[U-Boot] [PATCH 5/7] OMAP5: EMIF: Add support for DDR3 device

[R, Sricharan]

Hi Tom,

>>>>  ah, this is what is there on OMAP5.
>>>>    3 for DDR3
>>>>    4 for LPDDR2-S4,
>>>>    5 for LPDDR2-S2
>>>
>>>
>>>
>>> 4/5 are listed as reserved here :( http://www.ti.com/lit/pdf/spruh73
>>>
>>>
>>>> Atleast DDR3 encoding is same. So we can differentiate bw DDR3 and 2
>>>> in same way.
>>>> Is the reset value set correctly on am335x according to the device
>>>> connected?
>>>
>>>
>>>
>>> I'll have to do some checking as the code doesn't check that today.
>>>
>>  The reset value reflects the connected device correctly for lpddr2 and
>> ddr3.
>>  Is that the same case in am33xx platforms  ?
>
>
> Yes, it's 3 on DDR3 and 2 on (non-LP) DDR2.  I spent some time yesterday
> starting on splitting up emif-common (am33xx does not have DMM) and making
> am33xx link at least with the common code.  My idea is to add a few more
> defines and make that part of do_sdram_init be:
> if (!in_sdram) {
>  switch(reg_sdram_type) {
> #ifdef CONFIG_EMIF4_DDR2
>    case 2:
>      ddr2_init();
>      break;
> #ifdef CONFIG_EMFI4_DDR3
>    case 3:
>      ddr3_init();
>      break;
> #endif
> #ifdef CONFIG_EMIF4_LPDDR2
>    case 4:
>    case 5:
>      lpddr2_init();
>      break;
> #endif
>    default:
>      panic("Unsupported DDR type connected to EMIF4");
>  }
> }
>
> And move lpddr2_init to emif4-lpddr2.c, and so forth.  My experimenting
> yesterday lead me to conclude that EMIF_MOD_ID_REGISTER behaves as is
> documented for am33xx on omap4/5 so what we're doing should be feasible.

 ok, and there are only few omap checks that are in the common lpddr2
and ddr3 init
paths, that we should be able to get rid of, so we can have generic
lpddr2_init and
ddr3_init.  But anyways as you said we can do all these only after the
current code
works for DDR3 on your am33xx.

>  I'm going to try and locally make the ddr3 patches work on the am33xx DDR3
> board as a way to prove that out and if it works, and we're fine with what
> I've laid out above, start working towards that layout.
>
 yeah, sounds correct.
  How about something like this, similar to the above
#define DDR2 1
#define LPDDR2 2
#define DDR3 3
#define LPDDR3 4

and soc specific code returns the correct ddr type. For OMAP
get_ddr_type()
{
 switch(EMIF_SDRAM_CONFIG & SDRAM_TYPE) {
 case 3:
     return DDR3;
 case 4:
 case 5:
 default:
    return LPDDR2;
}
We can avoid the #ifdef s in the emif code and use the type returned by
SOC to do the required initialisations. ?

Thanks,
 Sricharan