From mboxrd@z Thu Jan  1 00:00:00 1970
From: arnd@arndb.de (Arnd Bergmann)
Date: Tue, 04 Nov 2014 11:22:48 +0100
Subject: [PATCH v2] ARM: BCM5301X: set customized AUXCTL
In-Reply-To: <54581230.9090205@gmail.com>
References: <1414851912-30630-1-git-send-email-hauke@hauke-m.de>
 <54580CD9.3050309@hauke-m.de> <54581230.9090205@gmail.com>
Message-ID: <2789449.fsn4D0imX4@wuerfel>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

On Monday 03 November 2014 15:39:28 Florian Fainelli wrote:
> On 11/03/2014 03:16 PM, Hauke Mehrtens wrote:
> > On 11/03/2014 11:00 AM, Arnd Bergmann wrote:
> >> On Saturday 01 November 2014 15:41:38 Hauke Mehrtens wrote:
> >>>  };
> >>>  
> >>>  DT_MACHINE_START(BCM5301X, "BCM5301X")
> >>> -       .l2c_aux_val    = 0,
> >>> +       .l2c_aux_val    = L310_AUX_CTRL_CACHE_REPLACE_RR |
> >>> +                         L310_AUX_CTRL_DATA_PREFETCH |
> >>> +                         L310_AUX_CTRL_INSTR_PREFETCH |
> >>> +                         L310_AUX_CTRL_EARLY_BRESP,
> >>
> >> Please put these into the dts file instead if the boot loader fails
> >> to set them correctly. We should not override them from source code
> >> for new platforms any more.
> > I checked again and I only need L310_AUX_CTRL_DATA_PREFETCH and
> > L310_AUX_CTRL_INSTR_PREFETCH, the others get activate by the existing
> > code. Should I make it possible to give l2c_aux_val through dt, or
> > should I create some boolean options for  L310_AUX_CTRL_DATA_PREFETCH
> > and  L310_AUX_CTRL_INSTR_PREFETCH?

Certainly not the entire l2c_aux_val, what I meant is to use
boolean options for the bits you need. I was under the impression
that by now we had defined bindings for all the bits that are required
on any platform.

> Last we talked about these prefetch bits with Russell, I think we kind
> of agreed that they could probably be enabled by default
> 
> There is already a large number of properties defined in
> Documentation/devicetree/bindings/arm/l2cc.txt, I suppose that having
> prefetch properties could work too.

I can't really help with this if we don't have a binding. Proving that
it's always correct to enable these seems better than defining a
binding.

	Arnd