**** BEGIN LOGGING AT Mon Jul 11 02:59:56 2011
Jul 11 06:13:25 <elFidel> ahoi
Jul 11 16:25:18 <ahigerd> REAL success this time!
Jul 11 16:25:36 <Tootoot222> =D
Jul 11 16:25:39 <Tootoot222> what'd you have to do?
Jul 11 16:26:06 <ahigerd> Seems that the problem is in the SHA-1 offloading. With the af_alg OpenSSL engine using software SHA-1 instead of trying to use the kernel's version, the handshaking succeeded.
Jul 11 16:26:25 <Tootoot222> ah, cool
Jul 11 16:26:31 <ahigerd> On a suggestion from the engine's developer I took out the line of code that enabled it, and it started working.
Jul 11 16:26:58 <kerio> so you only use the hw crypto for the algorithms that work?
Jul 11 16:27:06 <ahigerd> I'll see about contacting Marvell with a bug report on their SHA-1 stuff, but practically speaking this is faster anyway because hardware-offload of hash functions tends to suck for performance.
Jul 11 16:27:09 <kerio> cool story bro
Jul 11 16:27:22 <ahigerd> kerio: The weird thing is that the algorithm DOES work when tested in isolation.
Jul 11 16:27:41 <Tootoot222> ahigerd: so you don't really know what causes the problem, only that removing the sha-1 solves it?
Jul 11 16:27:45 <elFidel> interesting
Jul 11 16:27:55 <ahigerd> More or less, yeah, because calculating SHA-1 hashes through the chip produces the right results.
Jul 11 16:38:26 * ahigerd headscratches
Jul 11 16:38:39 <ahigerd> Curious. CPU time says I'm definitely seeing hardware offloading.
Jul 11 16:39:01 <ahigerd> But I'm actually seeing less throughput all around, contrary to other people's results.
Jul 11 16:39:27 <Tootoot222> less throughput compared to software algos?
Jul 11 16:39:34 <ahigerd> Yeah
Jul 11 16:39:50 <ahigerd> Hardware: Doing aes-128-cbc for 3s on 1024 size blocks: 33947 aes-128-cbc's in 0.08s
Jul 11 16:39:56 <ahigerd> Software: Doing aes-128-cbc for 3s on 1024 size blocks: 51861 aes-128-cbc's in 2.99s
Jul 11 16:40:17 <Tootoot222> uh
Jul 11 16:40:23 <Tootoot222> the first test went a lot less time?
Jul 11 16:40:29 <Tootoot222> 0.008
Jul 11 16:40:31 <ahigerd> That's CPU time, not real time
Jul 11 16:40:33 <Tootoot222> oh
Jul 11 16:40:47 <ahigerd> Real time is 3.00s in both cases -- it's seeing how many encrypts it can do in three seconds.
Jul 11 17:20:42 <ahigerd> Oh bloody heck
Jul 11 17:20:45 <ahigerd> I accidentally neutered the whole thing
Jul 11 17:20:51 <ahigerd> It's doing the work in userspace
Jul 11 17:21:00 <ahigerd> *rummage*
Jul 11 17:34:58 <kerio> u mad
**** ENDING LOGGING AT Mon Jul 11 20:00:49 2011

**** BEGIN LOGGING AT Mon Jul 11 20:02:14 2011
**** ENDING LOGGING AT Tue Jul 12 02:59:57 2011