[PATCH 2/2] firmware/memmap: Move kobject initialisation before kobject_add()

[Andrew Morton]

On Wed, 30 Jul 2008 22:14:24 +0200
Bernhard Walle <bwalle at suse.de> wrote:

> * Bernhard Walle <bwalle at suse.de> [2008-07-30 22:12]:
> >
> > Because kobject_init() call could be done from firmware_map_add_entry()
> > which is called before kmalloc() can be used (we use the early bootmem allocator
> > here), move that call to memmap_init() which is a late_initcall.
> 
> You can leave out that patch now since kobject_init() was adapted with
> the other patch ... But I think it would still make sense to include
> that one.

What one?  Your thises and thats are confusing.

If you think that "firmware/memmap: Move kobject initialisation before
kobject_add()" should still be applied then I'd disagree.  Adding a
GFP_KERNEL allocation into kobject_init() was a fairly significant
backward step.  It's _good_ that kobject_init() can be called this
early.  Let us strive to retain that robustness.