[U-Boot] [PATCH] 7/12 Multiadapter/multibus I2C, drivers part 4

[Heiko Schocher <hs@denx.de>]

Hello ksi,

ksi at koi8.net wrote:
> On Thu, 19 Feb 2009, Wolfgang Denk wrote:
> 
>> Dear ksi at koi8.net,
>>
>> In message <Pine.LNX.4.64ksi.0902191141090.18501@home-gw.koi8.net> you wrote:
[...]
>>> That means you'll have to rewrite the entire U-Boot. 99% of the boards have
>>> only one bus so they did not switch busses. That means they never called
>>> that i2c_set_bus_num() relying on i2c_init() in libxxx/board.c instead.
>> I cannot follow your argument.
>>
>> Yes, the status quo is as you describe, it relies on  i2c_init()  and
>> is   simple-minded   and  does  not  support  an  arbitry  number  of
>> arbitrarily complex I2C bus trees and multiplexors and expanders  and
>> what  else.  But  it  was  sufficient  for the first 10 years and 500
>> boards of U-Boot development.
>>
>> Now we are discussion a major redesign, so what is the big problem of
>> changing this part? "rewrite the entire U-Boot"? Please stay serious.
>> Compared to the other changes you suggest, this is  not  that  big  a
>> part.
> 
> No, my changes are limited. Look, somebody must initialize an adapter. As
> for now it is done with a single i2c_init() usually in libxxx/board.c. Then
> the entire code assumes adapter is already initialized and just issues
> i2c_read/write() as it see fits. 99% of this code is written on assumption
> that there is only one I2C bus so it doesn't use i2c_set_bus_num() or
> whatever, it just fires up i2c_read() and that's it.
> 
> This would perfectly work with my changes without modifying that code -- the
> only bus is bus number 0 so there is nothing wrong with not setting the bus
> for each I2C access; it is already at that only bus.
> 
> Now, if we have adapter initialization moved to i2c_set_bus() all that code
> will cease to work because i2c_set_bus() is never called thus adapter will
> never be initialized and all those i2c_read() and friends will fail.

But you can call i2c_set_bus_num instead of i2c_init in lib_xxx/board.c
and all old boards will work as they did. Ok, we didn;t can get rid of
initializing bus 0, but with moving init() in i2c_set_bus_num, we only
init additional hardwareadapters only, if we need them.

> That means that we should read through each and every board's code to find
> where i2c functions are used and add i2c_set_bus() calls as needed. That is
> not INSTEAD of that big rewrite, that is _IN ADDITION_ to it. That is a very
> sizeable chunk of additional changes.

No, we have not to do this. See above. And if some board use multibus,
yes they have to do a i2c_set_bus_num before they access the i2c bus,
because they must know on which bus they have to read, and the actual
bus could be changed.

> I DID think of adding adapter initialization to i2c_set_bus() initially but
> then it turned out it generated more problems than it solved (and it solved
> none) so I dropped that idea.

Please tell us the problems, so we can think of it, maybe I overlook something.
Thats the reason why we discuss this here.

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