This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: PATCH: Add access restriction check for application registers(was Re: Is this a gas bug?)
- From: Jim Wilson <wilson at specifixinc dot com>
- To: "H. J. Lu" <hjl at lucon dot org>
- Cc: Jie Zhang <jiez at citiz dot net>, Andreas Schwab <schwab at suse dot de>,george dot gan at intel dot com,"binutils at sources dot redhat dot com" <binutils at sources dot redhat dot com>,"bug-binutils at gnu dot org" <bug-binutils at gnu dot org>
- Date: 27 May 2004 13:15:08 -0700
- Subject: Re: PATCH: Add access restriction check for application registers(was Re: Is this a gas bug?)
- References: <1085333524.2233.56.camel@mars> <jesmdrow2p.fsf@sykes.suse.de><1085411758.869.20.camel@mars> <20040526182849.GA1631@lucon.org>
On Wed, 2004-05-26 at 11:28, H. J. Lu wrote:
> * config/tc-ia64.c (ar_is_only_in_integer_unit): New.
> (ar_is_only_in_memory_unit): New.
> (generate_unwind_image): Silence gcc on 32bit host.
> (md_assemble): Check AR access.
Thanks. This looks good.
The differences between ar_is_in_integer_unit and
ar_is_only_in_integer_unit do not look right though. It appears to me
that ar_is_in_integer_unit is broken, as it should accept a strict
superset of the regs accepted by ar_is_only_in_integer_unit, but it does
not. The difference is regs 67 to 111. I think we need to modify
ar_is_in_integer_unit to accept them.
Otherwise, the assembler will take
mov ar100 = r10
and assemble it into a mov.m instruction, which is wrong according to my
reading of the docs.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com