[U-Boot] [PATCH v2] arm: cm4008, cm41xx: set gd->ram_size in dram_init

[Yann Vernier]

On Thursday 05 July 2012 16:48:22 you wrote:
> On 05.07.2012 14:11, Yann Vernier wrote:
> > Leave dram_init_banksize to set up the bank info data.
> > ram_size was previously uninitialized. Also, generalize
> > lowlevel assembly to support more RAM options.
> > 
> > Signed-off-by: Yann Vernier <yann.vernier at orsoc.se>
> > ---
> > 
> > Changes for v2:
> >    - Update to use CONFIG_SYS_SDRAM_ constants
> >    - Update cm41xx also
> >    - Map SDRAM to match configuration
> > 
> > ---
> > 
> >  arch/arm/cpu/arm920t/ks8695/lowlevel_init.S |    8 +++-----
> >  board/cm4008/cm4008.c                       |    5 +++--
> >  board/cm41xx/cm41xx.c                       |    5 +++--
> >  include/configs/cm4008.h                    |    5 ++---
> >  include/configs/cm41xx.h                    |    5 ++---
> >  5 files changed, 13 insertions(+), 15 deletions(-)
> > 
> > diff --git a/arch/arm/cpu/arm920t/ks8695/lowlevel_init.S
> > b/arch/arm/cpu/arm920t/ks8695/lowlevel_init.S index e9f1227..df13de6
> > 100644
> > --- a/arch/arm/cpu/arm920t/ks8695/lowlevel_init.S
> > +++ b/arch/arm/cpu/arm920t/ks8695/lowlevel_init.S
> > 
> > @@ -131,11 +131,9 @@ highflash:
> >  	 * before relocating, we have to setup RAM timing
> >  	 */
> >  	
> >  	ldr	r1, =(KS8695_IO_BASE+KS8695_SDRAM_CTRL0)
> > 
> > -#if (PHYS_SDRAM_1_SIZE == 0x02000000)
> > -	ldr	r2, =0x7fc0000e		/* 32MB */
> > -#else
> > -	ldr	r2, =0x3fc0000e		/* 16MB */
> > -#endif
> > +	/* 8 column address bits, 4 banks, 32 bits data width */
> > +	ldr	r2,
> > =((CONFIG_SYS_SDRAM_BASE+CONFIG_SYS_SDRAM_SIZE-0x10000)<<(22-16) | \ +		
> >      (CONFIG_SYS_SDRAM_BASE>>(16-12) | 0x00e))
> 
> ugh ... magic. Will it work for every possible setting?
> How about one setting CONFIG_SYS_SDRAM_BASE != 0x0? I the base address
> related to register content in any way?
> 
> Would be great to have a little documentation here why it is shifted
> that way, unfortunately I can not find a KS8695 spec in the net. Don't
> get me wrong it is ok to do so. I think it is way better than hard
> coding the values like before but would be great to understand why it is
> done that way.
> 
> Best regards
> 
> Andreas Bießmann


It's not a guarded secret, although it is far from obvious where to find it. 
One way is micrel.com - Products - Ethernet ICs / ARM based SOC's - HW Design 
Kit. Within that zip file, 
KS8695X_EVAL_HW_RV4.0_DP/RegDescription/KS8695X Register Description v1.1.pdf

My personal thought on the matter is that the existing code is rather heavy on 
magic numbers, ergo the comment once I figured out what the first ROM 
reconfiguration was even for. 


The memory mapping registers allow you to reconfigure the address recognition 
for each memory region. The low 16 bits are not configurable, so you can only 
relocate to 64KiB aligned addresses. Bits 25-16 are configured as an inclusive 
range, with the top value in bits 31-22 of the register, and the bottom in 
bits 21-12. The lower bits configure things like SDRAM geometry or SRAM/ROM 
access times, with one value meaning disabled. 

Come to think of it, I fear I've miscalculated where the high region actually 
is. That comment in lowlevel_init should read 48MB and 32MB, not 768 and 512. 
I'll fix that and insert a little more explanatory text.