[PATCH RESEND 0/2] KSM: use linear_page_index in collect_procs_ksm()

[PATCH RESEND 0/2] KSM: use linear_page_index in collect_procs_ksm()

[xu.xin16]

From 65241f800da589be1981181d1bc3efd69a7d0385 Mon Sep 17 00:00:00 2001
From: xu xin <xu.xin16@zte.com.cn>
Date: Wed, 8 Jul 2026 11:13:47 +0800
Subject: [PATCH 0/2] KSM: use linear_page_index in collect_procs_ksm()

In collect_procs_ksm() which is used to collect processes when the error
hit an ksm page, there is the same issue with rmap_walk_ksm (see the
previous discussion at [1]). So we apply the similar logic changes to
the collect_procs_ksm().

The patch [1/2] move the initializaion of addr from the position inside
loop to the position before the loop, since the variable will not change
in the loop.

The patch [2/2] optimize collect_procs_ksm by passing a suitable page offset
range to the anon_vma_interval_tree_foreach loop to reduce ineffective
checks.

[1] https://lore.kernel.org/all/20260703162253688u8Str9eFLR8TGCmo7nIOF@zte.com.cn/

xu xin (2):
  mm/ksm: Initialize the addr only once in collect_procs_ksm
  mm/ksm: Use precise linear_page_index instead of the whole address
    space

 mm/ksm.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

-- 
2.25.1

[PATCH 1/2] mm/ksm: Initialize the addr only once in collect_procs_ksm

[xu.xin16]

From: xu xin <xu.xin16@zte.com.cn>

Similar with 318d87b8fa7 ("ksm: initialize the addr only once in rmap_walk_ksm"),
Only initialize the addr only once in rmap_walk_ksm because the addr variable
doesn't change across iterations.

Signed-off-by: Xu Xin <xu.xin16@zte.com>
---
 mm/ksm.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mm/ksm.c b/mm/ksm.c
index c1dcc3cb4831..337eb04a9340 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -3289,7 +3289,7 @@ void collect_procs_ksm(const struct folio *folio, const struct page *page,
 		rcu_read_lock();
 		for_each_process(tsk) {
 			struct anon_vma_chain *vmac;
-			unsigned long addr;
+			const unsigned long addr = rmap_item->address & PAGE_MASK;
 			struct task_struct *t =
 				task_early_kill(tsk, force_early);
 			if (!t)
@@ -3299,7 +3299,6 @@ void collect_procs_ksm(const struct folio *folio, const struct page *page,
 			{
 				vma = vmac->vma;
 				if (vma->vm_mm == t->mm) {
-					addr = rmap_item->address & PAGE_MASK;
 					add_to_kill_ksm(t, page, vma, to_kill,
 							addr);
 				}
-- 
2.25.1

Re: [PATCH 1/2] mm/ksm: Initialize the addr only once in collect_procs_ksm

[David Hildenbrand (Arm)]

On 7/8/26 06:19, xu.xin16@zte.com.cn wrote:
> From: xu xin <xu.xin16@zte.com.cn>
> 
> Similar with 318d87b8fa7 ("ksm: initialize the addr only once in rmap_walk_ksm"),
> Only initialize the addr only once in rmap_walk_ksm because the addr variable
> doesn't change across iterations.
> 
> Signed-off-by: Xu Xin <xu.xin16@zte.com>
> ---

Acked-by: David Hildenbrand (Arm) <david@kernel.org>

-- 
Cheers,

David

Re: [PATCH 1/2] mm/ksm: Initialize the addr only once in collect_procs_ksm

[David Hildenbrand (Arm)]

On 7/8/26 06:19, xu.xin16@zte.com.cn wrote:
> From: xu xin <xu.xin16@zte.com.cn>
> 
> Similar with 318d87b8fa7 ("ksm: initialize the addr only once in rmap_walk_ksm"),

checkpatch.pl should complain here

"Similar as commit 318d87b8fa7 ("ksm: initialize the addr only once in
rmap_walk_ksm")," ...

> Only initialize the addr only once in rmap_walk_ksm because the addr variable

s/Only/only/, but you une one "only" too much here.

