Patchwork [U-Boot] remove (double) LED initialization in arm920t start.s

login
register
mail settings
Submitter Jens Scharsig
Date Dec. 18, 2010, 12:08 p.m.
Message ID <4D0CA427.1070706@scharsoft.de>
Download mbox | patch
Permalink /patch/76064/
State Accepted
Commit 009b54079e65ddf004041ccaccef3b8647ce8a0b
Delegated to: Albert ARIBAUD
Headers show

Comments

Jens Scharsig - Dec. 18, 2010, 12:08 p.m.
* remove LED initialization in front of relocation and bss init

Signed-off-by: Jens Scharsig <js_at_ng@scharsoft.de>
---

 * prevents run C function on an uninitialized environment
 
 arch/arm/cpu/arm920t/start.S |    3 ---
 1 files changed, 0 insertions(+), 3 deletions(-)

-- 1.7.1
Andreas Bießmann - Dec. 23, 2010, 11:58 a.m.
Dear Jens Scharsig,

Am 18.12.2010 um 13:08 schrieb Jens Scharsig:

> * remove LED initialization in front of relocation and bss init
> 
> Signed-off-by: Jens Scharsig <js_at_ng@scharsoft.de>
> ---
> 
> * prevents run C function on an uninitialized environment

You are right, we do not have stack/bss setup when we call these two c-functions.

But in that case it is ok to do so cause we do just pc relative accesses here, do not use bss variables and last we do not use the stack.

Beside the fact that we could use it here as it is done since atmel posted their at91rm9200 code we could ask if we need it or not.

I think it is a nice helper for initialisation debugging and I used it from time to time for my relocation investigations. But I think we do _not_ need it in every arm920t/at91 board for every bootup.
It should be
 a) enclosed in some ifdef to enable them conditionally on compile time
 b) documented in a README and completely removed (some lines later too)

Beside that one could state
 c) completely removed (the whole coloured_LED stuff) as this is atmel specific and other SoC use another interface to switch LED
if we decide to not use these LED features for bootup debugging.

AFAIK this is also used in arm926ejs (at91sam only), Reinhard can you please comment on that?

regards

Andreas Bießmann
Reinhard Meyer - Dec. 23, 2010, 12:29 p.m.
On 23.12.2010 12:58, Andreas Bießmann wrote:
> Dear Jens Scharsig,
>
> Am 18.12.2010 um 13:08 schrieb Jens Scharsig:
>
>> * remove LED initialization in front of relocation and bss init
>>
>> Signed-off-by: Jens Scharsig<js_at_ng@scharsoft.de>
>> ---
>>
>> * prevents run C function on an uninitialized environment
>
> You are right, we do not have stack/bss setup when we call these two c-functions.
>
> But in that case it is ok to do so cause we do just pc relative accesses here, do not use bss variables and last we do not use the stack.
>
> Beside the fact that we could use it here as it is done since atmel posted their at91rm9200 code we could ask if we need it or not.
>
> I think it is a nice helper for initialisation debugging and I used it from time to time for my relocation investigations. But I think we do _not_ need it in every arm920t/at91 board for every bootup.
> It should be
>   a) enclosed in some ifdef to enable them conditionally on compile time
>   b) documented in a README and completely removed (some lines later too)
>
> Beside that one could state
>   c) completely removed (the whole coloured_LED stuff) as this is atmel specific and other SoC use another interface to switch LED
> if we decide to not use these LED features for bootup debugging.
>
> AFAIK this is also used in arm926ejs (at91sam only), Reinhard can you please comment on that?

Quite funny, incidentally right this morning I tried to figure out how to use LEDs to assist
a "blackbox" sort of system that has only 3 RG LEDs to signal u-boot phases and errors before
kernel and application are ultimately started.

This part of u-boot is quite messy, look at "status_led.h" for example...

I agree the AT91 colored LED specialty should be removed, once we can agree on a useful
common LED approach.

And I think that should be working the other way.
It is pointless to try to factorize it into a few LEDs with specific meanings.

We should have one function like "show_status(int status)"
with various statuses defined like:
LED_STATUS_RESET, _BEFORE_RELOC, _AFTER_RELOC, _SENT_DHCP, _STARTING_KERNEL,
_CRASHED, ...

Then it is up to the board specific code to drive multicolored LEDs, an LCDisplay,
some seven-segment displays or even relais.

But I guess that's too futuristic, and as always there is a lot of existing code that
does it in various different ways :(

Best Regards,
Reinhard
Wolfgang Denk - Dec. 23, 2010, 12:51 p.m.
Dear Reinhard Meyer,

In message <4D1340BD.2090206@emk-elektronik.de> you wrote:
>
> We should have one function like "show_status(int status)"
> with various statuses defined like:
> LED_STATUS_RESET, _BEFORE_RELOC, _AFTER_RELOC, _SENT_DHCP, _STARTING_KERNEL,
> _CRASHED, ...

