[RFC PATCH] mm: memblock: optimize memblock_find_in_range_node() to minimize the search work

[RFC PATCH] mm: memblock: optimize memblock_find_in_range_node() to minimize the search work

[Lin Feng]

The memblock array is in ascending order and we traverse the memblock array in
reverse order so we can add some simple check to reduce the search work.

Tejun fix a underflow bug in 5d53cb27d8, but I think we could break there for
the same reason.

Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Lin Feng <linfeng@cn.fujitsu.com>
---
 mm/memblock.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/mm/memblock.c b/mm/memblock.c
index 6259055..a710557 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -111,11 +111,18 @@ phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
 	end = max(start, end);
 
 	for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
+		/*
+		 * exclude the regions out of the candidate range, since it's
+		 * likely to find a suitable range, we ignore the worst case.
+		 */
+		if (this_start >= end)
+			continue;
+
 		this_start = clamp(this_start, start, end);
 		this_end = clamp(this_end, start, end);
 
 		if (this_end < size)
-			continue;
+			break;
 
 		cand = round_down(this_end - size, align);
 		if (cand >= this_start)
-- 
1.7.11.7

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC PATCH] mm: memblock: optimize memblock_find_in_range_node() to minimize the search work

[Tejun Heo]

On Fri, Jan 04, 2013 at 05:24:53PM +0800, Lin Feng wrote:
> The memblock array is in ascending order and we traverse the memblock array in
> reverse order so we can add some simple check to reduce the search work.
> 
> Tejun fix a underflow bug in 5d53cb27d8, but I think we could break there for
> the same reason.
> 
> Cc: Tejun Heo <tj@kernel.org>
> Signed-off-by: Lin Feng <linfeng@cn.fujitsu.com>
> ---
>  mm/memblock.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 6259055..a710557 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -111,11 +111,18 @@ phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
>  	end = max(start, end);
>  
>  	for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
> +		/*
> +		 * exclude the regions out of the candidate range, since it's
> +		 * likely to find a suitable range, we ignore the worst case.
> +		 */
> +		if (this_start >= end)
> +			continue;
> +
>  		this_start = clamp(this_start, start, end);
>  		this_end = clamp(this_end, start, end);
>  
>  		if (this_end < size)
> -			continue;
> +			break;

I don't know.  This only saves looping when memblocks are below the
requested size, right?  I don't think it would matter in any way and
would prefer to keep the logic as simple as possible.

Thanks.

-- 
tejun

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC PATCH] mm: memblock: optimize memblock_find_in_range_node() to minimize the search work

[Lin Feng]

On 01/04/2013 11:01 PM, Tejun Heo wrote:
> On Fri, Jan 04, 2013 at 05:24:53PM +0800, Lin Feng wrote:
>> The memblock array is in ascending order and we traverse the memblock array in
>> reverse order so we can add some simple check to reduce the search work.
>>
>> Tejun fix a underflow bug in 5d53cb27d8, but I think we could break there for
>> the same reason.
>>
>> Cc: Tejun Heo <tj@kernel.org>
>> Signed-off-by: Lin Feng <linfeng@cn.fujitsu.com>
>> ---
>>  mm/memblock.c | 9 ++++++++-
>>  1 file changed, 8 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/memblock.c b/mm/memblock.c
>> index 6259055..a710557 100644
>> --- a/mm/memblock.c
>> +++ b/mm/memblock.c
>> @@ -111,11 +111,18 @@ phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
>>  	end = max(start, end);
>>  
>>  	for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
>> +		/*
>> +		 * exclude the regions out of the candidate range, since it's
>> +		 * likely to find a suitable range, we ignore the worst case.
>> +		 */
>> +		if (this_start >= end)
>> +			continue;
>> +
>>  		this_start = clamp(this_start, start, end);
>>  		this_end = clamp(this_end, start, end);
>>  
>>  		if (this_end < size)
>> -			continue;
>> +			break;
> 
> I don't know.  This only saves looping when memblocks are below the
> requested size, right?  I don't think it would matter in any way and
> would prefer to keep the logic as simple as possible.
Hi Tejun,

You're right, when we hit the 'if (this_end < size)' branch, it's nearly 
the end of the whole search loops. I just got an impression that is 
there any candidate range after we hit the if clause when I first read
this code, so... ;-)

thanks,
linfeng
> 
> Thanks.
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>