Re: [XFRM] Make XFRM core subsystem af-independent

[YOSHIFUJI Hideaki / 吉藤英明 <yoshfuji@xxxxxxxxxxxxxx>]

In article <20040918001216.GB24235@xxxxxxxxxxxxxxxxxxx> (at Sat, 18 Sep 2004 
10:12:16 +1000), Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> says:

> On Fri, Sep 17, 2004 at 05:06:07PM -0700, David S. Miller wrote:
> > 
> > Compiler is smarter than you think :-)
> 
> Sparc bigots :)
> 
> Unfortunately it can't do that on architectures where the arguments
> are pushed onto the stack in reverse order, i.e., i386.

I've tested on i386.

.globl ip_route_output_key
        .type    ip_route_output_key,@function
ip_route_output_key:
        movl 4(%esp),%edx
        movl 8(%esp),%eax
        pushl $0
        pushl $0
        pushl %eax
        pushl %edx
        call ip_route_output_flow
        addl $16,%esp
        ret

This is not good.

We, however, now have (up to 3) register parameters on i386 (CONFIG_REGPARM), 
and assembly code is:

.globl ip_route_output_key
        .type   ip_route_output_key,@function
ip_route_output_key:
        pushl   $0
        xorl    %ecx, %ecx
        call    ip_route_output_flow
        popl    %ecx
        ret

This is (not good as sparc, but) not bad
(well, much better than before at least).

-- 
Hideaki YOSHIFUJI @ USAGI Project <yoshfuji@xxxxxxxxxxxxxx>
GPG FP: 9022 65EB 1ECF 3AD1 0BDF  80D8 4807 F894 E062 0EEA