Re: [PATCH][RFC] Discontigmem support for the x440

["Martin J. Bligh" <mbligh@aracnet.com>]

> +/* Identify which cnode a physical address resides on */
> +int pfn_to_nid(unsigned long pfn)
> +{
> +	int	i;
> +	struct node_memory_chunk_s *nmcp;
> +
> +	/* We've got a sorted list.  Binary search here?  Do we care?? */
> +	nmcp = node_memory_chunk;
> +	for (i = num_memory_chunks; --i >= 0; nmcp++)
> +		if (pfn >= nmcp->start_pfn && pfn <= nmcp->end_pfn)
> +			return (int)nmcp->nid;
> +
> +	return -1;
> +}

This is called a lot, and it's large and inefficient. Can you turn it 
into an array lookup like the NUMA-Q implementation, and inline it? 

All the clumps and chunks stuff can go, I think.

> diff -Nru a/drivers/acpi/events/evevent.c b/drivers/acpi/events/evevent.c
> --- a/drivers/acpi/events/evevent.c	Wed Feb  5 19:15:58 2003
> +++ b/drivers/acpi/events/evevent.c	Wed Feb  5 19:15:58 2003
> @@ -104,6 +104,7 @@
>  
>  	ACPI_FUNCTION_TRACE ("ev_handler_initialize");
>  
> +	return_ACPI_STATUS (0);
>  
>  	/* Install the SCI handler */

That used to be wrapped in ifdef CONFIG_SUMMIT, which seems much safer
to me ... any reason for the change?

And all the kludge stuff needs to go, but you know that already ;-)

M.