From mboxrd@z Thu Jan 1 00:00:00 1970 From: Hans de Goede Subject: Re: [PATCH v10 0/8] ARM: berlin: add AHCI support Date: Sat, 19 Jul 2014 12:31:19 +0200 Message-ID: <53CA48F7.5020405@redhat.com> References: <1405686607-8126-1-git-send-email-antoine.tenart@free-electrons.com> <20140718135758.GB13012@htj.dyndns.org> <53CA4608.3040208@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <53CA4608.3040208@redhat.com> Sender: linux-ide-owner@vger.kernel.org To: Tejun Heo , =?ISO-8859-1?Q?Antoine_T=E9nart?= Cc: sebastian.hesselbarth@gmail.com, kishon@ti.com, alexandre.belloni@free-electrons.com, thomas.petazzoni@free-electrons.com, zmxu@marvell.com, jszhang@marvell.com, linux-arm-kernel@lists.infradead.org, linux-ide@vger.kernel.org, devicetree@vger.kernel.org, linux-kernel@vger.kernel.org List-Id: devicetree@vger.kernel.org Hi, On 07/19/2014 12:18 PM, Hans de Goede wrote: > The problem is that: > > 1) We need to enable resources before we can do ahci_save_initial_config() > 2) We must do ahci_save_initial_config() before we can do ata_host_alloc_pinfo() > 3) Therefor we don't have port_info at enable_resources time, which is when we > want to enable the phys (and we cannot just enable the phys elsewhere as > enable_resouces gets used on e.g. resume too). > > So I think it is best to just make the phy pointers an array inside > ahci_host_priv, with a comment that the array indexes must match port > indexes. So looking at "[PATCH v10 4/8] ata: libahci: allow to use multiple PHYs" I see that currently the phy array indexes do not necessarily match the port indexes. Since you already allocate the phys array at nports size, I suggest simply making the array sparse, leaving in NULL entries for unused ports, and adjusting enable / disable_phys to check for NULL pointers. This way we still have a 1:1 way to map ports <-> phys if we want to do something with phys on a per port basis in the future. Note please also add a check that reg < nports so that we don't use the array out of bounds if there is an error in the dts. Regards, Hans