Re: [RFC PATCH 1/4] workqueue: move the wq_update_unbound_numa_attrs_buf allocation location.

[wanghaibin]

On 2016/1/7 23:48, Tejun Heo wrote:

> On Thu, Jan 07, 2016 at 08:38:56PM +0800, wanghaibin wrote:
>> the wq_update_unbound_numa_attrs_buf will be useful, only when the
>> wq_numa_enabled is true.
>> if there is something wrong to cause the wq_numa_enable false, it
>> can just return without the wq_update_unbound_numa_attrs_buf
>> allocation.
>>
>> This doesn't introduce any functional changes.
> 
> I don't see what the point is with this change.
> 

What I Meant To Say,  if (WARN_ON(node == NUMA_NO_NODE)) is true, this cause
wq_numa_enabled to be set the false.  That is, the wq_update_unbound_numa_attrs_buf
will be useless.

It can free the the wq_update_unbound_numa_attrs_buf while the WARN_ON condition is
true;
Or, better way is that only when the wq_numa_enabled is true, we will allocate the
wq_update_unbound_numa_attrs_buf;

However, just like your said, the WARN_ON condition should never happen,
Maybe this change is not useless too :) .

> Thanks.
>