linux-next: fix vector initialisation when !CONFIG_KGDB

linux-next: fix vector initialisation when !CONFIG_KGDB

[Ben Dooks]

[-- Attachment #1: Type: text/plain, Size: 803 bytes --]

It seems in the changes for adding CONFIG_KGDB that the ARM
trap initialisation has been turned off if the configuration
is not enabled. If CONFIG_KGDB is not set, the system locks
up as soon as the kernel turns the IRQs on for the first
time in init/main.c.

The fix is to remove the dependency on calling early_trap_init()
on CONFIG_KGDB, so that the vectors are initialised before
the init sequence enables the IRQs.

Signed-off-by: Ben Dooks <ben-linux@fluff.org>

--- linux-2.6.26-rc6-q1.orig/arch/arm/kernel/setup.c	2008-06-18 15:17:31.000000000 +0100
+++ linux-2.6.26-rc6-q1/arch/arm/kernel/setup.c	2008-06-18 15:17:45.000000000 +0100
@@ -855,9 +855,7 @@ void __init setup_arch(char **cmdline_p)
 #endif
 #endif
 
-#if defined(CONFIG_KGDB)
 	early_trap_init();
-#endif
 }


[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: linux-next: fix vector initialisation when !CONFIG_KGDB

[Jason Wessel]

Ben Dooks wrote:
> It seems in the changes for adding CONFIG_KGDB that the ARM
> trap initialisation has been turned off if the configuration
> is not enabled. If CONFIG_KGDB is not set, the system locks
> up as soon as the kernel turns the IRQs on for the first
> time in init/main.c.
>
> The fix is to remove the dependency on calling early_trap_init()
> on CONFIG_KGDB, so that the vectors are initialised before
> the init sequence enables the IRQs.
>
> Signed-off-by: Ben Dooks <ben-linux@fluff.org>
>
> --- linux-2.6.26-rc6-q1.orig/arch/arm/kernel/setup.c	2008-06-18 15:17:31.000000000 +0100
> +++ linux-2.6.26-rc6-q1/arch/arm/kernel/setup.c	2008-06-18 15:17:45.000000000 +0100
> @@ -855,9 +855,7 @@ void __init setup_arch(char **cmdline_p)
>  #endif
>  #endif
>  
> -#if defined(CONFIG_KGDB)
>  	early_trap_init();
> -#endif
>  }
>
>   

That is absolutely correct.  Thanks for catching this.

I folded this into the general ARM kgdb patch as there is no sense in
generating a patch bisecting failure.  The change is in the kgdb-next
branch which feeds into the linux-next, which means it should show up in
the next day or so.

Thanks,
Jason.