[RFC] mm/compaction: ignore block suitable after check large free page

[RFC] mm/compaction: ignore block suitable after check large free page

[Yisheng Xie]

If the migrate target is a large free page and we ignore suitable,
it may not good for defrag. So move the ignore block suitable after
check large free page.

Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
---
 mm/compaction.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/mm/compaction.c b/mm/compaction.c
index 0fdfde0..4bf2a5d 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
 static bool suitable_migration_target(struct compact_control *cc,
 							struct page *page)
 {
-	if (cc->ignore_block_suitable)
-		return true;
-
 	/* If the page is a large free page, then disallow migration */
 	if (PageBuddy(page)) {
 		/*
@@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
 			return false;
 	}
 
+	if (cc->ignore_block_suitable)
+		return true;
+
 	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
 	if (migrate_async_suitable(get_pageblock_migratetype(page)))
 		return true;
-- 
1.7.12.4

[RFC] mm/compaction: ignore block suitable after check large free page

[Yisheng Xie]

If the migrate target is a large free page and we ignore suitable,
it may not good for defrag. So move the ignore block suitable after
check large free page.

Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
---
 mm/compaction.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/mm/compaction.c b/mm/compaction.c
index 0fdfde0..4bf2a5d 100644
--- a/mm/compaction.c
+++ b/mm/compaction.c
@@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
 static bool suitable_migration_target(struct compact_control *cc,
 							struct page *page)
 {
-	if (cc->ignore_block_suitable)
-		return true;
-
 	/* If the page is a large free page, then disallow migration */
 	if (PageBuddy(page)) {
 		/*
@@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
 			return false;
 	}
 
+	if (cc->ignore_block_suitable)
+		return true;
+
 	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
 	if (migrate_async_suitable(get_pageblock_migratetype(page)))
 		return true;
-- 
1.7.12.4

--
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] mm/compaction: ignore block suitable after check large free page

[Vlastimil Babka]

On 03/10/2017 05:20 AM, Yisheng Xie wrote:
> If the migrate target is a large free page and we ignore suitable,
> it may not good for defrag. So move the ignore block suitable after
> check large free page.

Right. But in practice I expect close to no impact, because direct
compaction shouldn't have to be called if there's a >=pageblock_order
page already available.

> Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
> ---
>  mm/compaction.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/compaction.c b/mm/compaction.c
> index 0fdfde0..4bf2a5d 100644
> --- a/mm/compaction.c
> +++ b/mm/compaction.c
> @@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
>  static bool suitable_migration_target(struct compact_control *cc,
>  							struct page *page)
>  {
> -	if (cc->ignore_block_suitable)
> -		return true;
> -
>  	/* If the page is a large free page, then disallow migration */
>  	if (PageBuddy(page)) {
>  		/*
> @@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>  			return false;
>  	}
>  
> +	if (cc->ignore_block_suitable)
> +		return true;
> +
>  	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>  	if (migrate_async_suitable(get_pageblock_migratetype(page)))
>  		return true;
> 

Re: [RFC] mm/compaction: ignore block suitable after check large free page

[Vlastimil Babka]

On 03/10/2017 05:20 AM, Yisheng Xie wrote:
> If the migrate target is a large free page and we ignore suitable,
> it may not good for defrag. So move the ignore block suitable after
> check large free page.

Right. But in practice I expect close to no impact, because direct
compaction shouldn't have to be called if there's a >=pageblock_order
page already available.

> Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
> ---
>  mm/compaction.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/compaction.c b/mm/compaction.c
> index 0fdfde0..4bf2a5d 100644
> --- a/mm/compaction.c
> +++ b/mm/compaction.c
> @@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
>  static bool suitable_migration_target(struct compact_control *cc,
>  							struct page *page)
>  {
> -	if (cc->ignore_block_suitable)
> -		return true;
> -
>  	/* If the page is a large free page, then disallow migration */
>  	if (PageBuddy(page)) {
>  		/*
> @@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>  			return false;
>  	}
>  
> +	if (cc->ignore_block_suitable)
> +		return true;
> +
>  	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>  	if (migrate_async_suitable(get_pageblock_migratetype(page)))
>  		return true;
> 

