From: Tobias Rauter Subject: [RFC] Freescale DCP crypto acceleration Date: Wed, 8 May 2013 20:41:12 +0200 Message-ID: <1368038473-6979-1-git-send-email-tobias.rauter@gmail.com> Cc: "Christoph G. Baumann" To: Tobias Rauter , linux-crypto@vger.kernel.org, Herbert Xu Return-path: Received: from mail-ee0-f53.google.com ([74.125.83.53]:47359 "EHLO mail-ee0-f53.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751031Ab3EHSlp (ORCPT ); Wed, 8 May 2013 14:41:45 -0400 Received: by mail-ee0-f53.google.com with SMTP id d49so1129614eek.26 for ; Wed, 08 May 2013 11:41:43 -0700 (PDT) Sender: linux-crypto-owner@vger.kernel.org List-ID: Hi! I am working on a driver for the data co processor of freescale's imx28 (should also work on mx23 and mx5). I noticed that Christoph sent a patch for this device some days ago to the crypto mailing list. As far as I see this is basically the Freescale-SDK driver with some modifications for mainline. This was also my first approach. Since this driver has some problems (duplicated code, mutex in atomic context) I decided to get the reference manual and write the driver from scratch. To prevent double work (in case Christoph or someone else does the same thing), I want to share my work here now. This is my first real kernel driver so it would be nice if someone could take some time to find the noob-mistakes and send me feedback. Things I'm working on: - get rid of the global device - at the moment the number of entries of the in/out scatterlist is limited to 20. I am working on a ringbuffer where the CPU fills the co-processors buffer while it iterates through the pages. - overall performance - ATM each scatterlist entry of the request has to be aligned to AES_BLOCK_SIZE. Since I was not able to find a documentation, I want to ask if my idea is the right approach to solve this problem: use a ablkcipher walk ,iterate through this walk and do dma_map_single for each buffer instead of the whole scatterlist. regards, Tobias