[dpdk-dev] [RFC v2, 1/2] cryptodev: add support to set session private data

Akhil Goyal akhil.goyal at nxp.com
Thu Jan 25 07:42:39 CET 2018


On 1/25/2018 1:16 AM, De Lara Guarch, Pablo wrote:
> 
> 
>> -----Original Message-----
>> From: Gujjar, Abhinandan S
>> Sent: Tuesday, January 23, 2018 8:54 AM
>> To: Doherty, Declan <declan.doherty at intel.com>; akhil.goyal at nxp.com; De
>> Lara Guarch, Pablo <pablo.de.lara.guarch at intel.com>;
>> Jerin.JacobKollanukkaran at cavium.com
>> Cc: dev at dpdk.org; Vangati, Narender <narender.vangati at intel.com>;
>> Gujjar, Abhinandan S <abhinandan.gujjar at intel.com>; Rao, Nikhil
>> <nikhil.rao at intel.com>
>> Subject: [RFC v2, 1/2] cryptodev: add support to set session private data
>>
>> Update rte_crypto_op to indicate private data offset.
>>
>> The application may want to store private data along with the
>> rte_cryptodev that is transparent to the rte_cryptodev layer.
>> For e.g., If an eventdev based application is submitting a
>> rte_cryptodev_sym_session operation and wants to indicate event
>> information required to construct a new event that will be enqueued to
>> eventdev after completion of the rte_cryptodev_sym_session operation.
>> This patch provides a mechanism for the application to associate this
>> information with the rte_cryptodev_sym_session session.
>> The application can set the private data using
>> rte_cryptodev_sym_session_set_private_data() and retrieve it using
>> rte_cryptodev_sym_session_get_private_data().
> 
> Hi Abhinandan,
> 
>>
>> Signed-off-by: Abhinandan Gujjar <abhinandan.gujjar at intel.com>
>> Signed-off-by: Nikhil Rao <nikhil.rao at intel.com>
>> ---
>> Notes:
>>          V2:
>> 	1. Removed enum rte_crypto_op_private_data_type
>> 	2. Corrected formatting
>>
>>   lib/librte_cryptodev/rte_crypto.h    |  8 ++++++--
>>   lib/librte_cryptodev/rte_cryptodev.h | 32
>> ++++++++++++++++++++++++++++++++
>>   2 files changed, 38 insertions(+), 2 deletions(-)
>>
>> diff --git a/lib/librte_cryptodev/rte_crypto.h
>> b/lib/librte_cryptodev/rte_crypto.h
>> index 95cf861..14c87c8 100644
>> --- a/lib/librte_cryptodev/rte_crypto.h
>> +++ b/lib/librte_cryptodev/rte_crypto.h
>> @@ -84,8 +84,12 @@ struct rte_crypto_op {
>>   	 */
>>   	uint8_t sess_type;
>>   	/**< operation session type */
>> -
>> -	uint8_t reserved[5];
>> +	uint16_t private_data_offset;
>> +	/**< Offset to indicate start of private data (if any). The private
>> +	 * data may be used by the application to store information which
>> +	 * should remain untouched in the library/driver
> 
> Is this the offset for the private data after the crypto operation?
>  From your title, it looks like it is for the session private data, but then, this shouldn't be here.
> If it is for the crypto operation, I suggest you to separate it in another patch.
> Also, you should indicate where the offset starts from. For the IV, the offset is counted
> from the start of the rte_crypto_op, so I think it should be the same, to keep consistency.
> 
> For the session private data, we see two options:
> 
> 1 - Add a  "valid" private data field in the rte_cryptodev_sym_session structure,
> so when it is set, it indicates that the session contains private data
> (a single bit would be enough, 1 to indicate there is, and 0 to indicate there is not).
> This would go into the beginning of the structure, so this would require an ABI deprecation notice.
> This also assumes that the private data starts just after the session header
> 
> 2 -  Do not add an extra "valid" private data field in rte_cryptodev_sym_session structure,
> and add a small header in the private data, which contains the "valid" bit.
> Then, when calling sym_session_get_private_data, this bit should be checked.
> Note that the object that holds the session structure needs to be big enough to hold this value.
> If the object has only space for the sess_private_data array, then the session has no private data.
> Therefore, this approach might be less performant, but with no ABI deprecation required.
> 
> I would recommend you to send a deprecation notice for option 1, then check the performance of both option,
> and if needed, make the change in the structure, in 18.05.
> 
> Regards,
> Pablo
> 

My thoughts are also inline with Pablo.

-Akhil


More information about the dev mailing list