Re: mmotm 2008-10-02-16-17 uploaded

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 8 Oct 2008 08:47:53 +0900
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

> On Tue, 07 Oct 2008 09:18:03 -0700
> Yinghai Lu <yinghai@kernel.org> wrote:
> 
> > KAMEZAWA Hiroyuki wrote:
> > > I'm sorry if alread fixed.
> > > 
> > > Folloing function in git-x86.patch touches invalid address.
> > > ==
> > > +void __init early_cpu_init(void)
> > > +{
> > > +       struct cpu_dev **cdev;
> > > +       int count = 0;
> > > +
> > > +       printk("KERNEL supported cpus:\n");
> > > +       for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
> > > +               struct cpu_dev *cpudev = *cdev;
> > > +               unsigned int j;
> > > +
> > > +               if (count >= X86_VENDOR_NUM)
> > > +                       break;
> > > +               cpu_devs[count] = cpudev;
> > > +               count++;
> > > +
> > > +               for (j = 0; j < 2; j++) {
> > > +                       if (!cpudev->c_ident[j])
> > > +                               continue;
> > > +                       printk("  %s %s\n", cpudev->c_vendor,
> > > +                               cpudev->c_ident[j]);
> > > +               }
> > > +       }
> > > +
> > > +       early_identify_cpu(&boot_cpu_data);
> > >  }
> > > ==
> > > printk("  %s %s\n", cpudev->c_vendor, cpudev->c_ident[j]);
> > > touches invalid address. (following is printk() result by
> > > replacing %s with %p.
> > > ==
> > >   ffffffff8066e38a ffffffff8066e383
> > >   ffffffff8066e3ab ffffffff8066e3a2
> > >   ffffffff8066e3af ffffffff8066e3b7
> > >   807064c0c7c74855 08ec834853e58948
> > >   807064c0c7c74855 74c085fffffe9fe8
> > > ==
> > > and the kernel never boots on my box.
> > 
> > could be merging problem again.
> > 
> > please check in arch/x86/kernel/vmliux_64.lds.S
> > 
> > it should be like 
> > 
> >   __con_initcall_end = .;
> >   __x86_cpu_dev_start = .;
> >   .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
> >         *(.x86_cpu_dev.init)
> >   }
> >   __x86_cpu_dev_end = .;
> > 
> >   DYN_ARRAY_INIT(8)
> > 
> >   SECURITY_INIT

The above is what's presently in Ingo's "tip" tree.

> Oh, yes. like this.
> ==
>  __con_initcall_end = .;
>   __x86_cpu_dev_start = .;
>   .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
>         *(.x86_cpu_dev.init)
>   }
>   SECURITY_INIT
>   __x86_cpu_dev_end = .;
> ==
> 
> I'll try next version when it comes.

If that fixes it then Ingo's tree will need fixing too, I suppose.