--
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] mm/compaction: ignore block suitable after check large free page

[Yisheng Xie]

Hi Vlastimil,

Thanks for comment.
On 2017/3/10 15:30, Vlastimil Babka wrote:
> On 03/10/2017 05:20 AM, Yisheng Xie wrote:
>> If the migrate target is a large free page and we ignore suitable,
>> it may not good for defrag. So move the ignore block suitable after
>> check large free page.
> 
> Right. But in practice I expect close to no impact, because direct
> compaction shouldn't have to be called if there's a >=pageblock_order
> page already available.
> 
Maybe you are right and this change is just based on logical analyses.

Presently, only in direct compaction, we increase the compaction priority,
and ignore suitable at MIN_COMPACT_PRIORITY. I have a silly question, can
we do the similar thing in kcompactd? maybe by doing most work in kcompactd,
we can get better perf of slow path.

Thanks
Yisheng Xie

>> Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
>> ---
>>  mm/compaction.c | 6 +++---
>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index 0fdfde0..4bf2a5d 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
>>  static bool suitable_migration_target(struct compact_control *cc,
>>  							struct page *page)
>>  {
>> -	if (cc->ignore_block_suitable)
>> -		return true;
>> -
>>  	/* If the page is a large free page, then disallow migration */
>>  	if (PageBuddy(page)) {
>>  		/*
>> @@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>>  			return false;
>>  	}
>>  
>> +	if (cc->ignore_block_suitable)
>> +		return true;
>> +
>>  	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>>  	if (migrate_async_suitable(get_pageblock_migratetype(page)))
>>  		return true;
>>
> 
> 
> .
> 

Re: [RFC] mm/compaction: ignore block suitable after check large free page

[Yisheng Xie]

Hi Vlastimil,

Thanks for comment.
On 2017/3/10 15:30, Vlastimil Babka wrote:
> On 03/10/2017 05:20 AM, Yisheng Xie wrote:
>> If the migrate target is a large free page and we ignore suitable,
>> it may not good for defrag. So move the ignore block suitable after
>> check large free page.
> 
> Right. But in practice I expect close to no impact, because direct
> compaction shouldn't have to be called if there's a >=pageblock_order
> page already available.
> 
Maybe you are right and this change is just based on logical analyses.

Presently, only in direct compaction, we increase the compaction priority,
and ignore suitable at MIN_COMPACT_PRIORITY. I have a silly question, can
we do the similar thing in kcompactd? maybe by doing most work in kcompactd,
we can get better perf of slow path.

Thanks
Yisheng Xie

>> Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
>> ---
>>  mm/compaction.c | 6 +++---
>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index 0fdfde0..4bf2a5d 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
>>  static bool suitable_migration_target(struct compact_control *cc,
>>  							struct page *page)
>>  {
>> -	if (cc->ignore_block_suitable)
>> -		return true;
>> -
>>  	/* If the page is a large free page, then disallow migration */
>>  	if (PageBuddy(page)) {
>>  		/*
>> @@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>>  			return false;
>>  	}
>>  
>> +	if (cc->ignore_block_suitable)
>> +		return true;
>> +
>>  	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>>  	if (migrate_async_suitable(get_pageblock_migratetype(page)))
>>  		return true;
>>
> 
> 
> .
> 

--
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] mm/compaction: ignore block suitable after check large free page

[Vlastimil Babka]

On 03/10/2017 10:53 AM, Yisheng Xie wrote:
> Hi Vlastimil,
> 
> Thanks for comment.
> On 2017/3/10 15:30, Vlastimil Babka wrote:
>> On 03/10/2017 05:20 AM, Yisheng Xie wrote:
>>> If the migrate target is a large free page and we ignore suitable,
>>> it may not good for defrag. So move the ignore block suitable after
>>> check large free page.
>>
>> Right. But in practice I expect close to no impact, because direct
>> compaction shouldn't have to be called if there's a >=pageblock_order
>> page already available.
>>
> Maybe you are right and this change is just based on logical analyses.