> doesn't change across iterations.
> 
> Signed-off-by: Xu Xin <xu.xin16@zte.com>
> ---
>  mm/ksm.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/mm/ksm.c b/mm/ksm.c
> index c1dcc3cb4831..337eb04a9340 100644
> --- a/mm/ksm.c
> +++ b/mm/ksm.c
> @@ -3289,7 +3289,7 @@ void collect_procs_ksm(const struct folio *folio, const struct page *page,
>  		rcu_read_lock();
>  		for_each_process(tsk) {
>  			struct anon_vma_chain *vmac;
> -			unsigned long addr;
> +			const unsigned long addr = rmap_item->address & PAGE_MASK;
>  			struct task_struct *t =
>  				task_early_kill(tsk, force_early);
>  			if (!t)
> @@ -3299,7 +3299,6 @@ void collect_procs_ksm(const struct folio *folio, const struct page *page,
>  			{
>  				vma = vmac->vma;
>  				if (vma->vm_mm == t->mm) {
> -					addr = rmap_item->address & PAGE_MASK;
>  					add_to_kill_ksm(t, page, vma, to_kill,
>  							addr);
>  				}


-- 
Cheers,

David

[PATCH RESEND 2/2] mm/ksm: Use precise linear_page_index instead of the whole address space

[xu.xin16]

From: xu xin <xu.xin16@zte.com.cn>

Since an earlier commit introduced the linear_page_index to struct
ksm_rmap_item[1], allowing for optimizing the RMAP walk, we can use it
to locate more precisely all processes that hit the KSM page, which will
decrease a lot of unvalid iterations.

Link: https://lore.kernel.org/all/20260703162357853iIa-RP7if9hRlAIuTh5La@zte.com.cn/ [1]

Signed-off-by: Xu Xin <xu.xin16@zte.com.cn>
---
 mm/ksm.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/mm/ksm.c b/mm/ksm.c
index 337eb04a9340..2791ce5bd44b 100644
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -3290,12 +3290,13 @@ void collect_procs_ksm(const struct folio *folio, const struct page *page,
 		for_each_process(tsk) {
 			struct anon_vma_chain *vmac;
 			const unsigned long addr = rmap_item->address & PAGE_MASK;
+			const unsigned long index = rmap_item->linear_page_index;
 			struct task_struct *t =
 				task_early_kill(tsk, force_early);
 			if (!t)
 				continue;
-			anon_vma_interval_tree_foreach(vmac, &av->rb_root, 0,
-						       ULONG_MAX)
+			anon_vma_interval_tree_foreach(vmac, &av->rb_root, index,
+						       index)
 			{
 				vma = vmac->vma;
 				if (vma->vm_mm == t->mm) {
-- 
2.25.1

Re: [PATCH RESEND 2/2] mm/ksm: Use precise linear_page_index instead of the whole address space

[David Hildenbrand (Arm)]

On 7/8/26 06:22, xu.xin16@zte.com.cn wrote:
> From: xu xin <xu.xin16@zte.com.cn>
> 
> Since an earlier commit introduced the linear_page_index to struct
> ksm_rmap_item[1], allowing for optimizing the RMAP walk, we can use it
> to locate more precisely all processes that hit the KSM page, which will
> decrease a lot of unvalid iterations.

It's awkward to reference a commit by a mailing list link. The other patch is
not upstream yet. Best to wait for that to go in first, so you can properly
reference the commit.

-- 
Cheers,

David

Re: [PATCH RESEND 2/2] mm/ksm: Use precise linear_page_index instead of the whole address space

[xu.xin16]

> > From: xu xin <xu.xin16@zte.com.cn>
> > 
> > Since an earlier commit introduced the linear_page_index to struct
> > ksm_rmap_item[1], allowing for optimizing the RMAP walk, we can use it
> > to locate more precisely all processes that hit the KSM page, which will
> > decrease a lot of unvalid iterations.
> 
> It's awkward to reference a commit by a mailing list link. The other patch is
> not upstream yet. Best to wait for that to go in first, so you can properly
> reference the commit.

Ok, I'm going to resend this after that referenced patch is upstream.

Re: [PATCH RESEND 2/2] mm/ksm: Use precise linear_page_index instead of the whole address space

[David Hildenbrand (Arm)]

On 7/9/26 05:10, xu.xin16@zte.com.cn wrote:
>>> From: xu xin <xu.xin16@zte.com.cn>
>>>
>>> Since an earlier commit introduced the linear_page_index to struct
>>> ksm_rmap_item[1], allowing for optimizing the RMAP walk, we can use it
>>> to locate more precisely all processes that hit the KSM page, which will
>>> decrease a lot of unvalid iterations.
>>
>> It's awkward to reference a commit by a mailing list link. The other patch is
>> not upstream yet. Best to wait for that to go in first, so you can properly
>> reference the commit.
> 
> Ok, I'm going to resend this after that referenced patch is upstream.

Alternatively, just mention that we now have linear_page_index available that we
can use here, without referring to the other patch/commit.

That would probably be cleanest.

-- 
Cheers,

David