From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tejun Heo Subject: Re: [PATCH] Re: [git patches] libata updates - (improve post-reset device ready test) regression Date: Sat, 10 May 2008 15:34:11 +0900 Message-ID: <482541E3.6020001@gmail.com> References: <20080506154847.GA15299@havoc.gtf.org> <20080507145616.GA2210@gentoox2.trippelsdorf.de> <48231DAF.7070900@garzik.org> <4823BD8E.40109@gmail.com> <4823D4EC.7060407@garzik.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from rv-out-0506.google.com ([209.85.198.230]:26065 "EHLO rv-out-0506.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752009AbYEJGeV (ORCPT ); Sat, 10 May 2008 02:34:21 -0400 Received: by rv-out-0506.google.com with SMTP id l9so1890548rvb.1 for ; Fri, 09 May 2008 23:34:20 -0700 (PDT) In-Reply-To: <4823D4EC.7060407@garzik.org> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Jeff Garzik Cc: Markus Trippelsdorf , linux-ide@vger.kernel.org, Andrew Morton , Linus Torvalds , LKML , Takashi Iwai , marc.c.dionne@gmail.com, dl9pf@gmx.de, bug-track@fisher-privat.net, sitsofe@yahoo.com Jeff Garzik wrote: > Tejun Heo wrote: >> This means that we need to make custom readiness tests for controllers >> using 0x77 or 0x7f. Eeeek... Both groups of controllers are behaving >> in incorrect way. Controllers shouldn't use 0x77 or 0x7f for either >> busy or ready states - it's invalid for both, yet, some use the 77/7f >> for busy while others use them for ready state. Great. :-( > > I think that's assuming too much? PATA and SATA are quite different > here... in PATA the status is mostly the value from the device directly > off the wires. in SATA, it may be from the device or from the > controller. And "smart" or firmware-based controllers may generate > their own status, too, apart from the device's status. > > So that results in varied status returns, and not all the time is a > definite "ready" or "not ready" obvious. I think it's pretty safe to say that these weird ready values are from TF emulation on controller side. The ready/not ready distinction is probably too simplistic but those values aren't supposed to appear during post-reset readiness test. Sorry about the big regression. Heh... It's amazing how all the controllers I tested didn't show the problem and I did test a good number of combinations. I still think it would be better to have a unified readiness test function. The problem is subtle (device misdetection on hotplug of certain drives) and went unnoticed quite some time for JMB ahcis && test coverage over those things can't be good. I'll try to think about something better. Thanks. -- tejun