question about the usage of DECL_INCOMING_RTL in a stack-smashing protector

[Hiroaki Etoh <ETOH at jp dot ibm dot com>]

I am developing a stack protection tool (
http://www.research.ibm.com/trl/projects/security/ssp). The tool is a kind 
of program translator in a RTL code level.  In order to prevent the 
corruption of pointers from stack smashing attacks, the tool changes the 
location of pointers in function arguments to a pseudo register, which 
will be assigned to the stack when the register is spilled out from the 
assignment of hard registers. 

To maintain the location of arguments for gdb, I am setting the pseudo 
register to DECL_INCOMING_RTL after rtl generation.   I examined that 
register allocator allocates a hard register or a stack region for that 
pseudo and it also changes DECL_INCOMING_RTL to the right location.

I'd like to know issues about setting a pseudo register to 
DECL_INCOMING_RTL.

Hiroaki Etoh,  Tokyo Research Laboratory, IBM Japan
----- Forwarded by Hiroaki Etoh/Japan/IBM on 2004/01/19 10:04 -----

Josef Zlomek <zlomekj@suse.cz> wrote on 2004/01/02 22:24:43:

> I noticed you are setting a pseudoregister to DECL_INCOMING_RTL.
> IMHO (according to comment in tree.h for DECL_INCOMING_RTL and my 
experience
> with it) it may contain only memory or hard registers. When 
registerallocator
> allocates a hard register for that pseudo you should change 
DECL_INCOMING_RTL
> too.  Or maybe you should set DECL_RTL instead of DECL_INCOMING_RTL.
> Please ask on gcc@gcc.gnu.org mailing list.
> 
> Josef