[PATCH] nfsd: Simplify the allocation of slab caches in nfsd_drc_slab_create

[PATCH] nfsd: Simplify the allocation of slab caches in nfsd_drc_slab_create

[Kunwu Chan]

Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.
Make the code cleaner and more readable.

Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
---
 fs/nfsd/nfscache.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/fs/nfsd/nfscache.c b/fs/nfsd/nfscache.c
index 5c1a4a0aa605..64ce0cc22197 100644
--- a/fs/nfsd/nfscache.c
+++ b/fs/nfsd/nfscache.c
@@ -166,8 +166,7 @@ nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct nfsd_cacherep *rp,
 
 int nfsd_drc_slab_create(void)
 {
-	drc_slab = kmem_cache_create("nfsd_drc",
-				sizeof(struct nfsd_cacherep), 0, 0, NULL);
+	drc_slab = KMEM_CACHE(nfsd_cacherep, 0);
 	return drc_slab ? 0: -ENOMEM;
 }
 
-- 
2.39.2

Re: [PATCH] nfsd: Simplify the allocation of slab caches in nfsd_drc_slab_create

[Benjamin Coddington]

On 1 Feb 2024, at 3:19, Kunwu Chan wrote:

> Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
> to simplify the creation of SLAB caches.
> Make the code cleaner and more readable.
>
> Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
> ---
>  fs/nfsd/nfscache.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/fs/nfsd/nfscache.c b/fs/nfsd/nfscache.c
> index 5c1a4a0aa605..64ce0cc22197 100644
> --- a/fs/nfsd/nfscache.c
> +++ b/fs/nfsd/nfscache.c
> @@ -166,8 +166,7 @@ nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct nfsd_cacherep *rp,
>
>  int nfsd_drc_slab_create(void)
>  {
> -	drc_slab = kmem_cache_create("nfsd_drc",
> -				sizeof(struct nfsd_cacherep), 0, 0, NULL);
> +	drc_slab = KMEM_CACHE(nfsd_cacherep, 0);
>  	return drc_slab ? 0: -ENOMEM;
>  }
>
> -- 
> 2.39.2

I don't agree that the code is cleaner or more readable like this.  I really
dislike having to parse through the extra "simplification" to see what's
actually being called and sent.

Just my .02 worth.

Ben

Re: [PATCH] nfsd: Simplify the allocation of slab caches in nfsd_drc_slab_create

[Jeff Layton]

On Fri, 2024-02-02 at 09:13 -0500, Benjamin Coddington wrote:
> On 1 Feb 2024, at 3:19, Kunwu Chan wrote:
> 
> > Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
> > to simplify the creation of SLAB caches.
> > Make the code cleaner and more readable.
> > 
> > Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
> > ---
> >  fs/nfsd/nfscache.c | 3 +--
> >  1 file changed, 1 insertion(+), 2 deletions(-)
> > 
> > diff --git a/fs/nfsd/nfscache.c b/fs/nfsd/nfscache.c
> > index 5c1a4a0aa605..64ce0cc22197 100644
> > --- a/fs/nfsd/nfscache.c
> > +++ b/fs/nfsd/nfscache.c
> > @@ -166,8 +166,7 @@ nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct nfsd_cacherep *rp,
> > 
> >  int nfsd_drc_slab_create(void)
> >  {
> > -	drc_slab = kmem_cache_create("nfsd_drc",
> > -				sizeof(struct nfsd_cacherep), 0, 0, NULL);
> > +	drc_slab = KMEM_CACHE(nfsd_cacherep, 0);
> >  	return drc_slab ? 0: -ENOMEM;
> >  }
> > 
> > -- 
> > 2.39.2
> 
> I don't agree that the code is cleaner or more readable like this.  I really
> dislike having to parse through the extra "simplification" to see what's
> actually being called and sent.
> 
> Just my .02 worth.
> 
> Ben
> 

This will also result in a behavioral change. The "nfsd_drc" string is
lost with the above macro and (I think) the new name will be
"nfsd_cacherep". I'm not necessarily opposed to that, as I don't think
anything depends on the old name, but it should at least be noted in the
changelog.

