[PATCH v2 4/6] ARM: tegra: rework fuse.c

[Peter De Schrijver]

On Tue, Jan 07, 2014 at 09:47:05PM +0100, Stephen Warren wrote:
> On 01/07/2014 07:10 AM, Peter De Schrijver wrote:
> > On Mon, Jan 06, 2014 at 09:50:42PM +0100, Stephen Warren wrote:
> >> On 12/24/2013 06:32 AM, Peter De Schrijver wrote:
> >>> Reduce fuse.c to the minimum functionality required for the early bootstages.
> >>>
> >>> Also export tegra_read_straps() for use by the fuse driver.
> 
> >>> diff --git a/arch/arm/mach-tegra/fuse.c b/arch/arm/mach-tegra/fuse.c
> 
> >>> -int tegra_sku_id;
> >>> -int tegra_cpu_process_id;
> >>> -int tegra_core_process_id;
> >>>  int tegra_chip_id;
> >>> -int tegra_cpu_speedo_id;		/* only exist in Tegra30 and later */
> >>> -int tegra_soc_speedo_id;
> >>>  enum tegra_revision tegra_revision;
> >>
> >> It's a bit odd to remove most of this, but leave a few parts hanging
> >> around. Wouldn't it be better to the drivers/misc/fuse code to export
> >> this, so that /all/ the fuse logic was there, rather than part of it
> >> being left over in arch/arm/? We'll need to fix that up anyway when we
> >> start using these globals on ARMv8, so may as well get it right now.
> >> Also, I rather think that the new drivers/misc/fuse code shouldn't be a
> >> module or driver, so that we can guarantee it's always there to provide
> >> the globals and that they are initialized early enough...
> > 
> > tegra_revision is used in tegra_dt_init() to initialize soc_dev_attr->revision
> > Hence this needs to be available before the fuse driver is initialized.
> 
> Yes, the same for tegra_chip_id too.
> 
> My point is: Why not move all the globals into the fuse driver, and make
> an early call to that fuse driver to initialize all these globals.
> Basically, rework this patch series to simply move the code to
> drivers/misc/fuse/, and keep initializing it by function call rather
> than as a driver probe(). The code can still scan DT to get the required
> reg/clock/... resources, in a similar fashion to e.g. the Tegra timer or
> cpufreq drivers IIRC. Perhaps the sysfs exports could be associated with
> a driver still though - just initialize the globals early?

That's probably better indeed... Will look into that.

Cheers,

Peter.