rtems-CVS-050303 i.MX/MX1/csb336 BUG and BUGFIX

[Ralf Corsepius]

On Fri, 2005-03-04 at 01:58 +0100, Pavel Pisa wrote:
> Hello Joel and all others,
> 
> there has been introduced bug into bsp_vector_table definition.
Well, I agree, you are right, there is a new bug, but ... I think this
bug only exposes another, more severe bug in all arm BSPs :(

> It has been changed to "rtems_irq_hdl bsp_vector_table;".
> This means, that ldscript declares address of variable
> of the function pointer type.
> This is than silently casted from (rtems_irq_hdl) to (rtems_irq_hdl*)
> aka (long*) in
> 
>   rtems/c/src/lib/libbsp/arm/shared/irq/irq_init.c:rtems_irq_mngt_init()
>       vectorTable = (long *) VECTOR_TABLE;
This code is plain wrong.

> and uninitialized memory contents is used as base address for the vector table
> 
>       /* Initialize the vector table contents with default handler */
>       for (i=0; i<BSP_MAX_INT; i++)
>         *(vectorTable + i) = (long)(default_int_handler);
Worse, this code probably is non functionial.

I think,
volatile rtems_irq_hdl* vectorTable
is required, because as far as I understand the this code is supposed to
initialize hardware addresses.

> Next oneliner patch solves described problem
> 
> ---------------------------------------------------------------------------------------------
> diff --minimal -r -u -P -p rtems/c/src/lib/libcpu/arm/mc9328mxl/irq/irq.h.orig rtems/c/src/lib/libcpu/arm/mc9328mxl/irq/irq.h
> --- rtems/c/src/lib/libcpu/arm/mc9328mxl/irq/irq.h.orig 2005-03-04 00:57:56.000000000 +0100
> +++ rtems/c/src/lib/libcpu/arm/mc9328mxl/irq/irq.h 2005-03-04 01:00:05.000000000 +0100
> @@ -114,7 +114,7 @@ typedef void (*rtems_irq_enable)    (con
>  typedef void (*rtems_irq_disable)   (const struct __rtems_irq_connect_data__*);
>  typedef int  (*rtems_irq_is_enabled)(const struct __rtems_irq_connect_data__*);
>  
> -extern rtems_irq_hdl bsp_vector_table;
> +extern rtems_irq_hdl bsp_vector_table[BSP_MAX_INT];
>  #define VECTOR_TABLE bsp_vector_table
>                  
>  typedef struct __rtems_irq_connect_data__ {

I'd assume bsp_vector_table should be volatile, too.
> ---------------------------------------------------------------------------------------------
> 
> But I have big problem to analyze and find this hidden bug.

Me thinks, this code needs to be redesigned - :(

Ralf