From: Herbert Xu Subject: Re: [PATCH 0/2] crypto: add new driver for Marvell CESA Date: Wed, 29 Apr 2015 17:49:02 +0800 Message-ID: <20150429094902.GC9874@gondor.apana.org.au> References: <1428591523-1780-1-git-send-email-boris.brezillon@free-electrons.com> <20150428215232.6834906c@bbrezillon> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "David S. Miller" , linux-crypto@vger.kernel.org, Rob Herring , Pawel Moll , Mark Rutland , Ian Campbell , Kumar Gala , devicetree@vger.kernel.org, Tawfik Bayouk , Lior Amsalem , Nadav Haklai , Eran Ben-Avi , Thomas Petazzoni , Gregory CLEMENT , Jason Cooper , Sebastian Hesselbarth , Andrew Lunn , linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, Arnaud Ebalard To: Boris Brezillon Return-path: Content-Disposition: inline In-Reply-To: <20150428215232.6834906c@bbrezillon> Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-crypto.vger.kernel.org On Tue, Apr 28, 2015 at 09:52:32PM +0200, Boris Brezillon wrote: > > In particular, I'd like to discuss the threaded-irq approach taken in > this driver (other drivers are using tasklets). > The main reason behind this choice is the fact that crypto engines > are quite fast, and I'm worried about the CPU contention that might > happen in case of fully loaded crypto engines (the CPU might spend most > of its time in interrupt context until the crypto queue is emptied). > Using threaded-irq allows preemption of the crypto completion > operation, thus authorizing another thread (with higher priority) to be > scheduled. ITOH, the tasklet approach provide slightly performances (I > don't recall the exact numbers, but Arnaud did some tests). Either approach is fine with me. Cheers, -- Email: Herbert Xu Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt