[PATCH] maple_tree: Remove redundant mte_to_node() in mte_dead_node()

[PATCH] maple_tree: Remove redundant mte_to_node() in mte_dead_node()

[I Hsin Cheng]

In mte_dead_node(), it already assign "node" as "mte_to_node(enode)" in
the first place, calling "mte_parent(enode)" will result in the same
"mte_to_node(enode)" again which is redundant.

Refactor mte_dead_node() and utilize ma_dead_node() to perform the
parent check without the redundant "mte_to_node()".

Signed-off-by: I Hsin Cheng <richard120310@gmail.com>
---
 lib/maple_tree.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/lib/maple_tree.c b/lib/maple_tree.c
index f7153ade1be5..362f85c62678 100644
--- a/lib/maple_tree.c
+++ b/lib/maple_tree.c
@@ -584,13 +584,10 @@ static __always_inline bool ma_dead_node(const struct maple_node *node)
  */
 static __always_inline bool mte_dead_node(const struct maple_enode *enode)
 {
-	struct maple_node *parent, *node;
+	struct maple_node *node;
 
 	node = mte_to_node(enode);
-	/* Do not reorder reads from the node prior to the parent check */
-	smp_rmb();
-	parent = mte_parent(enode);
-	return (parent == node);
+	return ma_dead_node(node);
 }
 
 /*
-- 
2.43.0

Re: [PATCH] maple_tree: Remove redundant mte_to_node() in mte_dead_node()

[Liam R. Howlett]

* I Hsin Cheng <richard120310@gmail.com> [250210 03:35]:

> In mte_dead_node(), it already assign "node" as "mte_to_node(enode)" in
> the first place, calling "mte_parent(enode)" will result in the same
> "mte_to_node(enode)" again which is redundant.

This is a very confusing way of saying "avoid calling mte_to_node() in
the mte_parent() call by using the ma_dead_node() instead."

In fact, the subject is wrong as well, since the mte_to_node() was
removed from the call path of mte_dead_node(), and not the function
itself.

> 
> Refactor mte_dead_node() and utilize ma_dead_node() to perform the
> parent check without the redundant "mte_to_node()".
> 
> Signed-off-by: I Hsin Cheng <richard120310@gmail.com>

The code looks right, but the subject and change log are not.  Please
respin the patch, something like this:

maple_tree: Use ma_dead_node() in mte_dead_node()

Using ma_dead_node() in mte_dead_node() avoids decoding the maple enode
for a second time to find the parent.

Feel free to change it as you'd like, but I couldn't follow what you
meant.

> ---
>  lib/maple_tree.c | 7 ++-----
>  1 file changed, 2 insertions(+), 5 deletions(-)
> 
> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> index f7153ade1be5..362f85c62678 100644
> --- a/lib/maple_tree.c
> +++ b/lib/maple_tree.c
> @@ -584,13 +584,10 @@ static __always_inline bool ma_dead_node(const struct maple_node *node)
>   */
>  static __always_inline bool mte_dead_node(const struct maple_enode *enode)
>  {
> -	struct maple_node *parent, *node;
> +	struct maple_node *node;
>  
>  	node = mte_to_node(enode);
> -	/* Do not reorder reads from the node prior to the parent check */
> -	smp_rmb();
> -	parent = mte_parent(enode);
> -	return (parent == node);
> +	return ma_dead_node(node);
>  }
>  
>  /*
> -- 
> 2.43.0
> 

Re: [PATCH] maple_tree: Remove redundant mte_to_node() in mte_dead_node()

[I Hsin Cheng]

On Mon, Feb 10, 2025 at 09:51:42AM -0500, Liam R. Howlett wrote:
> * I Hsin Cheng <richard120310@gmail.com> [250210 03:35]:
> 
> > In mte_dead_node(), it already assign "node" as "mte_to_node(enode)" in
> > the first place, calling "mte_parent(enode)" will result in the same
> > "mte_to_node(enode)" again which is redundant.
> 
> This is a very confusing way of saying "avoid calling mte_to_node() in
> the mte_parent() call by using the ma_dead_node() instead."
> 
> In fact, the subject is wrong as well, since the mte_to_node() was
> removed from the call path of mte_dead_node(), and not the function
> itself.
> 
> > 
> > Refactor mte_dead_node() and utilize ma_dead_node() to perform the
> > parent check without the redundant "mte_to_node()".
> > 
> > Signed-off-by: I Hsin Cheng <richard120310@gmail.com>
> 
> The code looks right, but the subject and change log are not.  Please
> respin the patch, something like this:
> 
> maple_tree: Use ma_dead_node() in mte_dead_node()
> 
> Using ma_dead_node() in mte_dead_node() avoids decoding the maple enode
> for a second time to find the parent.
> 
> Feel free to change it as you'd like, but I couldn't follow what you
> meant.
> 
> > ---
> >  lib/maple_tree.c | 7 ++-----
> >  1 file changed, 2 insertions(+), 5 deletions(-)
> > 
> > diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> > index f7153ade1be5..362f85c62678 100644
> > --- a/lib/maple_tree.c
> > +++ b/lib/maple_tree.c
> > @@ -584,13 +584,10 @@ static __always_inline bool ma_dead_node(const struct maple_node *node)
> >   */
> >  static __always_inline bool mte_dead_node(const struct maple_enode *enode)
> >  {
> > -	struct maple_node *parent, *node;
> > +	struct maple_node *node;
> >  
> >  	node = mte_to_node(enode);
> > -	/* Do not reorder reads from the node prior to the parent check */
> > -	smp_rmb();
> > -	parent = mte_parent(enode);
> > -	return (parent == node);
> > +	return ma_dead_node(node);
> >  }
> >  
> >  /*
> > -- 
> > 2.43.0
> >

Hello Liam,

Thanks for your kindly review!

> In fact, the subject is wrong as well, since the mte_to_node() was
> removed from the call path of mte_dead_node(), and not the function
> itself.

I see, I'll rephrase the whole commit and send a new patch later, what I
wrote is indeed too confusing. Thank you.

Best regards,
I Hsin Cheng