From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([140.186.70.92]:36271) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1RiUFg-00045g-6K for qemu-devel@nongnu.org; Wed, 04 Jan 2012 12:03:49 -0500 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1RiUFa-0004aE-MZ for qemu-devel@nongnu.org; Wed, 04 Jan 2012 12:03:48 -0500 Received: from plane.gmane.org ([80.91.229.3]:52273) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1RiUFa-0004a2-G9 for qemu-devel@nongnu.org; Wed, 04 Jan 2012 12:03:42 -0500 Received: from public by plane.gmane.org with local (Exim 4.69) (envelope-from ) id 1RiUFX-0001J4-MB for qemu-devel@nongnu.org; Wed, 04 Jan 2012 18:03:39 +0100 Sender: Paolo Bonzini Message-ID: <4F048662.1020702@redhat.com> Date: Wed, 04 Jan 2012 18:03:30 +0100 From: Paolo Bonzini MIME-Version: 1.0 References: <1313152395-25248-1-git-send-email-morita.kazutaka@lab.ntt.co.jp> <20111223133850.GA12770@lst.de> <20111229120626.GA32331@lst.de> <20111230103500.GA1740@stefanha-thinkpad.localdomain> <20120102153959.GA22823@lst.de> <20120103081655.GB28636@stefanha-thinkpad.localdomain> <20120104155825.GA11210@lst.de> In-Reply-To: <20120104155825.GA11210@lst.de> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] coroutine bug?, was Re: [PATCH] sheepdog: use coroutines List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Christoph Hellwig Cc: public-kwolf-H+wXaHxf7aLQT0dZR+AlfA@plane.gmane.org, public-sheepdog-CJ+3F33aHDCwRfvgX43A7Q@plane.gmane.org, Stefan Hajnoczi , public-qemu-devel-qX2TKyscuCcdnm+yROfE0A@plane.gmane.org On 01/04/2012 04:58 PM, Christoph Hellwig wrote: > Btw, what is the plan forward for the block I/O interface? Only > qcow2, sheepdog and nbd implement the coroutine interfaces Actually all formats implement a coroutine interface except raw, qed and vdi (IIRC). But the "minor" formats do not implement scatter/gather I/O, and use a bounce buffer instead. Among the protocols, only nbd implements coroutines, the others use AIO. > , and none of the > hardware emulations calls them directly. Yes, AIO is often simpler to use for hardware emulation (or anyway it has smaller advantages). But I do expect that SD could easily switch from synchronous to coroutines, for example. Floppy is a mess. IMO both coroutines and AIO have their uses. Converting qed to coroutines would make sense, but this is much less true for converting raw and even less for rbd or iscsi (the external libraries are callback-based). On the other hand, changing all formats to support scatter/gather I/O should not be hard, and would enable some nice simplifications in the code. I have enough other cleanups on my plate, though. :/ Paolo