Skip to contentSkip to conversation

mm: remove lru_add_drain_all in alloc_contig_range

PATCH Superseded Minchan Kim 5 participants 9 messages Feb 9, 2021 – Feb 10, 2021

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm: remove lru_add_drain_all in alloc_contig_range

__alloc_contig_migrate_range already has lru_add_drain_all call via migrate_prep. It's necessary to move LRU taget pages into LRU list to be able to isolated. However, lru_add_drain_all call after __alloc_contig_migrate_range is called is pointless.

This patch removes it.

---
 mm/page_alloc.c | 2 --
 1 file changed, 2 deletions(-)
mm/page_alloc.c +0 -2
index 6446778cbc6b..f8fbee73dd6d 100644
@@ -8603,8 +8603,6 @@ int alloc_contig_range(unsigned long start, unsigned long end,
 	 * isolated thus they won't get removed from buddy.
 	 */
 
-	lru_add_drain_all();
-
 	order = 0;
 	outer_start = start;
 	while (!PageBuddy(pfn_to_page(outer_start))) {
Signature · 1 line
2.30.0.478.g8a0d178c01-goog
#1 lore raw
Quoted 18 lines · Minchan Kim

__alloc_contig_migrate_range already has lru_add_drain_all call via migrate_prep. It's necessary to move LRU taget pages into LRU list to be able to isolated. However, lru_add_drain_all call after __alloc_contig_migrate_range is called is pointless.

This patch removes it.

---
  mm/page_alloc.c | 2 --
  1 file changed, 2 deletions(-)
mm/page_alloc.c +0 -0
index 6446778cbc6b..f8fbee73dd6d 100644

mm/page_alloc.c int alloc_contig_range(unsigned long start, unsigned long end, · line 8608

Quoted 8 lines
mm/page_alloc.c +0 -2
@@ -8603,8 +8603,6 @@ int alloc_contig_range(unsigned long start, unsigned long end,
  	 * isolated thus they won't get removed from buddy.
  	 */
 
-	lru_add_drain_all();
-
  	order = 0;
  	outer_start = start;
  	while (!PageBuddy(pfn_to_page(outer_start))) {

I was expecting some magical reason why this is still required but I am not able to find a compelling one. Maybe this is really some historical artifact.

Let's see if other people know why this call here still exists.

Signature · 3 lines
Thanks,

David / dhildenb
#2 lore raw
Quoted 31 lines · David Hildenbrand
Quoted 17 lines · Minchan Kim

__alloc_contig_migrate_range already has lru_add_drain_all call via migrate_prep. It's necessary to move LRU taget pages into LRU list to be able to isolated. However, lru_add_drain_all call after __alloc_contig_migrate_range is called is pointless.

This patch removes it.

---
  mm/page_alloc.c | 2 --
  1 file changed, 2 deletions(-)
mm/page_alloc.c +0 -0
index 6446778cbc6b..f8fbee73dd6d 100644

mm/page_alloc.c int alloc_contig_range(unsigned long start, unsigned long end, · line 8608

Quoted 8 lines
mm/page_alloc.c +0 -2
@@ -8603,8 +8603,6 @@ int alloc_contig_range(unsigned long start, unsigned long end,
  	 * isolated thus they won't get removed from buddy.
  	 */
-	lru_add_drain_all();
-
  	order = 0;
  	outer_start = start;
  	while (!PageBuddy(pfn_to_page(outer_start))) {
 

I was expecting some magical reason why this is still required but I am not able to find a compelling one. Maybe this is really some historical artifact.

Quoted 1 line

Let's see if other people know why this call here still exists.

I also stumbled upon this while working on adding hugetlb support for alloc_acontig_range [1].
I have to confess I puzzled me a bit.

I saw it going back to when the function was first introduced by

commit 041d3a8cdc18dc375a128d90bbb753949a81b1fb

Author: Michal Nazarewicz <[email protected]>
Date:   Thu Dec 29 13:09:50 2011 +0100

mm: page_alloc: introduce alloc_contig_range()

It does not make much sense to me. At this point our pages are free, so we do not care about LRU handling here.
But I might be missing something.

[1] https://lore.kernel.org/linux-mm/20210208103935.GA32103@linux/T/#md651fc6e73c656105179382f92f8b2d6073051d1

Signature · 2 lines
Oscar Salvador
SUSE L3
#3 lore raw
Quoted 17 lines · Oscar Salvador
Quoted 5 lines · David Hildenbrand

I was expecting some magical reason why this is still required but I am not able to find a compelling one. Maybe this is really some historical artifact.

Let's see if other people know why this call here still exists.

I also stumbled upon this while working on adding hugetlb support for alloc_acontig_range [1].
I have to confess I puzzled me a bit.

I saw it going back to when the function was first introduced by

commit 041d3a8cdc18dc375a128d90bbb753949a81b1fb

Author: Michal Nazarewicz <[email protected]>
Date:   Thu Dec 29 13:09:50 2011 +0100
Quoted 6 lines

mm: page_alloc: introduce alloc_contig_range()

It does not make much sense to me. At this point our pages are free, so we do not care about LRU handling here.
But I might be missing something.

AFAICS, at the time page migration used putback_lru_page() to release the migration source page. This would put the page on lru pvec even if it was in fact not mapped anywhere anymore, and only the drain would actually free it. Seems Minchan optimized this in 2016 by c6c919eb90e0 ("mm: use put_page() to free page instead of putback_lru_page()")

Quoted 3 lines

[1] https://lore.kernel.org/linux-mm/20210208103935.GA32103@linux/T/#md651fc6e73c656105179382f92f8b2d6073051d1

reviewer Acked-by
#4 lore raw
Quoted 25 lines · Vlastimil Babka
Quoted 17 lines · Oscar Salvador
Quoted 5 lines · David Hildenbrand

I was expecting some magical reason why this is still required but I am not able to find a compelling one. Maybe this is really some historical artifact.

