Re: the creation of boot_cpu_init() is wrong and accessing uninitialised data

[Andrew Morton <akpm@osdl.org>]

On Mon, 26 Jun 2006 22:36:32 -0500
James Bottomley <James.Bottomley@SteelEye.com> wrote:

> On Mon, 2006-06-26 at 20:04 -0700, Andrew Morton wrote:
> > I think arch code should do it before calling start_kernel(), really.
> > It's
> > just such a basic part of the kernel framework.
> 
> Hmm ... well, getting at current_thread_info()->cpu is possible, but
> nasty to audit, I would have thought (given that we're in assembler
> before start_kernel is called).

Well.  It's the assembly code which chose to call start_kernel().  It could
call something else.

> > A less wholesome but perhaps simpler solution would be to call the new
> > setup_smp_processor_id() on entry to start_kernel().
> 
> I was wondering about simply replacing boot_cpu_init() with
> smp_prepare_boot_cpu().  By and large they do the same thing on most
> archs, and mostly they don't seem to depend on setup_arch() having been
> called.

That won't fix the other bugs - we're presently calling printk() prior to
setup_arch(), and printk uses smp_procesor_id().

> However, introducing setup_smp_processor_id() will also work ... I'll
> see if I can do it in an easy way.

It's a bit odd - I think non-zero BSPs happen a bit more often than
only-on-voyager.