[U-Boot] How do ARM platform initialize DDR?

[Tom Rini]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 09/19/2013 05:08 PM, York Sun wrote:
> On 09/19/2013 01:57 PM, Tom Rini wrote:
>> On Tue, Sep 17, 2013 at 08:59:24AM -0700, York Sun wrote:
>> 
>>> Albert,
>>> 
>>> Pardon me if this is a dumb question. I have been working on
>>> powerpc platforms in the past. Now we (the developers I work
>>> with) are exploring ARM cores. I am searching how memory is
>>> initialized and found different solutions. Some platforms have
>>> memory ready before u-boot even starts, some simply write to a
>>> set of registers. I understand many platforms don't share the
>>> IP of DDR controller. I am wondering if there is generic DDR
>>> driver used by many ARM platforms, like the one we have for 
>>> powerpc/mpc85xx SoCs.
>> 
>> Thinking back, as a rule of thumb, PowerPC has SPD I2C data 
>> available, usually.  That's not the rule for ARM.  One of a few
>> choices happen: 1) ROM sets up DDR. 2) U-Boot/SPL sets up the DDR
>> controller.
>> 
> 
> So for ARM platforms, the majority don't have the flexibility of
> using different DIMMs and/or clocks?

It's a different world.  Again, thinking back to the PowerPC boards
I've seen, they had "regular" DDR sockets.  Most ARM boards don't.
You can design your board with whatever, and I know in prototyping
folks make do swapping chips in and out (and if you look at the omap
code, you can see where we have code to calculate the timing values
and print them, or use provided hard-coded values).

>> The problem is that the DDR controller is usually
>> vendor-specific. Perhaps the flip-side here is that there's not
>> so much a generic DDR driver for mpc85xx but simply one vendor
>> for mpc85xx.  Taking arch/powerpc/cpu/mpc85xx/ddr-gen3.c as what
>> you're talking about, 
>> arch/arm/cpu/armv7/omap-common/emif-common.c would be an
>> ARM-world example (the 'EMIF' is found on a large variety of TI
>> parts, not just "omap" ones).
> 
> Does it make sense to share the Freescale DDR driver across ARM
> and Powerpc? Or does it make more sense to selectively copy the
> mpc8xxx DDR driver to Freescale ARM subfolder to start with. If the
> similarity sustains then we merge them into common driver. If not,
> we maintain two separated drivers.
> 
> For those who is not familiar with, Freescale is extending products
> to ARM cores. I am expecting peripherals stay relatively close, so
> many driver can be reused.

I've been wondering when this would start to be visible.  If we are
able to share the DDR controller code between mpc85xx and the ARM
stuff you're talking about, we'll have to sort out someplace within
drivers/ to place it, to avoid #include nightmares I suspect.  Other
drivers should be easier to share at least :)

- -- 
Tom
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=INiu
-----END PGP SIGNATURE-----