Re: [PATCH 1/5] try parent numa_node at first before using default

[Cornelia Huck <cornelia.huck@de.ibm.com>]

On Thu, 12 Jul 2007 07:47:52 +0200,
Stefan Richter <stefanr@s5r6.in-berlin.de> wrote:

> The patch does nothing for all subsystems which do
> 
> 	device_initialize(&dev);
> 	dev->parent = pd;
> 	device_add(&dev);
> 
> Let's avoid to add infrastructure which does nothing, or only does
> something by accident.

I agree. dev->parent now is only expected to be set in device_add(),
not in device_initialize(), so the sequence above is perfectly fine.
(Why should you need it, anyway? parent information becomes necessary
only when something is added to the tree.)
> 
> The alternatives are:
> 
>   - Change all subsystems to set dev->parent before device_initialize().
>     *Document* that the device_initialize() API has this requirement.
>     This is counter-intuitive, amounts to some work across the kernel,
>     and could be gotten wrong again in future code because it's a
>     counter-intuitive API.

Yes. We shouldn't do that.
> 
>   - Move your code from device_initialize() to device_add().  One minor
>     drawback is that node-specific allocations based on the device's
>     numa_node would not be optimized before device_add(), but there is
>     probably no need for this.  Driver probes come after device_add().

I'd expect most allocations to be done when probing, so this shouldn't
hurt much.
> 
>   - Let subsystems explicitly call set_dev_node() on their own.
> 
> 
> Also keep in mind that either device_move() should update the numa_node,
> or the subsystems which call device_move() should explicitly update it
> on their own.   (Unless they know that their devices will always stay at
> the same NUMA node even when switching parents.)

I'd trust the subsystems to know best whether something regarding NUMA
changed.