> >> Hello, > >> The Linux-kernel maintainer rejected the /dev/crypto solution and has > >> added similar - but incompatible - functionality in the mainline kernel. > >> Check AF_ALG. There is comparison with it in cryptodev-linux pages. > > Thanks for your reply and great work on the benchmark comparisons. > > Could you explain why the performance is poor with AF_ALG for the NULL > > cipher case? > > > It has additional overhead imposed by the choice of its API. As far as I > remember it goes through the networking API for no particular reason.
Yes I will run some tests on real hardware accelerators here. I think you didn't get a chance to do the same which is atleast one of the reasons why the maintainer was not convinced. We can show some results and make a better proposition. Either case, I feel cryptodev-linux shouldn't be left out as it can be made an option for one looking for a different kernel-userspace interface for crypto. Having more options is not a bad thing IMO. As long as there is no conflict, should be merged in. A counter argument can be, af_alg is not yet merged into openssl as well. > > >> If you check openssl several parts of this patch have been incorporated. > >> The openssl team is pretty slow in including patches though and I plan > >> to give up maintaining an updated version for them. > > I guess from (2) you meant that the patch(es) have already been > > expected. So now do you mean some patches are still pending? > > > Yes. OK. For now, for any tests I'm using OpenSSL 1.0.1e on Cortex-A8 based beaglebone having the omap-* hardware accelerators on am33xx (which atleast seen to work with ocf-linux) Regards, Joel _______________________________________________ Cryptodev-linux-devel mailing list Cryptodev-linux-devel@gna.org https://mail.gna.org/listinfo/cryptodev-linux-devel