Let's see if other people know why this call here still exists.

I also stumbled upon this while working on adding hugetlb support for alloc_acontig_range [1].
I have to confess I puzzled me a bit.

I saw it going back to when the function was first introduced by

commit 041d3a8cdc18dc375a128d90bbb753949a81b1fb

Author: Michal Nazarewicz <[email protected]>
Date:   Thu Dec 29 13:09:50 2011 +0100
Quoted 6 lines

mm: page_alloc: introduce alloc_contig_range()

It does not make much sense to me. At this point our pages are free, so we do not care about LRU handling here.
But I might be missing something.

Quoted 5 lines

AFAICS, at the time page migration used putback_lru_page() to release the migration source page. This would put the page on lru pvec even if it was in fact not mapped anywhere anymore, and only the drain would actually free it. Seems Minchan optimized this in 2016 by c6c919eb90e0 ("mm: use put_page() to free page instead of putback_lru_page()")

This would be a great addition to the changelog. Thanks a lot Vlastimil, you saved me from some archeology. With that mentioned feel free to add

Quoted 5 lines
Quoted 4 lines

[1] https://lore.kernel.org/linux-mm/20210208103935.GA32103@linux/T/#md651fc6e73c656105179382f92f8b2d6073051d1

Signature · 2 lines
Michal Hocko
SUSE Labs
#5 lore raw
In reply to Vlastimil Babka
Quoted 5 lines · Vlastimil Babka

AFAICS, at the time page migration used putback_lru_page() to release the migration source page. This would put the page on lru pvec even if it was in fact not mapped anywhere anymore, and only the drain would actually free it. Seems Minchan optimized this in 2016 by c6c919eb90e0 ("mm: use put_page() to free page instead of putback_lru_page()")

Ok, I see. Thanks for this valuable information Vlastimil.

Signature · 2 lines
Oscar Salvador
SUSE L3
#6 lore raw
In reply to Vlastimil Babka
Quoted 25 lines · Vlastimil Babka
Quoted 17 lines · Oscar Salvador
Quoted 5 lines · David Hildenbrand

I was expecting some magical reason why this is still required but I am not able to find a compelling one. Maybe this is really some historical artifact.

Let's see if other people know why this call here still exists.

I also stumbled upon this while working on adding hugetlb support for alloc_acontig_range [1].
I have to confess I puzzled me a bit.

I saw it going back to when the function was first introduced by

commit 041d3a8cdc18dc375a128d90bbb753949a81b1fb

Author: Michal Nazarewicz <[email protected]>
Date:   Thu Dec 29 13:09:50 2011 +0100
Quoted 6 lines

mm: page_alloc: introduce alloc_contig_range()

It does not make much sense to me. At this point our pages are free, so we do not care about LRU handling here.
But I might be missing something.

Quoted 7 lines

AFAICS, at the time page migration used putback_lru_page() to release the migration source page. This would put the page on lru pvec even if it was in fact not mapped anywhere anymore, and only the drain would actually free it. Seems Minchan optimized this in 2016 by c6c919eb90e0 ("mm: use put_page() to free page instead of putback_lru_page()")

Quoted 1 line

[1] https://lore.kernel.org/linux-mm/20210208103935.GA32103@linux/T/#md651fc6e73c656105179382f92f8b2d6073051d1

Thanks for digging history, Vlastimil!

Acked-by on v1 1/1
#7 lore raw
In reply to Minchan Kim
Quoted 8 lines · Minchan Kim

__alloc_contig_migrate_range already has lru_add_drain_all call via migrate_prep. It's necessary to move LRU taget pages into LRU list to be able to isolated. However, lru_add_drain_all call after __alloc_contig_migrate_range is called is pointless.

This patch removes it.

Quoted 18 lines
---
 mm/page_alloc.c | 2 --
 1 file changed, 2 deletions(-)
mm/page_alloc.c +0 -2
index 6446778cbc6b..f8fbee73dd6d 100644
@@ -8603,8 +8603,6 @@ int alloc_contig_range(unsigned long start, unsigned long end,
 	 * isolated thus they won't get removed from buddy.
 	 */
 
-	lru_add_drain_all();
-
 	order = 0;
 	outer_start = start;
 	while (!PageBuddy(pfn_to_page(outer_start))) {
Reviewed-by on v1 1/1
#8 lore raw
In reply to Minchan Kim
Quoted 8 lines · Minchan Kim

__alloc_contig_migrate_range already has lru_add_drain_all call via migrate_prep. It's necessary to move LRU taget pages into LRU list to be able to isolated. However, lru_add_drain_all call after __alloc_contig_migrate_range is called is pointless.

This patch removes it.

I agree with Michal that the changelog could benefit from Vastimil's clarification, so with that:

Quoted 21 lines
---
 mm/page_alloc.c | 2 --
 1 file changed, 2 deletions(-)
mm/page_alloc.c +0 -2
index 6446778cbc6b..f8fbee73dd6d 100644
@@ -8603,8 +8603,6 @@ int alloc_contig_range(unsigned long start, unsigned long end,
 	 * isolated thus they won't get removed from buddy.
 	 */
 
-	lru_add_drain_all();
-
 	order = 0;
 	outer_start = start;
 	while (!PageBuddy(pfn_to_page(outer_start))) {
Signature · 1 line
2.30.0.478.g8a0d178c01-goog
Signature · 2 lines
Oscar Salvador
SUSE L3