[PATCH] bcm53xx: initial support for the BCM5301/BCM470X SoC with ARM CPU

[Domenico Andreoli]

On Wed, Jul 24, 2013 at 09:21:43PM +0200, Hauke Mehrtens wrote:
> On 07/24/2013 02:44 AM, Domenico Andreoli wrote:
> > On Wednesday, July 24, 2013, Hauke Mehrtens <hauke at hauke-m.de
> > <mailto:hauke at hauke-m.de>> wrote:
> >> On 07/19/2013 03:36 AM, Domenico Andreoli wrote:
> >>> On Tue, Jul 16, 2013 at 05:35:21PM +0200, Hauke Mehrtens wrote:
> >>>> On 07/16/2013 05:20 PM, Thomas Petazzoni wrote:
> >>>>>
> >>>>>> diff --git a/arch/arm/mach-bcm53xx/Kconfig
> > b/arch/arm/mach-bcm53xx/Kconfig
> >>>>>> new file mode 100644
> >>>>>> index 0000000..1e16e87
> >>>>>> --- /dev/null
> >>>>>> +++ b/arch/arm/mach-bcm53xx/Kconfig
> >>>>>> @@ -0,0 +1,10 @@
> >>>>>> +config ARCH_BCM53XX
> >>>>>> +  bool "Broadcom BCM47XX / BCM53XX ARM SoC"
> >>>>>
> >>>>> So the directory is named mach-bcm53xx, but you also handle BCM47xx
> >>>>> SoCs. This doesn't sound really easy to follow.
> >>>
> >>> At the time of the BCM281XX merge we considered that such directories
> > would
> >>> mostly contain board files only, being these new entries DT based. Hence
> >>> the choice of mach-bcm to collect all of them.
> >>>
> >>> I think you should then put this stuff there.
> >>
> >> So you think I should move the file from
> >> arch/arm/mach-bcm53xx/bcm53xx.c to arch/arm/mach-bcm/bcm53xx.c ?
> > 
> > yes
> > 
> > this looks more comsistent with the actual soc name:
> > arch/arm/mach-bcm/bcm530xx.c
> > 
> > but I find also acceptable the complete name of the "parent" soc (as I'm
> > doing with the bcm4760), so: arch/arm/mach-bcm/bcm53010.c
> 
> I do not think these Broadcom ARM SoCs (bcm4760, BCM5301X, bcm11351)
> have more in common than the vendor name, so I do not think it is a good
> idea to place them all at mach-bcm.

In an ideal DT-only world (as basically is for Broadcom ARM SoCs), whatever
two or more SoCs share can hopefully modelled in a driver and as such
would go in the drivers/ subtree. What's left is really SoC specific and,
again hopefully, very minimal.

There should not be any SoC so weird to require a whole subdirectory full
of that SoC specific stuff. So grouping everything by vendor name looks
quite appealing to me.

Sharing the same subdir requires people working in it to talk and find
agreements some more than the sparsely populated subdir did (because there
is not a single maintainer who owns it). I think this is a big advantage
Broadcomers can start with right now.

> 
> >> Most (All?) of the Router SoCs in the BCM53XX and BCM47XX line are from
> >> the same family and they should be supported by the same code, so there
> >> is no reason to have a bcm47xx.c and a bcm53xx.c.
> >>
> >>>> Yes the BCM53XX and BCM47XX SoCs are technically from the same line. I
> >>>> do not know why there are two different names, probably marketing.
> >>>>
> >>>> Earlier versions of these SoC lines (also BCM47XX and BCM53XX) used a
> >>>> MIPS core and they are supported by arch/mips/bcm47xx/. I use BCM53XX to
> >>>> not conflict with the MIPS part now. I know this could still cause
> >>>> problems and people will get confused, but I do not know a better name.
> >>>
> >>> I'll throw also BCM476x (bcm4760 and bcm4761) on the table. These are
> > ARM11,
> >>> single-core SoCs.
> >>>
> >>> Can we agree on a nice naming so that we actually clarify this silliness?
> >>
> >> We could name it BCM5301X like it is named the the vendor source code.
> >> Currently there are the following SoCs of this family available:
> >> BCM53010, BCM53011, BCM53012, BCM53014, BCM53015, BCM53016, BCM53017,
> >> BCM53018, BCM4707, BCM4708 and BCM4709.
> > 
> > There is not a clear single pattern here, whatever choice won't be 100%
> > right. Users will need to read some help to pick the right choice.
> > BCM53010? But mine is not a strong opinion.
> > 
> >> Internal there are two different
> >> chip IDs available 53010 (BCM4707) and 53018 (BCM53018).
> > 
> > what do you mean with "internal" here?
> 
> There is some register in the chip where the chipid is stored, according
> to the vendor code these two values are currently possible. This was
> changed for every generation with the MIPS SoCs, while one generation
> had some SoCs.

Only two chipids for all these SoCs? Not that changes anything to me,
I'm only trying to understand the beast.

> 
> Hauke