From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: [PATCHSET 4/4] implement PMP support, take 4 Date: Thu, 12 Jul 2007 16:27:08 -0400 Message-ID: <46968E9C.3060607@garzik.org> References: <1183287253677-git-send-email-htejun@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]:55758 "EHLO mail.dvmed.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758001AbXGLU1L (ORCPT ); Thu, 12 Jul 2007 16:27:11 -0400 In-Reply-To: <1183287253677-git-send-email-htejun@gmail.com> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Tejun Heo Cc: Alan Cox , linux-ide@vger.kernel.org, Forrest Zhao Tejun Heo wrote: > Hello, all. > > his is the fourth take of libata-pmp patchset. This patchset contains > 15 patches and implements PMP support. > > #01-04: implement libata PMP support > #05-09: implement sata_sil24 PMP support > #10-15: implement ahci PMP support > > Changes from the last take[L] are. > > * updated to fit new #upstream > * ACPI support for PMP fan-out ports added > * improved general PMP EH behavior > * various PMP chip specific quirks for smooth handling of first gen > products which at times are pretty quirky > * AHCI PMP support added > > Tested extensively with sil3214/32, ICH9R, JMB360/3 combined with > sil3726, 4726 and 5744 port multipliers. Disks from all major vendors > and an SATAPI device is used as downstream devices. sil5744 and 3726 > behave very nicely on all controllers. Hotplugging and suspend/resume > work well too. There are some remaining issues. > > * When attached to AHCI controller (ICH9R, JMB360/3), NCQ doesn't > work. Things look okay as long as only one command is in flight but > as soon as the second command is issued (to the same device), hell > breaks loose and all sorts of errors occur including device abort, > bad PMP number and interface fatal error. This behavior is common > across all PMP chips. I've disabled NCQ if PMP is attached to an > ahci controller for now. > > * Sil4726 is a bit too quirky. Without the first fan-out port > occupied, the thing acts really weirdly. As long as the first port > is occupied, it works okay. This is reportedly fixed by new > firmware. I tried to update the firmware but installing 3124 driver > on Windows was just too difficult for me and I gave up after thirty > painful minutes. Anyone up for testing? > > * Again, Sil4726 doesn't like ahci driver initialization sequence. > During driver initialization, the whole controller is reset and the > PHY seems to go offline and back online. Sil4726 doesn't like it > and goes out for lunch for several secs after that which makes > initial probing think that the port is empty. After probing is > complete, Sil4726 comes back and gets detected but at that point > detection is asynchronous and if boot device is on one of fan-out > ports, it can be a problem. Again, this might have been fixed by > new firmware. Anyone up for testing? > > * The long boot delay on ASUS boards with on-board PMP chips should be > fixed by this patchset but I don't have such a board or Sil4723 > which is used on those baords, so I'm not sure. Anyone up for > testing? > > This patchset is against > > libata-dev#upstream (4e1ae96828f6cee7b89ab8ca474c150fe211afd8) > + [1] misc-updates patchset > + [2] libata-link patchset, take 4 > + [3] libata-pmp-prep patchset, take 4 patchset overall seems OK to me