host_pointer_to_address() === (CORE_ADDR) (void*) (val)

[Donn Terry]

I was looking at Alpha/NT at one point in my life, and it
also is sign-extended pointers.  Whether there are any other
instances of 32 bit Alpha implementations, I don't know.

Based on what I remember of that old code, this seems an
emminently reasonable thing to do.

Donn

> -----Original Message-----
> From: Andrew Cagney [ mailto:ac131313@cygnus.com ]
> Sent: Friday, May 26, 2000 4:02 AM
> To: GDB Patches
> Subject: host_pointer_to_address() === (CORE_ADDR) (void*) (val)
> 
> 
> Hello,
> 
> If you look through code like procfs.c, you will find casts like:
> 
>       /* Stop looping if the callback returns non-zero.  */
>       if ((funcstat = (*func) (fd, (CORE_ADDR) map->pr_vaddr)) != 0)
> 
> The problem is with ``(CORE_ADDR) map->pr_vaddr''.  
> ``map->pr_vaddr'' is
> a ``void *'' (32bits) and ``CORE_ADDR'' is a ``long long'' (64bits). 
> Some (one? mips/n32) targets assume that addresses sign-extended while
> others assume zero extension.  The consequence is that the 
> above code is
> potentially dangerous.  GCC complains with a warning while some native
> compilers refuse to accept it at all :-/
> 
> To address this, I'd like to propose two wrappers to JimB's
> POINTER_TO_ADDRESS and ADDRESS_TO_POINTER:
> 
> 	CORE_ADDR host_pointer_to_address (void *ptr);
> 		... check operation is sane ...
> 		return POINTER_TO_ADDRESS (builtin_type_ptr, ptr);
> 	void *address_to_host_pointer (CORE_ADDR addr);
> 		void *ptr;
> 		... check operation is sane ...
> 		ADDRESS_TO_POINTER (builtin_type_ptr, &ptr, addr)
> 		return ptr;
> 
> which would be used when ever a host void* <-> CORE_ADDR 
> conversion was
> required.  The above code would then be rewritten as:
> 
> 	funcstat = func (fd, host_pointer_to_address (map->pr_vaddr));
> 	if (funcstat != 0)
> 	  ....
> 
> this would ensure that the code passed around a correct ``CORE_ADDR''
> (and not a truncated value that just happens to work).  This becomes
> very important when you start debugging n32 abi's (32 bit pointers)
> where the target has 64 bit registers - the full sign extended value
> needs to be read/written.
> 
> Thoughts?  Making sense?  I'll post a full patch for irix[56]/n32 once
> I've seen some test results.
> 
> 	Andrew
>