From mboxrd@z Thu Jan 1 00:00:00 1970 From: Robert Hancock Subject: Re: [RFC] add DMA setup FIS auto-activate feature Date: Thu, 23 Jul 2009 00:18:33 -0600 Message-ID: <4A6800B9.10706@gmail.com> References: <1248320734.9035.22.camel@sli10-desk.sh.intel.com> <4A67E00F.40808@pobox.com> <20090723054007.GA7473@sli10-desk.sh.intel.com> <4A67F8ED.2070204@kernel.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from an-out-0708.google.com ([209.85.132.248]:59913 "EHLO an-out-0708.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750990AbZGWGQW (ORCPT ); Thu, 23 Jul 2009 02:16:22 -0400 Received: by an-out-0708.google.com with SMTP id d40so938670and.1 for ; Wed, 22 Jul 2009 23:16:22 -0700 (PDT) In-Reply-To: <4A67F8ED.2070204@kernel.org> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Tejun Heo Cc: Shaohua Li , Jeff Garzik , linux-ide On 07/22/2009 11:45 PM, Tejun Heo wrote: > Shaohua Li wrote: >>> IIRC, not all NCQ-capable controllers support this feature... SATA 1.0 >>> hardware may not behave properly in response, no? >> It appears the AHCI spec doesn't define a HBA capability about this >> feature, but I'm likely wrong as I'm not quite familar with SATA. > > IIRC, AA is in the ahci spec from rev 1.0, so if controllers implement > the spec correctly, it should work. But even for ahcis, if we enable > it by default, I'm fairly sure we'll be met by a number of unpleasant > surprises. Not sure whether doing that would be worth the trouble or > not. There's definitely a controller dependency here, as SATA 1.0 hardware won't respond properly if this feature is turned on. Likely we need a host flag to indicate whether the controller supports auto-activate. It seems like AHCI shouldn't be a problem, but it's not clear if any other controller types would support it. As far as busted hardware not handling it properly.. well if it just ignores the enabling then that's not a problem as things will just work as before. There could be devices that claim support, don't ignore it but don't handle it properly.. but realistically the only way we can find out if that's the case is turning it on and see what breaks.