[PATCH] ARM: AM33XX: board-generic: Add of_dev_auxdata to fix dev_id for CAN module

[Hiremath, Vaibhav]

On Wed, Aug 08, 2012 at 07:38:01, Rob Herring wrote:
> On 08/07/2012 10:53 AM, Hiremath, Vaibhav wrote:
> > On Tue, Aug 07, 2012 at 20:49:48, Rob Herring wrote:
> >> On 08/07/2012 08:37 AM, Vaibhav Hiremath wrote:
> >>> If the module requires only one clocksource to function, then
> >>> driver can very well call clk_get() api with "con_id = NULL",
> >>>
> >>>        clk = clk_get(dev, NULL);
> >>>
> >>> And platform code should respect this and return valid clk handle.
> >>> That means, now the clk_get() api would rely on dev_id, and platform
> >>> code should either have clk node with matching dev_id (with con_id=NULL)
> >>> or return dummy clk node.
> >>>
> >>> With DT support, we can fix the dev_id for particular module
> >>> using "struct of_dev_auxdata" to satisfy above clk framework requirement.
> >>>
> >>> In case of AM33XX, we required this for the DCAN module, so this
> >>> patch adds "of_dev_auxdata" for both DCAN_0/1 instances.
> >>>
> >>> Signed-off-by: Vaibhav Hiremath <hvaibhav at ti.com>
> >>> Cc: Tony Lindgren <tony at atomide.com>
> >>> Cc: Paul Walmsley <paul at pwsan.com>
> >>> Cc: Benoit Cousson <b-cousson at ti.com>
> >>> Cc: Grant Likely <grant.likely at secretlab.ca>
> >>> ---
> >>> This patch is boot tested on BeagleBone platform, and checked for
> >>> clk_get() return value in d_can module driver.
> >>>
> >>>  arch/arm/mach-omap2/board-generic.c |   18 +++++++++++++++++-
> >>>  1 files changed, 17 insertions(+), 1 deletions(-)
> >>>
> >>> diff --git a/arch/arm/mach-omap2/board-generic.c b/arch/arm/mach-omap2/board-generic.c
> >>> index 6f93a20..c9b7903 100644
> >>> --- a/arch/arm/mach-omap2/board-generic.c
> >>> +++ b/arch/arm/mach-omap2/board-generic.c
> >>> @@ -37,11 +37,27 @@ static struct of_device_id omap_dt_match_table[] __initdata = {
> >>>  	{ }
> >>>  };
> >>>
> >>> +/*
> >>> + * Lookup table for attaching a specific name and platform_data pointer to
> >>> + * devices as they get created by of_platform_populate(). Ideally this table
> >>> + * would not exist, but the current clock implementation depends on some devices
> >>> + * having a specific name OR to satisfy NULL con_id requirement from driver.
> >>> + */
> >>> +static const struct of_dev_auxdata am33xx_auxdata_lookup[] __initconst = {
> >>> +	OF_DEV_AUXDATA("bosch,d_can", 0x481cc000, "d_can.0", NULL),
> >>> +	OF_DEV_AUXDATA("bosch,d_can", 0x481d0000, "d_can.1", NULL),
> >>> +	{ },
> >>> +};
> >>
> >> Adding an additional clkdev lookup accomplishes the same thing and is a
> >> cleaner solution.
> >>
> > 
> > That is also required and I have submitted patch for the same -
> > 
> > http://www.spinics.net/lists/arm-kernel/msg187998.html
> 
> That only adds the non-DT name. What I'm saying is you can have 2 lookup
> names for the same clock.
> 
> > 
> > With DT support, I am getting dev_id as -
> > 
> >  - Without "reg" property: d_can.16 and d_can.17,
> >    (I believe it changes dynamically here)
> > 
> >  - With "reg" property: 481cc000.d_can and 481d0000.d_can
> > 
> > Which is not so intuitive, I would like to see them as per Spec/TRM, so 
> > doesn't this patch make sense?
> 
> The spec doesn't have a base address for the module? This name is going
> to get used in sysfs anyway, and the old name should be going away.
> 

This is certainly going to change user interface change, and I thought we 
should not change the user interface irrespective of whether it is DT or 
non-DT.
Few points, which I do care about, are,

1. request_irq():
   For the driver if I am using dev_name(dev) as for the 3rd argument, which 
   will now different with DT support - brings change in user interface.

   Non-DT:
    #cat /proc/interrupts
               CPU0
    52:          0      INTC  xxx0
    55:          0      INTC  xxx1

   DT:
    #cat /proc/interrupts
               CPU0
    52:          0      INTC  481cc000.xxx
    55:          0      INTC  481d0000.xxx


2. SYSFS:
   For example,
   Without auxdata:
     /sys/devices/481cc000.xxx/
     /sys/devices/481d0000.xxx/

   With auxdata:
     /sys/devices/xxx.0/
     /sys/devices/xxx.1/

   From user perspective, I still would not want to change the existing 
   naming convention.

3. Certainly I don't want user to go back to spec everytime to find out 
   which address is mapped to which instance of module.
   Also, in addition to that, in case of SoC's like AM33xx (for that matter, 
   any embedded SoC) we have different domains and modules are placed in 
   different domains as per use-cases.
   For example, in this case, can0 is places in wakeup domain and can1 is 
   placed in per-domain. Some boards might use only one instance or some 
   board might use both of them.

   I still would like to keep instance number wherever user interface comes 
   into picture and populate base address as an additional info to user.


I may be missing something, may be I am not able to see the bigger picture 
here.

Grant/Benoit, May be you can conform and put your opinion on this.

Thanks,
Vaibhav