From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: [PATCH #upstream-fixes] sata_nv: disable ADMA mode by default Date: Thu, 10 Jan 2008 00:58:14 -0500 Message-ID: <4785B3F6.4000802@garzik.org> References: <4781F008.9070404@gmail.com> <4782422C.8020202@rtr.ca> <478246AE.3020500@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from srv5.dvmed.net ([207.36.208.214]:33500 "EHLO mail.dvmed.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750785AbYAJF6U (ORCPT ); Thu, 10 Jan 2008 00:58:20 -0500 In-Reply-To: <478246AE.3020500@gmail.com> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Tejun Heo Cc: Mark Lord , Robert Hancock , IDE/ATA development list , Allen Martin , Peer Chen , Andrew Morton , Linus Torvalds Tejun Heo wrote: > There still are remaining issues with ADMA support. Disable it by > default and warn when enabling. > > Signed-off-by: Tejun Heo > --- > Jeff, please hold off till Robert acks. Robert, what do you think? > > drivers/ata/sata_nv.c | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) Don't worry, I won't "pull the trigger" immediately and without a lot of discussion. (and please help keep nvidia cc'd on changes to sata_nv) Current sleepless (thus discountable? :)) thoughts: * definitely leaning towards adma=0 default. if distros are disabling it, and upstream is not, that's a big hint :) * By switching to the tried-and-true legacy-IDE-like interface, adma=0 seems to make a lot of problems go away. * It is so late in 2.6.24-rc, it seems unlikely that we have enough time for testing such a major, fundamental behavior change in sata_nv, this late in the game. If it weren't for the time factor, I would be in favor of applying the patch and getting test results. Overall, while I do have docs under NDA (the only one in the world outside of NV?), they are pretty bare bones. And the ADMA interface, while found on many thousands of NV chips, was only one rev -- CK804 -- and is no longer being used. NV uses AHCI now. I think ADMA is an experiment that failed, in both the software sense and the hardware sense. The effort Robert has put into the ADMA code, fixing many bugs (I think I have a fix from him still to be applied, during my absence) is frankly amazing given the limits, but IMO ADMA is just "not there." If docs were available and NV actively supported the ADMA mode, things would probably be different, but they aren't. Jeff