diff mbox series

[29/34] net: smc: fix opencoded find_and_set_bit() in smc_wr_tx_get_free_slot_index()

Message ID 20231118155105.25678-30-yury.norov@gmail.com
State New
Headers
Series biops: add atomig find_bit() operations |

Commit Message

Yury Norov Nov. 18, 2023, 3:51 p.m. UTC 
  The function opencodes find_and_set_bit() with a for_each() loop. Fix
it, and make the whole function a simple almost one-liner.

Signed-off-by: Yury Norov <yury.norov@gmail.com>
---
 net/smc/smc_wr.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

Comments

Alexandra Winter Nov. 20, 2023, 8:43 a.m. UTC | #1 
On 18.11.23 16:51, Yury Norov wrote:
> The function opencodes find_and_set_bit() with a for_each() loop. Fix
> it, and make the whole function a simple almost one-liner.
> 
> Signed-off-by: Yury Norov <yury.norov@gmail.com>
> ---
>  net/smc/smc_wr.c | 10 +++-------
>  1 file changed, 3 insertions(+), 7 deletions(-)
> 
> diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c
> index 0021065a600a..b6f0cfc52788 100644
> --- a/net/smc/smc_wr.c
> +++ b/net/smc/smc_wr.c
> @@ -170,15 +170,11 @@ void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
>  
>  static inline int smc_wr_tx_get_free_slot_index(struct smc_link *link, u32 *idx)
>  {
> -	*idx = link->wr_tx_cnt;
>  	if (!smc_link_sendable(link))
>  		return -ENOLINK;
> -	for_each_clear_bit(*idx, link->wr_tx_mask, link->wr_tx_cnt) {
> -		if (!test_and_set_bit(*idx, link->wr_tx_mask))
> -			return 0;
> -	}
> -	*idx = link->wr_tx_cnt;
> -	return -EBUSY;
> +
> +	*idx = find_and_set_bit(link->wr_tx_mask, link->wr_tx_cnt);
> +	return *idx < link->wr_tx_cnt ? 0 : -EBUSY;
>  }
>  
>  /**


My understanding is that you can omit the lines with
> -	*idx = link->wr_tx_cnt;
because they only apply to the error paths and you checked that the calling function
does not use the idx variable in the error cases. Do I understand this correct?

If so the removal of these 2 lines is not related to your change of using find_and_set_bit(),
do I understand that correctly?

If so, it may be worth mentioning that in the commit message.
Tony Lu Nov. 20, 2023, 9:56 a.m. UTC | #2 
The prefix tag and subject imply that it is a bugfix. I think, first, it
should be a new feature with net-next tag. Also please use net/smc as
prefix.

Thanks,
Tony Lu

On Sat, Nov 18, 2023 at 07:51:00AM -0800, Yury Norov wrote:
> The function opencodes find_and_set_bit() with a for_each() loop. Fix
> it, and make the whole function a simple almost one-liner.
> 
> Signed-off-by: Yury Norov <yury.norov@gmail.com>
> ---
>  net/smc/smc_wr.c | 10 +++-------
>  1 file changed, 3 insertions(+), 7 deletions(-)
> 
> diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c
> index 0021065a600a..b6f0cfc52788 100644
> --- a/net/smc/smc_wr.c
> +++ b/net/smc/smc_wr.c
> @@ -170,15 +170,11 @@ void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
>  
>  static inline int smc_wr_tx_get_free_slot_index(struct smc_link *link, u32 *idx)
>  {
> -	*idx = link->wr_tx_cnt;
>  	if (!smc_link_sendable(link))
>  		return -ENOLINK;
> -	for_each_clear_bit(*idx, link->wr_tx_mask, link->wr_tx_cnt) {
> -		if (!test_and_set_bit(*idx, link->wr_tx_mask))
> -			return 0;
> -	}
> -	*idx = link->wr_tx_cnt;
> -	return -EBUSY;
> +
> +	*idx = find_and_set_bit(link->wr_tx_mask, link->wr_tx_cnt);
> +	return *idx < link->wr_tx_cnt ? 0 : -EBUSY;
>  }
>  
>  /**
> -- 
> 2.39.2
Yury Norov Nov. 21, 2023, 1:41 p.m. UTC | #3 
On Mon, Nov 20, 2023 at 09:43:54AM +0100, Alexandra Winter wrote:
> 
> 
> On 18.11.23 16:51, Yury Norov wrote:
> > The function opencodes find_and_set_bit() with a for_each() loop. Fix
> > it, and make the whole function a simple almost one-liner.
> > 
> > Signed-off-by: Yury Norov <yury.norov@gmail.com>
> > ---
> >  net/smc/smc_wr.c | 10 +++-------
> >  1 file changed, 3 insertions(+), 7 deletions(-)
> > 
> > diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c
> > index 0021065a600a..b6f0cfc52788 100644
> > --- a/net/smc/smc_wr.c
> > +++ b/net/smc/smc_wr.c
> > @@ -170,15 +170,11 @@ void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
> >  
> >  static inline int smc_wr_tx_get_free_slot_index(struct smc_link *link, u32 *idx)
> >  {
> > -	*idx = link->wr_tx_cnt;
> >  	if (!smc_link_sendable(link))
> >  		return -ENOLINK;
> > -	for_each_clear_bit(*idx, link->wr_tx_mask, link->wr_tx_cnt) {
> > -		if (!test_and_set_bit(*idx, link->wr_tx_mask))
> > -			return 0;
> > -	}
> > -	*idx = link->wr_tx_cnt;
> > -	return -EBUSY;
> > +
> > +	*idx = find_and_set_bit(link->wr_tx_mask, link->wr_tx_cnt);
> > +	return *idx < link->wr_tx_cnt ? 0 : -EBUSY;
> >  }
> >  
> >  /**
> 
> 
> My understanding is that you can omit the lines with
> > -	*idx = link->wr_tx_cnt;
> because they only apply to the error paths and you checked that the calling function
> does not use the idx variable in the error cases. Do I understand this correct?
> 
> If so the removal of these 2 lines is not related to your change of using find_and_set_bit(),
> do I understand that correctly?
> 
> If so, it may be worth mentioning that in the commit message.

