Re: [PATCH v2] x86/platform/uv: UV support for sub-NUMA clustering

[Steve Wahl]

On Wed, Jan 25, 2023 at 12:28:07PM +0100, Ingo Molnar wrote:
> 
> * Steve Wahl <steve.wahl@xxxxxxx> wrote:
> 
> > +static int __init alloc_conv_table(int num_elem, unsigned short **table)
> > +{
> > +	int i;
> > +	size_t bytes;
> > +
> > +	bytes = num_elem * sizeof(*table[0]);
> > +	*table = kmalloc(bytes, GFP_KERNEL);
> > +	WARN_ON_ONCE(!*table);
> > +	if (!*table)
> > +		return -ENOMEM;
> 
> WARN_ON_ONCE() is pass-through on the condition, so you can write this in a 
> shorter form:
> 
> 	if (!WARN_ON_ONCE(!*table))
> 		return -ENOMEM;

That is nicer.  I will incorporate this (including all the repeats).

...
> 
> > +		WARN_ON_ONCE(!new_hub);
> > +		if (!new_hub)
> > +			return;
> 
> Same. Also a memory leak of at least uv_hub_info_list_blade?

You're right, and this is not the only place.  I will rework.

...
> > +
> > +	for_each_node(nodeid) {
> > +		__uv_hub_info_list[nodeid] = uv_hub_info_list_blade[uv_node_to_blade_id(nodeid)];
> > +	}
> 
> Unnecessary curly braces.

I will fix.

> Looks good otherwise - presumably it's both tested and backwards compatible 
> with older UV hardware?

Yes, in fact there was a major re-work before letting it escape to the
world because the previous version didn't function on older hardware.

Thank you for taking the time to review!

--> Steve
-- 
Steve Wahl, Hewlett Packard Enterprise