On Tue, 2006-07-25 at 19:26 +0100, Christian Limpach wrote:
> On 25/7/06 07:06, "Zachary Amsden" <zach at vmware.com> wrote:
>
> > 1) Init code fills in pv_ops struct
> > 2) Linux does a bunch of stuff - now about to set up pagetables on BSP
> > 3) set_fixaddr_top(pv_ops->get_linear_top()) (once only, in
> > arch/i386/mm/init.c)
>
> I think this is the best proposal so far, since it removes how having a way
> to control where the VA space ends and having fixaddrs (at all, at the end
> of VA space) would be tied together otherwise. And it leaves the code more
> readable with the set_fixaddr_top call on the main code path.
There are several issues to answer here. First, that "set_fixaddr_top"
above would need to be in an #ifdef or otherwise wrapped for the
non-paravirt case.
Secondly, *if* the only thing that will actually alter the fixaddr_top
is paravirt, it makes sense to keep it a constant except in the paravirt
case, ie:
#ifdef CONFIG_PARAVIRT
#define __FIXADDR_TOP (paravirt_ops.linear_top)
#else
#define __FIXADDR_TOP 0xfffff000
#endif
This is simpler and clearer than indirecting through another function
(set_fixaddr_top) and another variable (__FIXADDR_TOP).
Along similar lines, I dislike a function ->get_linear_top() because
obscures the fact that it cannot change. Having a "constant" in an ops
structure is a little odd, but it's clear (a-la kernel_rpl).
My 2c,
Rusty.
--
Help! Save Australia from the worst of the DMCA: http://linux.org.au/law
