[dpdk-dev] [PATCH] crypto/openssl: performance improvements

[De Lara Guarch, Pablo]

Hi,

> -----Original Message-----
> From: Akhil Goyal [mailto:akhil.goyal at nxp.com]
> Sent: Tuesday, August 15, 2017 7:45 AM
> To: De Lara Guarch, Pablo <pablo.de.lara.guarch at intel.com>;
> dev at dpdk.org; Doherty, Declan <declan.doherty at intel.com>
> Cc: hemant.agrawal at nxp.com
> Subject: Re: [PATCH] crypto/openssl: performance improvements
> 
> On 8/14/2017 7:47 PM, De Lara Guarch, Pablo wrote:
> > Hi Akhil,
> >
> >> -----Original Message-----
> >> From: Akhil Goyal [mailto:akhil.goyal at nxp.com]
> >> Sent: Friday, July 28, 2017 12:08 PM
> >> To: dev at dpdk.org; Doherty, Declan <declan.doherty at intel.com>
> >> Cc: De Lara Guarch, Pablo <pablo.de.lara.guarch at intel.com>;
> >> hemant.agrawal at nxp.com; Akhil Goyal <akhil.goyal at nxp.com>
> >> Subject: [PATCH] crypto/openssl: performance improvements
> >>
> >> key and algo are added in the openssl ctx during session
> >> initialization instead of adding it for each packet.
> >>
> >> Also in case of HMAC the openssl APIs HMAC_XXX give better
> >> performance for all HMAC cases.
> >>
> >> Signed-off-by: Akhil Goyal <akhil.goyal at nxp.com>
> >
> > Thanks for the patch, nice optimization!
> > Could you split this into two patches, as you are doing two different
> things here?
> > One for the first sentence and another one for the second sentence.
> > Also, as you do that, could you rename the title to be more explicit?
> > Like: crypto/openssl: initialize cipher key at session init
> >
> > Finally, I was looking at GCM, and I think it could benefit from this.
> > I will send a separate patch for it, unless you want to integrate it in this
> patchset yourself.
> >
> 
> Ok I would split the patches.
> For GCM I will try to incorporate in this patchset, if I get some performance
> improvement, or I would send a different patch later if some issue comes.

Thanks Ahkil. Since I am working on AES-CCM for this PMD, I have the change
already done. I have seen performance improvements, but it is not as straight forward
as the cipher algorithms, because GMAC is also affected, which is in a different code path,
but requires GCM to be set.

> 
> Thanks,
> Akhil
>