atomic_t *v);
| 
| These two routines increment and decrement by 1, respectively, the
| given atomic counter.  They return a boolean indicating whether the
| resulting counter value was zero or not.
| 
| It requires explicit memory barrier semantics around the operation as
| above.

Note the last paragraph.  "It requires" should be "They require", but the
sense would seem to be obvious.  However, it's not clear on a second reading
as to whether this is an instruction to the _caller_ or an instruction to the
arch _implementer_.

I suppose from reading the abstract at the top:

| This document is intended to serve as a guide to Linux port maintainers on
| how to implement atomic counter, bitops, and spinlock interfaces properly.

that it is meant to be read by the implementor and not the user/caller, in which
case, Nick is correct.

It seems I need to adjust my memory barrier doc, and perhaps I should adjust
atomic_ops.txt too to make it clearer.

David