From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261360AbTI0SL2 (ORCPT ); Sat, 27 Sep 2003 14:11:28 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262133AbTI0SL2 (ORCPT ); Sat, 27 Sep 2003 14:11:28 -0400 Received: from [24.76.142.122] ([24.76.142.122]:20230 "HELO signalmarketing.com") by vger.kernel.org with SMTP id S261360AbTI0SL1 (ORCPT ); Sat, 27 Sep 2003 14:11:27 -0400 Date: Sat, 27 Sep 2003 13:11:25 -0500 (CDT) From: Derek Foreman To: Jens Axboe cc: linux-kernel@vger.kernel.org Subject: Re: CDROM_SEND_PACKET oddity In-Reply-To: <20030927175445.GI15415@suse.de> Message-ID: References: <20030927114712.GJ3416@suse.de> <20030927122703.GK3416@suse.de> <20030927175445.GI15415@suse.de> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Sat, 27 Sep 2003, Jens Axboe wrote: > On Sat, Sep 27 2003, Jens Axboe wrote: > > On Sat, Sep 27 2003, Jens Axboe wrote: > > > On Fri, Sep 26 2003, Derek Foreman wrote: > > > > The example code from > > > > http://www.ussg.iu.edu/hypermail/linux/kernel/0202.0/att-0603/01-cd_poll.c > > > > > > > > Does not behave as expected on my 2.6.0-test5 system. While the command > > > > seems to be successfully sent - 2 of my drives report it as an invalid > > > > opcode - for the other 2 drives, the buffer comes back all zeros. > > > > (actually, the buffer's contents will remain in whatever state they're in > > > > before the ioctl is called) > > > > > > > > Sending the same command to those 2 drives with SG_IO results in the > > > > expected behaviour. > > > > > > Can you try current -bk? It has some fixes for CDROM_SEND_PACKET. > > > > > > However, cd_poll should be rewritten to use SG_IO. Pretty trivial > > > exercise. > > > > Actually, try this patch against current bk, it kills the > > CDROM_SEND_PACKET setup and use SG_IO internally instead. Should be much > > much better than what we have now. It's not tested here at all though, > > I'd appreciate it if you could give it a go. > > This has a better chance of working. Changes: > > - Don't export sg_io() anyways (leftover) > - Actually set ->cmdp and ->cmd_len > > still untested. The old one worked after I fixed it up to set cmdp and the sense buffer. I'll test this one out a little later today.