From mboxrd@z Thu Jan  1 00:00:00 1970
From: tchalamarla@caviumnetworks.com (Tirumalesh Chalamarla)
Date: Tue, 16 Feb 2016 11:50:44 -0800
Subject: [PATCH V2] AHCI: Workaround for ThunderX Errata#22536
In-Reply-To: <56C37AD3.4030407@caviumnetworks.com>
References: <1455319230-30201-1-git-send-email-tchalamarla@caviumnetworks.com>
 <20160214170152.GC3965@htj.duckdns.org> <56C147B2.9010303@caviumnetworks.com>
 <20160215183041.GH3965@htj.duckdns.org>
 <20160216144250.GD31343@rric.localdomain>
 <56C37524.5090000@caviumnetworks.com> <56C37AD3.4030407@caviumnetworks.com>
Message-ID: <56C37D94.8080403@caviumnetworks.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org



On 02/16/2016 11:38 AM, David Daney wrote:
> On 02/16/2016 11:14 AM, Tirumalesh Chalamarla wrote:
>>
>>
>> On 02/16/2016 06:42 AM, Robert Richter wrote:
>>> On 15.02.16 13:30:41, Tejun Heo wrote:
>>>> On Sun, Feb 14, 2016 at 07:36:18PM -0800, Tirumalesh Chalamarla wrote:
>>>>> There is no need for special Driver, AHCI is sufficient for
>>>>> ThunderX, the
>>>>> file only contains this interrupt handler,
>>>>> is it preferable if this interrupt handler in libahci.c with others,
>>>>> instead
>>>>> of separate file?
>>>>
>>>> Yeap, just fold it in ahci.c with surrounding #ifdef guard.
>>>
>>> Yes, please use #ifdef CONFIG_CAVIUM_ERRATUM_22536 ... and add a
>>> kconfig entry for this to arch/arm64/Kconfig.
>>>
>> Are you sure, this is not a workaround that is based on alternative
>> framework rather on pci device and vendor
>>
>> do you think CONFIG_ARCH_THUNDER a good alternative?
>
> No.  CONFIG_ARCH_THUNDER should be removed all together.
>
> Grouping a bunch of unrelated features under a single config variable
> creates a very brittle system.  What are you going to do when a new
> hardware revision is released?  Create CONFIG_ARCH_THUNDER2?  Which one
> of these two would you select if building a kernel?  It is a choice that
> we don't want to force users (kernel builders) to have to waste mental
> energy on.
>
> Instead, let's try to make things work out of the box without having to
> set a bunch of random config variables.
>
> If a generic arm64 kernel won't get too bloated, I would suggest just
> enabling the compilation of the code unconditionally (at least for
> arm64).  The use of the code would still be gated by the PCI version
> probe that is part of the patch.
>

exactly, that is my initial choice with v1, and only depends on vendor 
and device id.

but it seems a config is needed. how about ARCH_ARM64 then?

Thanks,
Tirumalesh.

> David Daney
>