I'm not opposing the change, it might be better for future-proofing the
function, just pointing out that it most likely won't have any visible
effect right now.

> Presently, only in direct compaction, we increase the compaction priority,
> and ignore suitable at MIN_COMPACT_PRIORITY. I have a silly question, can
> we do the similar thing in kcompactd? maybe by doing most work in kcompactd,
> we can get better perf of slow path.

That would need a very good evaluation at the very least. Migrating
pages into pageblocks other than movable ones brings the danger of later
unmovable/reclaimable allocations having to fallback to movable
pageblocks and causing permanent fragmentation. For direct compaction we
decided that it's better to risk permanent fragmentation than a
premature OOM, but for kcompactd there doesn't seem to be such
compelling reason.

> Thanks
> Yisheng Xie
> 
>>> Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
>>> ---
>>>  mm/compaction.c | 6 +++---
>>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>> index 0fdfde0..4bf2a5d 100644
>>> --- a/mm/compaction.c
>>> +++ b/mm/compaction.c
>>> @@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
>>>  static bool suitable_migration_target(struct compact_control *cc,
>>>  							struct page *page)
>>>  {
>>> -	if (cc->ignore_block_suitable)
>>> -		return true;
>>> -
>>>  	/* If the page is a large free page, then disallow migration */
>>>  	if (PageBuddy(page)) {
>>>  		/*
>>> @@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>>>  			return false;
>>>  	}
>>>  
>>> +	if (cc->ignore_block_suitable)
>>> +		return true;
>>> +
>>>  	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>>>  	if (migrate_async_suitable(get_pageblock_migratetype(page)))
>>>  		return true;
>>>
>>
>>
>> .
>>
> 

Re: [RFC] mm/compaction: ignore block suitable after check large free page

[Vlastimil Babka]

On 03/10/2017 10:53 AM, Yisheng Xie wrote:
> Hi Vlastimil,
> 
> Thanks for comment.
> On 2017/3/10 15:30, Vlastimil Babka wrote:
>> On 03/10/2017 05:20 AM, Yisheng Xie wrote:
>>> If the migrate target is a large free page and we ignore suitable,
>>> it may not good for defrag. So move the ignore block suitable after
>>> check large free page.
>>
>> Right. But in practice I expect close to no impact, because direct
>> compaction shouldn't have to be called if there's a >=pageblock_order
>> page already available.
>>
> Maybe you are right and this change is just based on logical analyses.

I'm not opposing the change, it might be better for future-proofing the
function, just pointing out that it most likely won't have any visible
effect right now.

> Presently, only in direct compaction, we increase the compaction priority,
> and ignore suitable at MIN_COMPACT_PRIORITY. I have a silly question, can
> we do the similar thing in kcompactd? maybe by doing most work in kcompactd,
> we can get better perf of slow path.

That would need a very good evaluation at the very least. Migrating
pages into pageblocks other than movable ones brings the danger of later
unmovable/reclaimable allocations having to fallback to movable
pageblocks and causing permanent fragmentation. For direct compaction we
decided that it's better to risk permanent fragmentation than a
premature OOM, but for kcompactd there doesn't seem to be such
compelling reason.

> Thanks
> Yisheng Xie
> 
>>> Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
>>> ---
>>>  mm/compaction.c | 6 +++---
>>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>> index 0fdfde0..4bf2a5d 100644
>>> --- a/mm/compaction.c
>>> +++ b/mm/compaction.c
>>> @@ -991,9 +991,6 @@ static bool too_many_isolated(struct zone *zone)
>>>  static bool suitable_migration_target(struct compact_control *cc,
>>>  							struct page *page)
>>>  {
>>> -	if (cc->ignore_block_suitable)
>>> -		return true;
>>> -
>>>  	/* If the page is a large free page, then disallow migration */
>>>  	if (PageBuddy(page)) {
>>>  		/*
>>> @@ -1005,6 +1002,9 @@ static bool suitable_migration_target(struct compact_control *cc,
>>>  			return false;
>>>  	}
>>>  
>>> +	if (cc->ignore_block_suitable)
>>> +		return true;
>>> +
>>>  	/* If the block is MIGRATE_MOVABLE or MIGRATE_CMA, allow migration */
>>>  	if (migrate_async_suitable(get_pageblock_migratetype(page)))
>>>  		return true;
>>>
>>
>>
>> .
>>
> 

