[dpdk-dev] net/failsafe: add default Tx mbuf fast free capability

[Ferruh Yigit]

On 12/21/2018 12:52 PM, Andrew Rybchenko wrote:
> Hi Ferruh,
> 
> On 12/21/18 3:43 PM, Ferruh Yigit wrote:
>> On 12/21/2018 12:28 PM, Andrew Rybchenko wrote:
>>> On 12/21/18 3:12 PM, Ferruh Yigit wrote:
>>>> On 10/12/2018 12:36 PM, Andrew Rybchenko wrote:
>>>>> From: Ivan Malov <ivan.malov at oktetlabs.ru>
>>>>>
>>>>> This capability is reported when supported by the current emitting
>>>>> sub-device. Failsafe PMD itself does not excercise fast free logic.
>>>> I think overlay device capability reporting already discussed a few times, the
>>>> question is if an overlay devices should claim a feature when it depends on
>>>> underlay devices?
>>> The capability may be reported by the failsafe since it is transparent from
>>> fast free logic point of view.
>> Why it is transparent? If one of the underlying device doesn't support/claim
>> this feature, application can't use this feature with failsafe, isn't it?
> 
> tx_offload_capa in failsafe is a mask to apply on sub-device capabilities.

I missed this one, I see why it is transparent.

Why failsafe doesn't set a full tx_offload_capa MASK but maintain a list?

> So, if the capability is not supported by any sub-device it will not be 
> reported.
> As well if there is the capability bit in the mask, it will not be 
> reported regardless
> sub-devices capabilities. The description for the patch above tries to 
> explain it -
> it looks like not that successful.
> 
>>>> Given that no ack/review given to the patch, I am updating it as rejected.
>>> Is it a new policy? I thought that it was vice versa before.
>> Hi Andrew,
>>
>> Yes policy is other-way around indeed, when there is no comment at all default
>> behavior is accept, but please take above paragraph as my comment to the patch.
> 
> Got it.
> 
>> And I was thinking it is a little controversial and there is no support to have
>> it, so lets don't get it. What do you think?
> 
> I see you motivation.
> 
>>>>> Signed-off-by: Ivan Malov <ivan.malov at oktetlabs.ru>
>>>>> Signed-off-by: Andrew Rybchenko <arybchenko at solarflare.com>
>>>>> ---
>>>>>    doc/guides/nics/features/failsafe.ini | 1 +
>>>>>    drivers/net/failsafe/failsafe_ops.c   | 1 +
>>>>>    2 files changed, 2 insertions(+)
>>>>>
>>>>> diff --git a/doc/guides/nics/features/failsafe.ini b/doc/guides/nics/features/failsafe.ini
>>>>> index e3c4c08f2..b6f3dcee6 100644
>>>>> --- a/doc/guides/nics/features/failsafe.ini
>>>>> +++ b/doc/guides/nics/features/failsafe.ini
>>>>> @@ -7,6 +7,7 @@
>>>>>    Link status          = Y
>>>>>    Link status event    = Y
>>>>>    Rx interrupt         = Y
>>>>> +Fast mbuf free       = Y
>>>>>    Queue start/stop     = Y
>>>>>    Runtime Rx queue setup = Y
>>>>>    Runtime Tx queue setup = Y
>>>>> diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
>>>>> index 7f8bcd4c6..e3add404b 100644
>>>>> --- a/drivers/net/failsafe/failsafe_ops.c
>>>>> +++ b/drivers/net/failsafe/failsafe_ops.c
>>>>> @@ -78,6 +78,7 @@ static struct rte_eth_dev_info default_infos = {
>>>>>    		DEV_RX_OFFLOAD_SECURITY,
>>>>>    	.tx_offload_capa =
>>>>>    		DEV_TX_OFFLOAD_MULTI_SEGS |
>>>>> +		DEV_TX_OFFLOAD_MBUF_FAST_FREE |
>>>>>    		DEV_TX_OFFLOAD_IPV4_CKSUM |
>>>>>    		DEV_TX_OFFLOAD_UDP_CKSUM |
>>>>>    		DEV_TX_OFFLOAD_TCP_CKSUM |
>>>>>
> 
>