Re: RFC: PATCH: PR x86_64/584: Detect call on protected symbol

["H. J. Lu" <hjl at lucon dot org>]

On Thu, Jan 20, 2005 at 12:25:19PM +1030, Alan Modra wrote:
> On Wed, Jan 19, 2005 at 04:40:32PM -0800, H. J. Lu wrote:
> > On Thu, Jan 20, 2005 at 12:05:44AM +0100, Jakub Jelinek wrote:
> > > On Wed, Jan 19, 2005 at 02:48:34PM -0800, H. J. Lu wrote:
> > > > X86_64 uses R_X86_64_PC32 for both branch and store/load. Linker
> > > > can't tell if a protected symbol reference is local or global just
> > > > by relocation. This patch disassembles the code to check for call.
> > > 
> > > Eh, why you need the disassembler there?
> > > Isn't it enough just to check if R_X86_64_PC32's r_offset > 0 and
> > > contents[r_offset - 1] == 0xe8?
> > 
> > Can we be sure that contents[r_offset - 1] == 0xe8 will be call?
> 
> Not if the PC32 reloc happens to be in data.  Also, I suspect some of

We can check if the input section is executable. The worst case is that
you may get a runtime error instead of a linktime one.

> the possible load/store insns might have a sib byte of 0xE8.

I think 0xe8 SIB is [RAX + RIP * 8], which can't have R_X86_64_PC32.


H.J.