Re: [dpdk-dev] [PATCH] ethdev: add new offload flag to keep CRC

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

On 6/20/2018 6:39 PM, Andrew Rybchenko wrote:
> On 06/20/2018 08:24 PM, Ferruh Yigit wrote:
>> On 6/20/2018 8:42 AM, Andrew Rybchenko wrote:
>>> On 06/19/2018 09:02 PM, Ferruh Yigit wrote:
>>>> DEV_RX_OFFLOAD_KEEP_CRC offload flag added. PMDs that supports keeping
>>>> CRC should advertise this offload capability.
>>>>
>>>> DEV_RX_OFFLOAD_CRC_STRIP flag will remain one more release
>>>> default behavior in PMDs are to keep the CRC until this flag removed
>>>>
>>>> Until DEV_RX_OFFLOAD_CRC_STRIP flag is removed:
>>>> - Setting both KEEP_CRC & CRC_STRIP is INVALID
>>>> - Setting only CRC_STRIP PMD should strip the CRC
>>>> - Setting only KEEP_CRC PMD should keep the CRC
>>>> - Not setting both PMD should keep the CRC
>>>>
>>>> A helper function rte_eth_dev_is_keep_crc() has been added to be able to
>>>> change the no flag behavior with minimal changes in PMDs.
>>>>
>>>> The PMDs that doesn't report the DEV_RX_OFFLOAD_KEEP_CRC offload can
>>>> remove rte_eth_dev_is_keep_crc() checks next release, related code
>>>> commented to help the maintenance task.
>>>>
>>>> And DEV_RX_OFFLOAD_CRC_STRIP has been added to virtual drivers since
>>>> they don't use CRC at all, when an application requires this offload
>>>> virtual PMDs should not return error.
>>>>
>>>> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
>>>> ---
>>> <...>
>>>
>>>> diff --git a/lib/librte_ethdev/rte_ethdev_driver.h b/lib/librte_ethdev/rte_ethdev_driver.h
>>>> index c9c825e3f..09a42f8c2 100644
>>>> --- a/lib/librte_ethdev/rte_ethdev_driver.h
>>>> +++ b/lib/librte_ethdev/rte_ethdev_driver.h
>>>> @@ -325,6 +325,26 @@ typedef int (*ethdev_uninit_t)(struct rte_eth_dev *ethdev);
>>>>  int __rte_experimental
>>>>  rte_eth_dev_destroy(struct rte_eth_dev *ethdev, ethdev_uninit_t ethdev_uninit);
>>>>  
>>>> +/**
>>>> + * PMD helper function to check if keeping CRC is requested
>>>> + *
>>>> + * @param rx_offloads
>>>> + *   offloads variable
>>>> + *
>>>> + * @return
>>>> + *   Return positive if keeping CRC is requested,
>>>> + *   zero if stripping CRC is requested
>>>> + */
>>>> +static inline int
>>>> +rte_eth_dev_is_keep_crc(uint64_t rx_offloads)
>>>> +{
>>>> +	if (rx_offloads & DEV_RX_OFFLOAD_CRC_STRIP)
>>>> +		return 0;
>>>> +
>>>> +	/* no KEEP_CRC or CRC_STRIP offload flags means keep CRC */
>>>> +	return 1;
>>>> +}
>>>> +
>>>>  #ifdef __cplusplus
>>>>  }
>>>>  #endif
>>> A couple of control questions about the function:
>>>  - shouldn't __rte_experimental be used?
>> This is an internal function, not API, so I think doesn't require to be
>> experimental.
> 
> Just to make my thoughts clear: description does not say that it is an internal.
> So, nothing prevents external entities to use it. Changes will be API breakage.

rte_ethdev_driver.h is not public header, it is just for PMDs.

> 
>>>  - if the function remains in the future, it will be a bit asymmetric vs other
>>>    offload flags. Right now it is clear why the function is introduced, but
>>>    it is the question if the function should remain or go away in the future
>>>    (as far as I know no other offload flag has similar function to check).
>> No other offloads don't have similar functions, this is kind special.
>>
>> There will be more changes related CRC next release, CRC_STRIP will be removed
>> and no flag will mean strip CRC. So the conditions to is_keep_crc will be changed.
>> This function is to manage this change easier, localize the information in to
>> single function to make it easy to update later.
> 
> It is perfectly clear why it is required right now and introduced (as I said
> from the very beginning).
> Yes, it is will be the history which explains why it is so, but if we make
> a step forward and discard the history it will look asymmetric -
> it will be a function which checks single bit. It is really minor and
> 100% up to you.

I see, right it will be just a wrapper to bit check. In this patch it helps to
revert to logic, from strip_crc to keep_crc. In next release I am OK to remove
function and return back to bit check in PMDs, will this be more reasonable?