Re: [Qemu-devel] [PATCH 2/2] Revived GUEST_BASE support for usermode emulation targets.

[malc]

On Fri, 17 Apr 2009, Mika Westerberg wrote:

> 
> On Apr 10, 2009, at 2:22 AM, malc wrote:
> 
> > On Thu, 9 Apr 2009, address@hidden wrote:
> > 
> > > From: Riku Voipio <address@hidden>
> > > 
> > > From: Mika Westerberg <address@hidden>
> > > 
> > > - Now GUEST_BASE is dynamic and can be set from command line.
> > > - Qemu checks /proc/sys/vm/mmap_min_addr and sets GUEST_BASE
> > > if needed.
> > > - Code generation supports GUEST_BASE for i386 and x86_64 hosts.
> > > 
> > > Signed-off-by: Riku Voipio <address@hidden>
> > [..snip..]
> > 
> > > diff --git a/tcg/i386/tcg-target.c b/tcg/i386/tcg-target.c
> > > index e0fd434..7cb51b5 100644
> > > --- a/tcg/i386/tcg-target.c
> > > +++ b/tcg/i386/tcg-target.c
> > > @@ -560,6 +560,12 @@ static void tcg_out_qemu_ld(TCGContext *s, const
> > > TCGArg *args,
> > >    /* add x(r1), r0 */
> > >    tcg_out_modrm_offset(s, 0x03, r0, r1, offsetof(CPUTLBEntry, addend) -
> > >                         offsetof(CPUTLBEntry, addr_read));
> > > +#elif defined(CONFIG_USE_GUEST_BASE)
> > > +    /*
> > > +     * Add guest_base to all loads.
> > > +     */
> > > +    tcg_out_mov(s, r0, addr_reg);    /* movl addr_reg, r0 */
> > > +    tcg_out_addi(s, r0, GUEST_BASE); /* addl $GUEST_BASE, r0 */
> > 
> > I think 'lea' is warranted here, and likewise in _st and x86_64 case.
> 
> Can you be more specific about that? Do you mean that we should change MOV
> into
> LEA?
> 
> I think that in this case, as we are just moving address (pointer) from one
> register to another, it should be ok.  For example gcc generates from this:
> 
> static char *
> f(char *p)
> {
>        return (p + 65536);
> }
> 
> following code:
> 
> 00000000 <f>:
>   0:   55                      push   %ebp
>   1:   89 e5                   mov    %esp,%ebp
>   3:   8b 45 08                mov    0x8(%ebp),%eax
>   6:   05 00 00 01 00          add    $0x10000,%eax
>   b:   c9                      leave
>   c:   c3                      ret
> 
> Or am I missing something?

Well, here parameter (p) is passed on the stack so it should first be
moved to register, so gcc emits pair of mov, add, in TCG case however

mov r1, r0
add r1, const

can (and should) be replaced with a simple:

lea r1, [r0 + const]

-- 
mailto:address@hidden