Date: Wed, 20 Mar 2013 15:32:01 -0700
From: Andrew Morton <akpm@linux-foundation.org>
To: Russ Anderson <rja@sgi.com>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org, tglx@linutronix.de,
        mingo@redhat.com, hpa@zytor.com, linux-ia64@vger.kernel.org
Subject: Re: [patch] mm: speedup in __early_pfn_to_nid
Message-Id: <20130320153201.29c19769f9b29470bab822b5@linux-foundation.org>
In-Reply-To: <20130318155619.GA18828@sgi.com>
References: <20130318155619.GA18828@sgi.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2472
Lines: 62

On Mon, 18 Mar 2013 10:56:19 -0500 Russ Anderson <rja@sgi.com> wrote:

> When booting on a large memory system, the kernel spends
> considerable time in memmap_init_zone() setting up memory zones.
> Analysis shows significant time spent in __early_pfn_to_nid().
> 
> The routine memmap_init_zone() checks each PFN to verify the
> nid is valid.  __early_pfn_to_nid() sequentially scans the list of
> pfn ranges to find the right range and returns the nid.  This does
> not scale well.  On a 4 TB (single rack) system there are 308
> memory ranges to scan.  The higher the PFN the more time spent
> sequentially spinning through memory ranges.
> 
> Since memmap_init_zone() increments pfn, it will almost always be
> looking for the same range as the previous pfn, so check that
> range first.  If it is in the same range, return that nid.
> If not, scan the list as before.
> 
> A 4 TB (single rack) UV1 system takes 512 seconds to get through
> the zone code.  This performance optimization reduces the time
> by 189 seconds, a 36% improvement.
> 
> A 2 TB (single rack) UV2 system goes from 212.7 seconds to 99.8 seconds,
> a 112.9 second (53%) reduction.
> 
> ...
>
> --- linux.orig/mm/page_alloc.c	2013-03-18 10:52:11.510988843 -0500
> +++ linux/mm/page_alloc.c	2013-03-18 10:52:14.214931348 -0500
> @@ -4161,10 +4161,19 @@ int __meminit __early_pfn_to_nid(unsigne
>  {
>  	unsigned long start_pfn, end_pfn;
>  	int i, nid;
> +	static unsigned long last_start_pfn, last_end_pfn;
> +	static int last_nid;
> +
> +	if (last_start_pfn <= pfn && pfn < last_end_pfn)
> +		return last_nid;
>  
>  	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
> -		if (start_pfn <= pfn && pfn < end_pfn)
> +		if (start_pfn <= pfn && pfn < end_pfn) {
> +			last_nid = nid;
> +			last_start_pfn = start_pfn;
> +			last_end_pfn = end_pfn;
>  			return nid;
> +		}
>  	/* This is a memory hole */
>  	return -1;

lol.  And yes, it seems pretty safe to assume that the kernel is
running single-threaded at this time.

arch/ia64/mm/numa.c's __early_pfn_to_nid might benefit from the same
treatment.  In fact if this had been implemented as a caching wrapper
around an unchanged __early_pfn_to_nid(), no ia64 edits would be
needed?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/