--
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] mm/compaction: ignore block suitable after check large free page

[Yisheng Xie]

Hi Vlastimil,

Thanks for comment.
On 2017/3/13 17:51, Vlastimil Babka wrote:
> On 03/10/2017 10:53 AM, Yisheng Xie wrote:
>> Hi Vlastimil,
>>
>> Thanks for comment.
>> On 2017/3/10 15:30, Vlastimil Babka wrote:
>>> On 03/10/2017 05:20 AM, Yisheng Xie wrote:
>>>> If the migrate target is a large free page and we ignore suitable,
>>>> it may not good for defrag. So move the ignore block suitable after
>>>> check large free page.
>>>
>>> Right. But in practice I expect close to no impact, because direct
>>> compaction shouldn't have to be called if there's a >=pageblock_order
>>> page already available.
>>>
>> Maybe you are right and this change is just based on logical analyses.
> 
> I'm not opposing the change, it might be better for future-proofing the
> function, just pointing out that it most likely won't have any visible
> effect right now.
Get it, maybe I should put these in the change log :)

> 
>> Presently, only in direct compaction, we increase the compaction priority,
>> and ignore suitable at MIN_COMPACT_PRIORITY. I have a silly question, can
>> we do the similar thing in kcompactd? maybe by doing most work in kcompactd,
>> we can get better perf of slow path.
> 
> That would need a very good evaluation at the very least. Migrating
> pages into pageblocks other than movable ones brings the danger of later
> unmovable/reclaimable allocations having to fallback to movable
> pageblocks and causing permanent fragmentation. For direct compaction we
> decided that it's better to risk permanent fragmentation than a
> premature OOM, but for kcompactd there doesn't seem to be such
> compelling reason.
Thanks for kindly explain.

> 
>> Thanks
>> Yisheng Xie

Re: [RFC] mm/compaction: ignore block suitable after check large free page

[Yisheng Xie]

Hi Vlastimil,

Thanks for comment.
On 2017/3/13 17:51, Vlastimil Babka wrote:
> On 03/10/2017 10:53 AM, Yisheng Xie wrote:
>> Hi Vlastimil,
>>
>> Thanks for comment.
>> On 2017/3/10 15:30, Vlastimil Babka wrote:
>>> On 03/10/2017 05:20 AM, Yisheng Xie wrote:
>>>> If the migrate target is a large free page and we ignore suitable,
>>>> it may not good for defrag. So move the ignore block suitable after
>>>> check large free page.
>>>
>>> Right. But in practice I expect close to no impact, because direct
>>> compaction shouldn't have to be called if there's a >=pageblock_order
>>> page already available.
>>>
>> Maybe you are right and this change is just based on logical analyses.
> 
> I'm not opposing the change, it might be better for future-proofing the
> function, just pointing out that it most likely won't have any visible
> effect right now.
Get it, maybe I should put these in the change log :)

> 
>> Presently, only in direct compaction, we increase the compaction priority,
>> and ignore suitable at MIN_COMPACT_PRIORITY. I have a silly question, can
>> we do the similar thing in kcompactd? maybe by doing most work in kcompactd,
>> we can get better perf of slow path.
> 
> That would need a very good evaluation at the very least. Migrating
> pages into pageblocks other than movable ones brings the danger of later
> unmovable/reclaimable allocations having to fallback to movable
> pageblocks and causing permanent fragmentation. For direct compaction we
> decided that it's better to risk permanent fragmentation than a
> premature OOM, but for kcompactd there doesn't seem to be such
> compelling reason.
Thanks for kindly explain.

> 
>> Thanks
>> Yisheng Xie


--
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>