On Thu, Mar 25, 2010 at 11:53 AM, Stefan Reinauer <stepan@coresystems.de> wrote:

On 3/25/10 5:45 PM, Myles Watson wrote:
The reason I object to the void* method was that it just masked the problem so that gcc couldn't spot it. Casting to void* and back to a struct seems equivalent to just having it declared two different ways.

I don't think it's masking the problem. It should be sufficient to let gcc know that it can't assume anything from the pointer since it comes from a void *. Also, it's not casting it back to a struct, because it never was a struct. So, while it might not be beautiful, I think it's at least not wrong.

The interesting thing is that gcc won't allow the cast if you don't have a variable for it. That's why I thought it was masking the problem.

In other words, when I tried:
((acpi_header_t*)((void*)&AmlCode)) ->length

It still has the type-punned-pointer warning.

> Which makes me think if there are not other ways to determine the size
> of an array. Maybe sizeof(AmlCode) ?
I tried that. It doesn't work because it's an incomplete type.
It would not if we included the file

----- 8< ------ dsdt_wrapper.c -------- 8< ----------

#include "dsdt.c"

int AmlCodeSize = sizeof(AmlCode);

The boards with multiple ssdts would have to be refactored.

A (maybe too obvious) variant would be to create a binary dsdt instead of a C file and pack that into CBFS. It would reduce coreboot size and allow to copy/decompress it right to cbmem

This would be fine with me. It's quite a bit more overhead than the double copy, though.

Thanks,
Myles