Re: [PATCH RFC 4/4] crypto: qat - Add new algif interface for userspace

[Tadeusz Struk <tadeusz.struk@intel.com>]

Hi,
On 11/06/2014 09:31 PM, Herbert Xu wrote:
> On Thu, Nov 06, 2014 at 09:24:22PM -0800, Tadeusz Struk wrote:
>>
>> What about aead? Using my algif_qat I can build a single request to HW
>> that preforms both encryption and authentication in one go, ever for
>> things like aes-cbc-hmac-sha1. This allows me to get great performance
>> of authenticated encryption from user space.
> 
> aead is simpler as it's already part of the API so we just need to
> add algif_aead.
> 
> Cheers,
> 

Another thing is that the algif_skcipher works in kind of synchronous way:

-------             ----------
| app |--write()--> | socket | --> build_sgl()
-------             ----------        |
   <----------------------------------/
-------             ----------                   ------
| app |--read()---> | socket | --send_to_hw()--> | HW |
-------             ----------        |          ------
                                      |            |
                                      |            |
   <-------------------------------wait() <--------/



What I do in the algif_qat is this:

-------             ----------                   ------
| app |--write()--> | socket | --send_to_hw()--> | HW |
-------             ----------        |          ------
   <----------------------------------/            |
                         <-------------------------/
-------             ----------
| app |--read()---> | socket |
-------             ----------
                        |
   <--------------------/

This way I can get much higher throughput than with algif_skcipher.
Regards,
Tadeusz