-- 
Jeff Layton <jlayton@kernel.org>

Re: [PATCH] nfsd: Simplify the allocation of slab caches in nfsd_drc_slab_create

[NeilBrown]

On Sat, 03 Feb 2024, Benjamin Coddington wrote:
> On 1 Feb 2024, at 3:19, Kunwu Chan wrote:
> 
> > Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
> > to simplify the creation of SLAB caches.
> > Make the code cleaner and more readable.
> >
> > Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
> > ---
> >  fs/nfsd/nfscache.c | 3 +--
> >  1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/fs/nfsd/nfscache.c b/fs/nfsd/nfscache.c
> > index 5c1a4a0aa605..64ce0cc22197 100644
> > --- a/fs/nfsd/nfscache.c
> > +++ b/fs/nfsd/nfscache.c
> > @@ -166,8 +166,7 @@ nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct nfsd_cacherep *rp,
> >
> >  int nfsd_drc_slab_create(void)
> >  {
> > -	drc_slab = kmem_cache_create("nfsd_drc",
> > -				sizeof(struct nfsd_cacherep), 0, 0, NULL);
> > +	drc_slab = KMEM_CACHE(nfsd_cacherep, 0);
> >  	return drc_slab ? 0: -ENOMEM;
> >  }
> >
> > -- 
> > 2.39.2
> 
> I don't agree that the code is cleaner or more readable like this.  I really
> dislike having to parse through the extra "simplification" to see what's
> actually being called and sent.
> 
> Just my .02 worth.
> 

In general I agree that wrappers like this can hinder as much as they
help - if not more.

In this particular case it doesn't seem to bother me.  This is probably
because it is only used in initialisation code and I don't look at that
nearly as much as code that uses the initialised things.
Initialisation/cleanup code often has a lot of boilerplate which can
make it look messy.  Reducing that, which I think this patch helps with,
can be a good thing.

So I agree that we should be cautious about using (or creating) new
wrapper macros, but in this case I am mildly in favour.

Thanks,
NeilBrown

Re: [PATCH] nfsd: Simplify the allocation of slab caches in nfsd_drc_slab_create

[Kunwu Chan]

Thank you to all the guys who responded to my emails.

On 2024/2/2 22:24, Jeff Layton wrote:
> On Fri, 2024-02-02 at 09:13 -0500, Benjamin Coddington wrote:
>> On 1 Feb 2024, at 3:19, Kunwu Chan wrote:
>>
>>> Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
>>> to simplify the creation of SLAB caches.
>>> Make the code cleaner and more readable.
>>>
>>> Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
>>> ---
>>>   fs/nfsd/nfscache.c | 3 +--
>>>   1 file changed, 1 insertion(+), 2 deletions(-)
>>>
>>> diff --git a/fs/nfsd/nfscache.c b/fs/nfsd/nfscache.c
>>> index 5c1a4a0aa605..64ce0cc22197 100644
>>> --- a/fs/nfsd/nfscache.c
>>> +++ b/fs/nfsd/nfscache.c
>>> @@ -166,8 +166,7 @@ nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct nfsd_cacherep *rp,
>>>
>>>   int nfsd_drc_slab_create(void)
>>>   {
>>> -	drc_slab = kmem_cache_create("nfsd_drc",
>>> -				sizeof(struct nfsd_cacherep), 0, 0, NULL);
>>> +	drc_slab = KMEM_CACHE(nfsd_cacherep, 0);
>>>   	return drc_slab ? 0: -ENOMEM;
>>>   }
>>>
>>> -- 
>>> 2.39.2
>>
>> I don't agree that the code is cleaner or more readable like this.  I really
>> dislike having to parse through the extra "simplification" to see what's
>> actually being called and sent.
>>
>> Just my .02 worth.
>>
>> Ben
>>
> 
Everyone has a different opinion. From newcomers like me, a simple code 
is more important than checking all the args of a call function to 
understand what it does.
Too many default arguments can cost us a lot of time that could be spent 
understanding the main logic of the module code, rather than wasting it 
on a single line of calls.

> This will also result in a behavioral change. The "nfsd_drc" string is
> lost with the above macro and (I think) the new name will be
> "nfsd_cacherep". I'm not necessarily opposed to that, as I don't think
> anything depends on the old name, but it should at least be noted in the
> changelog.
Thanks i'll update my v2 patch with a new commit msg to show the name 
change.

> 
-- 
Thanks,
   Kunwu