Hi,

On Friday, 30 of September 2005 00:29, Siddha, Suresh B wrote:
> On Fri, Sep 30, 2005 at 12:01:08AM +0200, Rafael J. Wysocki wrote:
> > On Thursday, 29 of September 2005 02:00, Siddha, Suresh B wrote:
> > > 
> > > My patch as such shouldn't change the behavior of the existing swsup
> > > code. I am making only boot_level4_pgt as initdata. But not the 
> > > init_level4_pgt.
> > 
> > Suresh, unfortunately the kernel does not boot with your patch, because
> 
> Did you try just only my patch on top of 2.6.14-rc2? You can get that
> patch from http://www.x86-64.org/lists/discuss/msg07313.html

The patch that I tested is attached.  I think it's the same one.  I've just applied
it on top of 2.6.14-rc2-git7, and it doesn't boot.

> > it clears init_level4_pgt.
> 
> what is wrong with zapping low mappings? Looks like someother change
> you are trying assumes that low mappings are always present in init_level4_pgt,
> which is wrong..

Nope.

The problem (as I see it) is this:
In x86_64_start_kernel() you copy boot_level4_pgt[] into init_level4_pgt[],
and you make the latter your current PGD by loading cr3 with its address.
Fine.  With this PGD you call start_kernel() which calls setup_arch(), which
calls zap_low_mappings(0) that fills init_level4_pgt[] (which at this moment
is still your current PGD) with zeros ...

> > +++ linux-2.6.14-rc2-git7/arch/x86_64/mm/init.c	2005-09-29 22:13:55.000000000 +0200
> > @@ -313,7 +313,7 @@
> >  void __cpuinit zap_low_mappings(int cpu)
> >  {
> >  	if (cpu == 0) {
> > -		pgd_t *pgd = pgd_offset_k(0UL);
> > +		pgd_t *pgd = boot_level4_pgt;
> >  		pgd_clear(pgd);
> 
> Don't do this. Its wrong.

Perhaps it is.
 
> Please send me your patch that you are having problems with. I can take 
> a look at it.

I'll post it in a while to the list.  I'm not touching the init code anyway and
the box hangs before it has a chance to send anything to the serial console.

Greetings,
Rafael