Re: [PATCH v1 4/5] devres: Simplify devm_percpu_match() implementation

[quic_zijuhu]

On 7/4/2024 9:17 PM, quic_zijuhu wrote:
> On 7/4/2024 6:34 PM, Greg KH wrote:
>> On Tue, Jul 02, 2024 at 10:51:53PM +0800, Zijun Hu wrote:
>>> Simplify devm_percpu_match() implementation by removing redundant
>>> conversions.
>>>
>>> Signed-off-by: Zijun Hu <quic_zijuhu@xxxxxxxxxxx>
>>> ---
>>> Previous discussion link:
>>> https://lore.kernel.org/lkml/1719496036-24642-1-git-send-email-quic_zijuhu@xxxxxxxxxxx/
>>>
>>> Changes since the original one:
>>>  - Select the simplier solution
>>>
>>>  drivers/base/devres.c | 5 ++---
>>>  1 file changed, 2 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/base/devres.c b/drivers/base/devres.c
>>> index e9b0d94aeabd..2ad6dacb3472 100644
>>> --- a/drivers/base/devres.c
>>> +++ b/drivers/base/devres.c
>>> @@ -1176,9 +1176,8 @@ static void devm_percpu_release(struct device *dev, void *pdata)
>>>  
>>>  static int devm_percpu_match(struct device *dev, void *data, void *p)
>>>  {
>>> -	struct devres *devr = container_of(data, struct devres, data);
>>> -
>>> -	return *(void **)devr->data == p;
>>> +	/* @data is already and must be (void *)devr->data */
>>> +	return *(void **)data == p;
>>
>> The compiler output should be identical here, right?  And container_of()
> yes, you are right.
>> enforces the placement so I'd prefer the original here as a comment
>> isn't going to enforce anything :)
>>
> i would like to show 2 different points related to the comments below:
> 
> 1) the comments is applicable for *ALL* kinds of devres match()s, and is
> *NOT* specific to devm_percpu_match() and also does *NOT* enforce anything.
> 
> may i remove the comments?
> 
> 2) the original implementation is *NOT* normative as explained below:
> include/linux/device.h:
> typedef int (*dr_match_t)(struct device *dev, void *res, void *match_data);
> void *devres_find(struct device *dev, dr_release_t release,
> 		  dr_match_t match, void *match_data);
> 
> devres API users maybe need to write their match() functions, for
> example, user of API devres_find(), but struct devres is a devres
> internal implementation defined within devres.c and is *NOT* exposed to
> API user by device.h, so API user should not use struct devres when
> implement their devres match() normally.
> but original devm_percpu_match() uses the struct devres.
> 

thanks for your code review.
may i know which of below options is your final preference?

A) remain current kernel design, namely, no changes.

B) changes shown by this [PATCH v1 4/5]

C) below changes we ever discussed by below link
https://lore.kernel.org/all/1719496036-24642-1-git-send-email-quic_zijuhu@xxxxxxxxxxx/

 static int devm_percpu_match(struct device *dev, void *data, void *p)
 {
-	struct devres *devr = container_of(data, struct devres, data);
+	void __percpu *ptr = *(void __percpu **)data;

-	return *(void **)devr->data == p;
+	return ptr == (void __percpu *)p;
 }

>> thanks,
>>
>> greg k-h
>