Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S966709AbcCPWw1 (ORCPT ); Wed, 16 Mar 2016 18:52:27 -0400 Received: from mga04.intel.com ([192.55.52.120]:36416 "EHLO mga04.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S965441AbcCPWwZ convert rfc822-to-8bit (ORCPT ); Wed, 16 Mar 2016 18:52:25 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.24,346,1455004800"; d="scan'208";a="670899208" From: "Dalessandro, Dennis" To: Doug Ledford CC: Al Viro , infinipath , "linux-rdma@vger.kernel.org" , "linux-fsdevel@vger.kernel.org" , "linux-kernel@vger.kernel.org" , Linus Torvalds , "Weiny, Ira" , "John, Jubin" Subject: RE: [WTF] utterly tasteless ABI in hfi1 (around ->write()/->write_iter()) Thread-Topic: [WTF] utterly tasteless ABI in hfi1 (around ->write()/->write_iter()) Thread-Index: AQHRfzrF50zhFlRNHUek7cW1H15VVp9cnCSAgABofID//6Xf4A== Date: Wed, 16 Mar 2016 22:52:21 +0000 Message-ID: References: <20160316041717.GJ17997@ZenIV.linux.org.uk> <56E97FD7.9020007@redhat.com> <20160316220029.GA5213@scvm10.sc.intel.com> In-Reply-To: <20160316220029.GA5213@scvm10.sc.intel.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [172.18.205.10] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT MIME-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1147 Lines: 24 Just wanted to re-send so it gets picked up by the lists. I was using a different server than usual and vger did not like it. Hopefully this attempt fares better. Sorry for the extra noise. -- On Wed, Mar 16, 2016 at 11:46:31AM -0400, Doug Ledford wrote: >Intel: For your own sake's you might want to consider doing something >simple such as using two files, one for regular commands and one for >SDMA commands, and modifying both the hfi1 and libpsm2 code bases. I >don't care, and I wouldn't force you to do it, but I've made my argument >to Al and he appears to be running it up the tree, so it might be >easiest to capitulate. This should be doable. We could have multiple files and have each implement only the appropriate handler, write() for control and writev() for data. Another option we are considering is combine the struct used for regular commands and the one for SDMA commands into a union in a structure that includes some magic value that tells us how to interpret the payload. This way whether it's write() or writev() won't matter we'll know how to handle it. We'll start with fixing hfi1 first and go from there. -Denny