[PATCH v30 03/20] iov_iter: skip copy if src == dst for direct data placement

[Jens Axboe]

On 7/15/25 11:08 AM, Chaitanya Kulkarni wrote:
> Jens,
> 
> On 7/15/25 08:06, Jens Axboe wrote:
>> On 7/15/25 7:27 AM, Aurelien Aptel wrote:
>>> From: Ben Ben-Ishay <benishay at nvidia.com>
>>>
>>> When using direct data placement (DDP) the NIC could write the payload
>>> directly into the destination buffer and constructs SKBs such that
>>> they point to this data. To skip copies when SKB data already resides
>>> in the destination buffer we check if (src == dst), and skip the copy
>>> when it's true.
>>>
>>> Signed-off-by: Ben Ben-Ishay <benishay at nvidia.com>
>>> Signed-off-by: Boris Pismenny <borisp at nvidia.com>
>>> Signed-off-by: Or Gerlitz <ogerlitz at nvidia.com>
>>> Signed-off-by: Yoray Zack <yorayz at nvidia.com>
>>> Signed-off-by: Shai Malin <smalin at nvidia.com>
>>> Signed-off-by: Aurelien Aptel <aaptel at nvidia.com>
>>> Reviewed-by: Chaitanya Kulkarni <kch at nvidia.com>
>>> Reviewed-by: Max Gurtovoy <mgurtovoy at nvidia.com>
>>> ---
>>>   lib/iov_iter.c | 9 ++++++++-
>>>   1 file changed, 8 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/lib/iov_iter.c b/lib/iov_iter.c
>>> index f9193f952f49..47fdb32653a2 100644
>>> --- a/lib/iov_iter.c
>>> +++ b/lib/iov_iter.c
>>> @@ -62,7 +62,14 @@ static __always_inline
>>>   size_t memcpy_to_iter(void *iter_to, size_t progress,
>>>   		      size_t len, void *from, void *priv2)
>>>   {
>>> -	memcpy(iter_to, from + progress, len);
>>> +	/*
>>> +	 * When using direct data placement (DDP) the hardware writes
>>> +	 * data directly to the destination buffer, and constructs
>>> +	 * IOVs such that they point to this data.
>>> +	 * Thus, when the src == dst we skip the memcpy.
>>> +	 */
>>> +	if (!(IS_ENABLED(CONFIG_ULP_DDP) && iter_to == from + progress))
>>> +		memcpy(iter_to, from + progress, len);
>>>   	return 0;
>>>   }
>> This seems like entirely the wrong place to apply this logic...
>>
> 
> do you have any specific preference where it needs to be moved ?
> or any other way you would prefer ?

In the caller? First of all, having any kind of odd kconfig check in the
iov iter code makes zero sense. Why would a copy helper need to care at
all about what kind of drivers are enabled in the kernel? It's a gross
hack and layering violation.

-- 
Jens Axboe