I'll add:

        If find_and_set_bit() doesn't acquire a bit, it returns
        ->wr_tx_cnt, and so explicit initialization of *idx with
        the same value is unneeded.

Makes sense?
Alexandra Winter Nov. 21, 2023, 3:39 p.m. UTC | #4 
On 21.11.23 14:41, Yury Norov wrote:
> On Mon, Nov 20, 2023 at 09:43:54AM +0100, Alexandra Winter wrote:
>>
>>
>> On 18.11.23 16:51, Yury Norov wrote:
>>> The function opencodes find_and_set_bit() with a for_each() loop. Fix
>>> it, and make the whole function a simple almost one-liner.
>>>
>>> Signed-off-by: Yury Norov <yury.norov@gmail.com>
>>> ---
>>>  net/smc/smc_wr.c | 10 +++-------
>>>  1 file changed, 3 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c
>>> index 0021065a600a..b6f0cfc52788 100644
>>> --- a/net/smc/smc_wr.c
>>> +++ b/net/smc/smc_wr.c
>>> @@ -170,15 +170,11 @@ void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
>>>  
>>>  static inline int smc_wr_tx_get_free_slot_index(struct smc_link *link, u32 *idx)
>>>  {
>>> -	*idx = link->wr_tx_cnt;
>>>  	if (!smc_link_sendable(link))
>>>  		return -ENOLINK;
>>> -	for_each_clear_bit(*idx, link->wr_tx_mask, link->wr_tx_cnt) {
>>> -		if (!test_and_set_bit(*idx, link->wr_tx_mask))
>>> -			return 0;
>>> -	}
>>> -	*idx = link->wr_tx_cnt;
>>> -	return -EBUSY;
>>> +
>>> +	*idx = find_and_set_bit(link->wr_tx_mask, link->wr_tx_cnt);
>>> +	return *idx < link->wr_tx_cnt ? 0 : -EBUSY;
>>>  }
>>>  
>>>  /**
>>
>>
>> My understanding is that you can omit the lines with
>>> -	*idx = link->wr_tx_cnt;
>> because they only apply to the error paths and you checked that the calling function
>> does not use the idx variable in the error cases. Do I understand this correct?
>>
>> If so the removal of these 2 lines is not related to your change of using find_and_set_bit(),
>> do I understand that correctly?
>>
>> If so, it may be worth mentioning that in the commit message.
> 
> I'll add:
> 
>         If find_and_set_bit() doesn't acquire a bit, it returns
>         ->wr_tx_cnt, and so explicit initialization of *idx with
>         the same value is unneeded.
> 
> Makes sense?
> 

Makes sense for the -EBUSY case, thank you. 
It does not explain that you also removed the line for the -ENOLINK case 
(which is ok, because the caller has also initialized it to link->wr_tx_cnt)
diff mbox series

Patch

diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c
index 0021065a600a..b6f0cfc52788 100644
--- a/net/smc/smc_wr.c
+++ b/net/smc/smc_wr.c
@@ -170,15 +170,11 @@  void smc_wr_tx_cq_handler(struct ib_cq *ib_cq, void *cq_context)
 
 static inline int smc_wr_tx_get_free_slot_index(struct smc_link *link, u32 *idx)
 {
-	*idx = link->wr_tx_cnt;
 	if (!smc_link_sendable(link))
 		return -ENOLINK;
-	for_each_clear_bit(*idx, link->wr_tx_mask, link->wr_tx_cnt) {
-		if (!test_and_set_bit(*idx, link->wr_tx_mask))
-			return 0;
-	}
-	*idx = link->wr_tx_cnt;
-	return -EBUSY;
+
+	*idx = find_and_set_bit(link->wr_tx_mask, link->wr_tx_cnt);
+	return *idx < link->wr_tx_cnt ? 0 : -EBUSY;
 }
 
 /**