Keep the existing code, and just turn the current set of numbers into
an enum.

Best regards,

Wolfgang Denk
Reinhard Meyer - Dec. 23, 2010, 1:18 p.m.
Dear Wolfgang Denk,
> Dear Reinhard Meyer,
>
> In message<4D1340BD.2090206@emk-elektronik.de>  you wrote:
>>
>> We should have one function like "show_status(int status)"
>> with various statuses defined like:
>> LED_STATUS_RESET, _BEFORE_RELOC, _AFTER_RELOC, _SENT_DHCP, _STARTING_KERNEL,
>> _CRASHED, ...
>
> Keep the existing code, and just turn the current set of numbers into
> an enum.

What current code?

1. show_boot_progress() is quite close, but has hundreds of values...
Not sure a enum would be useful there, the values have gaps; rather an
errno.h like implementation makes sense there.

2. the LED current code (in some non-ARM) places
has LED types (BOOT, ...) and LED modes (OFF, BLINK, ON).
That is totally different and quite useless for seven segment and
simple LCDisplays.

Best Regards,
Reinhard
Wolfgang Denk - Dec. 23, 2010, 3:05 p.m.
Dear Reinhard Meyer,

In message <4D134C08.8090609@emk-elektronik.de> you wrote:
>
> > Keep the existing code, and just turn the current set of numbers into
> > an enum.
> 
> What current code?

I meant the existing show_boot_progress() code.

> 1. show_boot_progress() is quite close, but has hundreds of values...
> Not sure a enum would be useful there, the values have gaps; rather an
> errno.h like implementation makes sense there.

I think it's just an implementation detail wether you use an enum or
#defines here.

> 2. the LED current code (in some non-ARM) places
> has LED types (BOOT, ...) and LED modes (OFF, BLINK, ON).
> That is totally different and quite useless for seven segment and
> simple LCDisplays.

Right.

Note that there is more code when it comes to seven segment and
character based LCD displays - please see doc/README.LED_display,
include/led-display.h, common/cmd_display.c and
drivers/misc/pdsp188x.c on one side, or the highly board specific
display code using the same framework in board/a4m072/a4m072.c on the
other side.


Best regards,

Wolfgang Denk
Jens Scharsig - Dec. 23, 2010, 4:17 p.m.
To explain the background

1st:

I've compiled my code for easier debugging without -Os option.
But the non optimized code needs a valid stack at this position.
It's only as secondary problem

2nd:

The functions will called later again.

3rd:

Some soc are not fully intitialized here (e.g Clock aso). So, we also
increase the boot speed.

regards

Jens
Albert ARIBAUD - Jan. 20, 2011, 8:56 p.m.
Le 18/12/2010 13:08, Jens Scharsig a écrit :
> * remove LED initialization in front of relocation and bss init
>
> Signed-off-by: Jens Scharsig<js_at_ng@scharsoft.de>
> ---
>
>   * prevents run C function on an uninitialized environment
>
>   arch/arm/cpu/arm920t/start.S |    3 ---
>   1 files changed, 0 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm/cpu/arm920t/start.S b/arch/arm/cpu/arm920t/start.S
> index 08f178d..764a06a 100644
> --- a/arch/arm/cpu/arm920t/start.S
> +++ b/arch/arm/cpu/arm920t/start.S
> @@ -119,9 +119,6 @@ start_code:
>   	orr	r0, r0, #0xd3
>   	msr	cpsr, r0
>
> -	bl	coloured_LED_init
> -	bl	red_LED_on
> -
>   #if	defined(CONFIG_AT91RM9200DK) || defined(CONFIG_AT91RM9200EK)
>   	/*
>   	 * relocate exception table
> -- 1.7.1

Applied to U-boot-arm, thanks.

Amicalement,

Patch

diff --git a/arch/arm/cpu/arm920t/start.S b/arch/arm/cpu/arm920t/start.S
index 08f178d..764a06a 100644
--- a/arch/arm/cpu/arm920t/start.S
+++ b/arch/arm/cpu/arm920t/start.S
@@ -119,9 +119,6 @@  start_code:
 	orr	r0, r0, #0xd3
 	msr	cpsr, r0
 
-	bl	coloured_LED_init
-	bl	red_LED_on
-
 #if	defined(CONFIG_AT91RM9200DK) || defined(CONFIG_AT91RM9200EK)
 	/*
 	 * relocate exception table