[coreboot] romcc segfault

[Stefan Reinauer]

On 4/12/11 10:05 PM, Russell Whitaker wrote:
>
>
> On Tue, 12 Apr 2011, Stefan Reinauer wrote:
>
>> Hi,
>>
>> just a heads up, I got romcc to segfault with the following sample 
>> program:
>>
>>
>> --------------------- 8< snip 8< --------------------------------------
>> typedef unsigned int u32;
>> #define DEFAULT_RCBA 0xfed1c000
>> #define GCS 0x3410
>> #define RCBA32(x) *((volatile u32 *)(DEFAULT_RCBA + x))
>>
>> void test(void)
>> {
>>        RCBA32(GCS) &= (~0x04);
>> }
>> --------------------- >8 snip >8 --------------------------------------
>>
>> I tried to simplify this further and this construct is still crashing 
>> romcc:
>>
>> --------------------- 8< snip 8< --------------------------------------
>>        u32 *gcs = (u32 *)(DEFAULT_RCBA + GCS);
>>        *gcs &= (~0x04);
>> --------------------- >8 snip >8 --------------------------------------
>>
>> while this one is working
>>
>> --------------------- 8< snip 8< --------------------------------------
>>        u32 *gcs = (u32 *)(DEFAULT_RCBA + GCS);
>>        u32 temp = *gcs;
>>        temp &= (~0x04);
>>        *gcs = temp;
>> --------------------- >8 snip >8 --------------------------------------
>>
>> Is that a construct that we just can't/don't want to support in romcc?
>> Or is there a chance we can fix that?
>>
>> Stefan
>
> Just for comparison used gcc -S to compile the 3 snippets.
> The first compiled without error.
>
[..]

Russel,

Not sure what you tried, but they all compile without error. You have to 
wrap example 2 and 3 with

typedef unsigned int u32;
#define DEFAULT_RCBA 0xfed1c000
#define GCS 0x3410
void test(void)
{

and

}

before actually running them through gcc.