If xas_find_marked() failed, there is no need to call xas_store() and
xas_clear_mark(). Just call xas_store() and xas_clear_mark() if
xas_find_marked() succeed.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
---
 lib/xarray.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/lib/xarray.c b/lib/xarray.c
index 3fac3f2cea9d..012d0ef5e847 100644
--- a/lib/xarray.c
+++ b/lib/xarray.c
@@ -1860,10 +1860,11 @@ int __xa_alloc(struct xarray *xa, u32 *id, void *entry,
 		xas_find_marked(&xas, limit.max, XA_FREE_MARK);
 		if (xas.xa_node == XAS_RESTART)
 			xas_set_err(&xas, -EBUSY);
-		else
+		else {
 			*id = xas.xa_index;
-		xas_store(&xas, entry);
-		xas_clear_mark(&xas, XA_FREE_MARK);
+			xas_store(&xas, entry);
+			xas_clear_mark(&xas, XA_FREE_MARK);
+		}
 	} while (__xas_nomem(&xas, gfp));
 
 	return xas_error(&xas);
-- 
2.30.0

On Fri, Nov 01, 2024 at 11:50:26PM +0800, Kemeng Shi wrote:
> If xas_find_marked() failed, there is no need to call xas_store() and
> xas_clear_mark(). Just call xas_store() and xas_clear_mark() if
> xas_find_marked() succeed.

No.  The point of the xas interfaces is that they turn into no-ops once
an error has occurred.

> -		else
> +		else {
>  			*id = xas.xa_index;
> -		xas_store(&xas, entry);
> -		xas_clear_mark(&xas, XA_FREE_MARK);
> +			xas_store(&xas, entry);
> +			xas_clear_mark(&xas, XA_FREE_MARK);
> +		}
>  	} while (__xas_nomem(&xas, gfp));
>  
>  	return xas_error(&xas);
> -- 
> 2.30.0
>

on 11/2/2024 12:02 AM, Matthew Wilcox wrote:
> On Fri, Nov 01, 2024 at 11:50:26PM +0800, Kemeng Shi wrote:
>> If xas_find_marked() failed, there is no need to call xas_store() and
>> xas_clear_mark(). Just call xas_store() and xas_clear_mark() if
>> xas_find_marked() succeed.
> 
> No.  The point of the xas interfaces is that they turn into no-ops once
> an error has occurred.
Yes, xas interfaces can tolerate error. The question is do we really need to
call xas_store(...) here if we already know there is no room to store new entry.
But no insistant on this as it's not a big deal anyway.

Will drop this on next version if you still disklike this.
Thanks.

> 
>> -		else
>> +		else {
>>  			*id = xas.xa_index;
>> -		xas_store(&xas, entry);
>> -		xas_clear_mark(&xas, XA_FREE_MARK);
>> +			xas_store(&xas, entry);
>> +			xas_clear_mark(&xas, XA_FREE_MARK);
>> +		}
>>  	} while (__xas_nomem(&xas, gfp));
>>  
>>  	return xas_error(&xas);
>> -- 
>> 2.30.0
>>
>

On Mon, Nov 04, 2024 at 09:55:42AM +0800, Kemeng Shi wrote:
> > No.  The point of the xas interfaces is that they turn into no-ops once
> > an error has occurred.
> Yes, xas interfaces can tolerate error. The question is do we really need to
> call xas_store(...) here if we already know there is no room to store new entry.
> But no insistant on this as it's not a big deal anyway.
> 
> Will drop this on next version if you still disklike this.

I very much dislike this.  The point is to make the callers simpler.