[U-Boot-Users] ARM cleanup - was: [GIT PULL] [ARM] Please pull fromgit://denx.de/git/u-boot-arm.git

[Peter Pearse]

Wolfgang

	I've got a bit lost here - can you explain a bit more. 
please excuse me if the following seems a bit basic - I just want to
understand
what I need to clean up.

You said:
--- snip ---
> > I'm afraid I don't understand. We are not talking about  any startup 
> > code,  but  about the code of the "reset" function, i. e. what 
> > casues the board to reset.
--- snip ---

I can't find a reference to the reset function.

There is do_reset() and set_cpu(), which the arm code supplies.
The only use of reset I can see is as a label in the ARM start.S.

The standard arm entry point to the U-Boot image is something like:-

.globl _start
_start:	b reset
	ldr	pc, _undefined_instruction
	ldr	pc, _software_interrupt
	ldr	pc, _prefetch_abort
	ldr	pc, _data_abort
	ldr	pc, _not_used
	ldr	pc, _irq
	ldr	pc, _fiq

which is a (never currently used) imitation of the ARM architecture vector
table.

So when U-Boot runs it jumps to code of the form

reset:
	Code to set the cpu to SVC32 mode

#ifndef CONFIG_SKIP_LOWLEVEL_INIT
	Code to e.g. remap RAM
#endif

#ifndef CONFIG_SKIP_RELOCATE_UBOOT
	Copy U-Boot code to RAM
#endif	/* CONFIG_SKIP_RELOCATE_UBOOT */

	Code to 
		set up the stack
		clear BSS
		
	Jump to board.c::start_armboot()

--- snip ---
>> The writers (and cut & pasters) of ARM code seem to have assumed that
>> reset:
>> is the entry to the code which runs immediately after the processor is 
>> reset, rather than the code which makes the processor reset.

> Well, if that was the assumption, that all this  code  has  obviously
never  been  tested  on any real hardware. 

>> Could we accept Ulf's code whilst I give this more consideration?

> It's so utterly wrong that I think we have to clean  this  up  before the
next  release. 
> If nobody submits a better patch, we will have to remove the bogus code.
--- snip ---

Is there anything that needs changing about the process I have described
above?

I could add ifdeffed code to really install the vector table at the ARM
reset point(s),
but it seems to me that the code will work OK as it is.

I do need to work with Ulf to put his code down into cpu_init_crit.
Both he & I missed the fact that CONFIG_BOOTBINFUNC got split & 
changed into the two CONFIG_SKIP defines.

Regards

Peter