On 3/27/20 10:24 AM, Johannes Thumshirn wrote:

On 27/03/2020 12:21, Christoph Hellwig wrote:

quoted

Because the two variants are rather pointless.  And this might get
more people to actually pass a useful node ID instead of copy and
pasting some old example code. 

So then everybody will copy the NUMA_NO_NODE ones and in X years someone 
will come up with a wrapper passing in NUMA_NO_NODE to 
blk_alloc_queue_node(), because he/she didn't read the commit history.

If you care about numa-node, then you are generally using it for other
allocations as well, so it should come naturally. If you don't, then
the copy/paste is fine. Better to expose it as the main API, so people
don't miss it.

-- 
Jens Axboe