Re: [PATCH]memblock: do double array and add merge directly path inmemblock_insert_region

[Tejun Heo]

Hello,

On Mon, Jun 10, 2013 at 01:34:30AM +0800, Wang YanQing wrote:
> If we allow do merge directly when insertion, we can get better performance,
> this patch add support to do merge directly in memblock_insert_region.

Do you have actual case where performance in this patch is
problematic?

> -static void __init_memblock memblock_insert_region(struct memblock_type *type,
> +static int __init_memblock memblock_insert_region(struct memblock_type *type,
>  						   int idx, phys_addr_t base,
> -						   phys_addr_t size, int nid)
> +						   phys_addr_t size, int nid, int merge)

bool merge and you need to update comment accordingly.

>  {
>  	struct memblock_region *rgn = &type->regions[idx];
>  
> -	BUG_ON(type->cnt >= type->max);
> +	if (merge && (base + size) == rgn->base &&
> +	        nid == memblock_get_region_node(rgn)) {
> +		rgn->base = base;
> +		rgn->size += size;
> +		type->total_size += size;
> +		return 0;
> +	}

What if merge is from upside down or this connects two disjoint
regions?

>  static int __init_memblock memblock_add_region(struct memblock_type *type,
>  				phys_addr_t base, phys_addr_t size, int nid)
...
>  		if (rbase > base) {
> -			nr_new++;
> -			if (insert)
> -				memblock_insert_region(type, i++, base,
> -						       rbase - base, nid);
> +			ret = memblock_insert_region(type, i++, base,
> +					rbase - base, nid, 1);
> +			if (ret) {
> +				return ret;
> +			}

Superflous { } and you need roll back what you've done upto this point
before returning.  That's the reason why the function is structured
the way it is.  If insertion fails due to array expansion failure, the
previous code doesn't change the memblock tables at all.  Your new
code leaves it half-updated.

> @@ -498,8 +488,11 @@ static int __init_memblock memblock_isolate_range(struct memblock_type *type,
>  			rgn->base = base;
>  			rgn->size -= base - rbase;
>  			type->total_size -= base - rbase;
> -			memblock_insert_region(type, i, rbase, base - rbase,
> -					       memblock_get_region_node(rgn));
> +			ret = memblock_insert_region(type, i, rbase, base - rbase,
> +					memblock_get_region_node(rgn), 0);
> +			if (ret) {
> +				return ret;
> +			}

Ditto, you need to roll back on failure.

> @@ -533,6 +529,7 @@ static int __init_memblock __memblock_remove(struct memblock_type *type,
>  
>  	for (i = end_rgn - 1; i >= start_rgn; i--)
>  		memblock_remove_region(type, i);
> +	memblock_merge_regions(type);

Why is this added?

Thanks.

-- 
tejun
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/