Hi Simon, Fabio, Stefano, Marek, Albert
On Fri, Mar 2, 2012 at 8:28 AM, Simon Glass sjg@chromium.org wrote:
+Graeme
Hi,
On Thu, Mar 1, 2012 at 5:23 AM, Fabio Estevam festevam@gmail.com wrote:
Hi,
Currently CONFIG_ARCH_CPU_INIT is used to select arch_cpu_init() function.
arch_cpu_init() does CPU level initialization, so why do we need to include CONFIG_ARCH_CPU_INIT in the include/configs/boardXYZ files, which are board related files ?
For example:
Let's say boards X, Y and Z are based on SoC S:
- If processor S has a arch_cpu_init() defined, then it means that
X,Y,Z need the code from arch_cpu_init() and then we need to define CONFIG_ARCH_CPU_INIT for each of these boards (actually all the boards based on this processor would need CONFIG_ARCH_CPU_INIT)
- If not all boards need the code inside arch_cpu_init() for
processor S, then it means that this code is not really CPU specific and then it should be moved to board code.
I was thinking in doing the following:
--- a/arch/arm/lib/board.c +++ b/arch/arm/lib/board.c @@ -224,10 +224,15 @@ void __dram_init_banksize(void) void dram_init_banksize(void) __attribute__((weak, alias("__dram_init_banksize")));
+int __arch_cpu_init(void) +{
- return 0;
+} +int arch_cpu_init(void)
- __attribute__((weak, alias("__arch_cpu_init")));
init_fnc_t *init_sequence[] = { -#if defined(CONFIG_ARCH_CPU_INIT) arch_cpu_init, /* basic arch cpu dependent setup */ -#endif #if defined(CONFIG_BOARD_EARLY_INIT_F) board_early_init_f, #endif
,so that CONFIG_ARCH_CPU_INIT is not needed anymore.
Before I go further in this route to remove CONFIG_ARCH_CPU_INIT from other places, I would like to know if this makes sense.
I consider arch_cpu_init() to be an architecture function, which should be defined in arch/arm/cpu/ somewhere. For tegra, the function is in arch/arm/cpu/armv7/tegra2/board.c
If it is a board function then it should be renamed to board_cpu_init() or similar.
So regarding your proposed change, I feel that the code size impact is small and it seems reasonable.
However, Graeme's forthcoming initcall mechanism may anyway may your change obsolete, since then architectures that need it can insert the call into the initcall sequence, and other archs need not.
I have been following this thread but have not had time to respond...
Simon's assesment is correct - My new INIT_CALL architecture (which I am still working on ... slowly ...) will move the visibility of CPU, SoC and arch specific initialisation away from the board configuration
It will also get rid of that #if defined(CONFIG_BOARD_EARLY_INIT_F) as well (and every other #ifdef in the init sequence arrays)
Finally, if we accept this change, then my generic board init series could/should be changed in the same way.
so it comes down to how to proceed - Live with the current implementation until INIT_CALL gets implemented (assuming it gets approved) or accept this proposed patch as a temporary 'solution' until INIT_CALL gets implemented
I think I should just post what I have of the INIT_CALL patches even though they are very rough and totally incomplete, at least you can all see how it is designed to work...
Regards,
Graeme