Re: [PATCH v2 1/2] net/nfp: standardize the use of RSS-related codes

[Ferruh Yigit <ferruh.yigit@amd.com>]

On 2/28/2023 8:46 AM, Chaoyong He wrote:
>> On 2/21/2023 3:55 AM, Chaoyong He wrote:
>>> From: Long Wu <long.wu@corigine.com>
>>>
>>> The usage of RTE_ETH_MQ_RX_RSS and RTE_ETH_MQ_RX_RSS_FLAG are
>> mixed in
>>> nfp_net_configure(), use RTE_ETH_MQ_RX_RSS_FLAG uniformly.
>>>
>>> Signed-off-by: Long Wu <long.wu@corigine.com>
>>> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
>>> Reviewed-by: Niklas Söderlund <niklas.soderlund@corigine.com>
>>> ---
>>>  drivers/net/nfp/nfp_common.c | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/net/nfp/nfp_common.c
>>> b/drivers/net/nfp/nfp_common.c index 907777a9e4..a545a10013 100644
>>> --- a/drivers/net/nfp/nfp_common.c
>>> +++ b/drivers/net/nfp/nfp_common.c
>>> @@ -171,7 +171,7 @@ nfp_net_configure(struct rte_eth_dev *dev)
>>>  	}
>>>
>>>  	/* Checking RX mode */
>>> -	if (rxmode->mq_mode & RTE_ETH_MQ_RX_RSS &&
>>> +	if (rxmode->mq_mode & RTE_ETH_MQ_RX_RSS_FLAG &&
>>>  	    !(hw->cap & NFP_NET_CFG_CTRL_RSS_ANY)) {
>>>  		PMD_INIT_LOG(INFO, "RSS not supported");
>>>  		return -EINVAL;
>>
>> They are same values, but when used as a mask, flag one can be more
>> proper.
>>
>> BUT,
>> Not sure how correct to '&' an enum element, enums supposed to be
>> abstraction on underneath values, right?
>> For this case user should know/care enum values which is wrong, perhaps to
>> OR flags to make enum elements was a mistake on our end.
>>
>> Anyway, with above usage all enum elements that has RSS_FLAG is taken
>> into account, like 'RTE_ETH_MQ_RX_VMDQ_RSS' (RSS mode with VMDq),
>> are you sure that is the intention, is there a chance you meant:
>>
>>  if (rxmode->mq_mode == RTE_ETH_MQ_RX_RSS && ...)
>>
>>
>> btw, what if mq_mode is 'RTE_ETH_MQ_RX_VMDQ_DCB', is it supported?
>>
> For now, we don't support 'RTE_ETH_MQ_RX_VMDQ_DCB'.
> 

My point was it is not checked in the driver, and that is the case for
most of the drivers, which means if user requests it, it will be
silently ignored, and if application depends on this feature this will
break the app.

So 'mq_mode' config option is not honored properly, for majority of the
cases all user/driver interested in is RSS, only a few drivers that
support other multi queue modes handles other cases.

We may improve this bit better in the ethdev layer, like providing a
mq_mode capability, etc...

>>
>> I just recognized there are various similar usage in drivers, and most of them
>> comes from same commit [1], we can discuss and fix that separately.
>>
> I saw you plan drop the RFC commit [1], so what we should do about this patch?
> 1. Keep our modify.
> 2. Follow your modify.
> 3. We just drop this commit. (your patch series will correct it anyway).
> Please give me an advice to guide my next move, thanks.
> 

Keep your modify,
when we continue to '&' 'rxmode->mq_mode' value, should use _FLAG as you
are updating in your patch.

>> [1]
>> 73fb89dd6a00 ("drivers/net: fix RSS hash offload flag if no RSS")