On Thu, Apr 15, 2010 at 4:21 PM, Tejun Heo <tj@kernel.org> wrote:
quoted text
> Hello,
>
> On 04/15/2010 10:31 AM, Minchan Kim wrote:
>> Hi, Tejun.
>>> This being a pretty cold path, I don't really see much benefit in
>>> converting it to alloc_pages_node_exact().  It ain't gonna make any
>>> difference.  I'd rather stay with the safer / boring one unless
>>> there's a pressing reason to convert.
>>
>> Actually, It's to weed out not-good API usage as well as some
>> performance gain.  But I don't think to need it strongly.
>> Okay. Please keep in mind about this and correct it if you confirms
>> it in future. :)
>
> Hmm... if most users are converting over to alloc_pages_node_exact(),
> I think it would be better to convert percpu too.  I thought it was a
> performance optimization (of rather silly kind too).  So, this is to
> weed out -1 node id usage?  Wouldn't it be better to update
> alloc_pages_node() such that it whines once per each caller if it's
> called with -1 node id and after updating most users convert the
> warning into WARN_ON_ONCE()?  Having two variants for this seems
> rather extreme to me.

Yes. I don't like it.
With it, someone who does care about API usage uses alloc_pages_exact_node but
someone who don't have a time or careless uses alloc_pages_node.
It would make API fragmentation and not good.
Maybe we can weed out -1 and make new API which is more clear.

* struct page *alloc_pages_any_node(gfp_t gfp_mask, unsigned int order);
* struct page *alloc_pages_exact_node(int nid, gfp_mask, unsigned int order);

So firstly we have to make sure users who use alloc_pages_node can
change alloc_pages_node with alloc_pages_exact_node.

After all of it was weed out, I will change alloc_pages_node with
alloc_pages_any_node.


-- 
Kind regards,
Minchan Kim
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/