From mboxrd@z Thu Jan 1 00:00:00 1970 From: Douglas Gilbert Subject: Re: [PATCH] normalize fixed and descriptor sense data Date: Sat, 28 Aug 2004 14:07:45 +1000 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <41300511.2060303@torque.net> References: <3DD91260-F839-11D8-A52F-00039398BB5E@ieee.org> Reply-To: dougg@torque.net Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from borg.st.net.au ([65.23.158.22]:49070 "EHLO borg.st.net.au") by vger.kernel.org with ESMTP id S268156AbUH1EI5 (ORCPT ); Sat, 28 Aug 2004 00:08:57 -0400 In-Reply-To: <3DD91260-F839-11D8-A52F-00039398BB5E@ieee.org> List-Id: linux-scsi@vger.kernel.org To: Pat LaVarre Cc: linux-scsi@vger.kernel.org Pat LaVarre wrote: > Doug G: > > > so deferred errors can > > be ignored in many contexts > > Any hope of Linux always reporting that deferred errors have occurred, > at least via dmesg if not in a critical dialog? I certainly hope so. One small change I made in my patch yesterday was to pass deferred errors back to the block SG_IO code. This could be a largish job. First put a more robust mechanism in place, then sweep the scsi mid level and perhaps upper level drivers and convert them to the new interface. The important thing to do is to think about the error checking code distributed around the place. It is reasonably easy to fix the wrong headed stuff but wider issues are often involved (e.g. should this be ignored and/or processed at a different level). > A deferred error, in the context of sense data ((bytes[0] & x7F) != x70) > is usually a purportedly rare and therefore interesting catastrophe of > some kind, yes? A failure to write seen while flushing cache. A > failure to eject after obediently faking "immediate"ly good status. Etc. > > > http://marc.theaimsgroup.com/?l=linux-scsi&m=109359716220586 Perhaps deferred errors should be picked up at the lowest level of the midlevel, logged and ignored at higher levels unless they are going back to a pass through interface. > Sorry I failed to click thru to the context for this thread. That is fine, the only other folks that I seemed to wake up were the style police :-) Doug Gilbert