From mboxrd@z Thu Jan 1 00:00:00 1970 From: Harvey Harrison Subject: Re: get/put unaligned helpers Date: Fri, 13 Feb 2009 14:07:02 -0800 Message-ID: <1234562823.5521.21.camel@brick> References: <20090211143521.GA18866@lst.de> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: Received: from wf-out-1314.google.com ([209.85.200.175]:38459 "EHLO wf-out-1314.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753342AbZBMWHG (ORCPT ); Fri, 13 Feb 2009 17:07:06 -0500 In-Reply-To: <20090211143521.GA18866@lst.de> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Christoph Hellwig Cc: linux-kernel@vger.kernel.org, Boaz Harrosh , Matthew Wilcox , James Bottomley , linux-scsi On Wed, 2009-02-11 at 15:35 +0100, Christoph Hellwig wrote: > I just loked into using the new get/put unaligned helpers. I must > say that I'm really unhappy about the lack of typing there. We did > put all the sparse infrastructure in place to make sure we do have > strong typechecking for LE/Be types, but using these helpers defeats > that. > > Can you please make these properly typed? Christoph, here's a quick summary of comments on the unaligned access bits I've previously submitted, please let me know if you have any additional ones. Proposed API: u16 load_le16_noalign(const __le16 *p) u32 load_le32_noalign(const __le32 *p) u64 load_le64_noalign(const __le64 *p) u16 load_be16_noalign(const __be16 *p) u32 load_be32_noalign(const __be32 *p) u64 load_be64_noalign(const __be64 *p) void store_le16_noalign(__le16 *p, u16 val) void store_le32_noalign(__le32 *p, u32 val) void store_le64_noalign(__le64 *p, u64 val) void store_be16_noalign(__be16 *p, u16 val) void store_be32_noalign(__be32 *p, u32 val) void store_be64_noalign(__be64 *p, u64 val) I'd also like to offer aligned versions like load_le16() store_le16() to make it symmetric. Then for arches that have no alignment constraints the unaligned helpers would degrade to the aligned helpers. load_le16() already exists as le16_to_cpup, but store_le16 would be a new API. Now, from discussions with the SCSI-folks, they really aren't interested in a typesafe interface as most of the time they are reading/writing into a u8 buffer at some offset. It seems a lot of the other places currently in-kernel also do the same. So I'm tempted to keep offering a version wrapped around the typesafe versions that takes a u8 *. As they were not interested in doing the following: load_be16_noalign((__be16 *)&buf[offset]); Perhaps a wrapper could be provided: load_be16_offset(u8 *buf, int offset) Thoughts? Harvey