[ltt-dev] [PATCH] Fix dirty page accounting in redirty_page_for_writepage()

[cl@linux.com (Christoph Lameter)]

On Thu, 30 Apr 2009, Ingo Molnar wrote:

> The new percpu APIs could be used in most of these places already,
> straight away. This is a really good TODO list for places to
> enhance.

Please look a the full list in the cpu alloc v3 patchset and not only
those that I listed here.

> Then a second set of patches could convert percpu_add() / etc. uses
> to __percpu_add() ... but that should be done by those architectures
> that need it (and to the extent they need it), because it's not
> really testable on x86.

Ok So we convert it and wait until the arch maintainers complain? I
definitely know that there is an IA64 issue with vm statistics.

> I dont really like the PER_CPU / CPU_INC etc. type of all-capitals
> APIs you introduced in the patches above:

I know. Patches would have to be redone against whatever API we agree on.

>
> +		__CPU_INC(bt->sequence);
> +	CPU_FREE(bt->sequence);
>
> was there any strong reason to go outside the well-established
> percpu_* name space and call these primitives as if they were
> macros?

They are macros and may do weird things with the variables. This goes back
to our disagreement last year on caps/lower case. I still think this kind
of preprocessor magic should be uppercase.

The reason not to use the percpu_* names was that they were x86 arch
specific (and thus not available) and did not differentiate in terms of
the irq/preemption context.