[U-Boot] [PATCH 1/3] i2c: add i2c_core and prepare for new multibus support

[Heiko Schocher]

Hello Simon,

Simon Glass wrote:
> Hi Heiko,
> 
> On Wed, Jan 18, 2012 at 10:35 PM, Heiko Schocher <hs at denx.de> wrote:
>> Hello Wolfgang, Timur, Simon,
>>
>> Wolfgang Denk wrote:
>>> Dear Simon Glass,
>>>
>>> In message <CAPnjgZ2kLPfMzzwgHkDJ4eL+wixqLv89+CvdVP7PCcy+XFaqNQ at mail.gmail.com> you wrote:
>>>>> I was really hoping we could get rid of the concept of a "current" i2c
>>>>> adapter, and just force all drivers to specify the I2C adapter they
>>>>> want to use for a given I2C operation. =A0That's how Linux operates, and
>>>>> it will prevent stuff like this:
>>>> I agree completely, it was one of the things I was going to ask for.
>>>> We should add a new parameter to calls instead IMO.
>>> Let's do one step at a time.  Now we go for multi-bus support.
>> Ok.
>>
>>> Implementing a new, better device interface is one of the next steps,
>>> then.
>> Some thoughts to the subject "get rid of concept of a current i2c"
>>
>> - First, it would be great to get rid of that ;-)
>>
>> - 2 reasons why we currently need the info, what is the current
>>  i2c adapter/i2c bus:
>>
>> - U-Boot principle says, don't init a device, if you don't use it.
>>  So, if we switch to another i2c adapter or i2c bus (A i2c bus is a
>>  combination if one i2c adpater and n i2c muxes), we must deinit
>>  the current adapter/bus. Ok, current code didn't this too, but
>>  this should a goal to keep in mind ... or we define, that this
>>  is not needed.
>>
>> - If we have i2c muxes, and you don't know your current i2c bus,
>>  you must on every i2c access also setup the i2c muxes on this
>>  i2c bus ... would we do this really?
> 
> Ignoring muxes, we can have more than one i2c adaptor inited at a time
> (i.e. de-initing is optional).

Ok for the point "more adapters inited at one time".

But we have to setup the muxes every time! We cannot ignore this.

> Perhaps reword this slightly. U-Boot can have knowledge of a current
> adaptor, mux settings and so on, and use this internally within the
> i2c layer to optimise performance and redundant i2c traffic. But the

Ok, thats what I mean. The "cur_i2c_bus" pointer should only be used
in the i2c subsystem! This pointer is in gd_t because it must be
writeable, also when running from flash ...

> pain is when the concept of a 'current adaptor' is exposed outside the
> i2c layer.

The "cur_i2c_bus" pointer is only used inside the i2c subsystem.

>>  if we have the current i2c bus info, we just have to check, if
>>  we are on this bus, and do the i2c access ... if we are not on
>>  this i2c bus, we can deinit it complete (the new i2c_core disconnects
>>  for example all i2c muxes on the i2c bus) and init the new bus.
>>  All this work is done in a central function i2c_set_bus_numer()
> 
> Yes but this function could become internal perhaps.

Yes, if we make the change, Timur suggested ... but that can be done
easy in a second step, as Wolfgang mentioned.

>> - Looking in the new multibus i2c_core.c file, we should get rid of
>>
>>  static unsigned int i2c_cur_bus __attribute__((section(".data"))) =
>>                                CONFIG_SYS_SPD_BUS_NUM;
>>
>>  and "cur_adap" renamed to "cur_i2c_bus" and should be a
>>  "struct i2c_bus_hose" pointer.
> 
> Sounds good. Heiko do you have time to take over your two patches I
> sent and tidy them up, or should I continue?

I did some work on this 2 patches, to get it working with soft_i2c.c
driver ... I try to rework this suggestions in, and send it to the
ML as a v2 ... if I do not find time, I speak up ;-)

bye,
Heiko
-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany