Re: [PATCH/RFT 0/13] x86: unify vmlinux.lds

[Ingo Molnar <mingo@elte.hu>]

* Sam Ravnborg <sam@ravnborg.org> wrote:

> On Wed, Apr 29, 2009 at 12:00:13PM +0200, Ingo Molnar wrote:
> > 
> > Ok, first problem hit in testing:
> > 
> > Freeing unused kernel memory: 1800k freed
> > BUG: Bad page state in process swapper  pfn:01c00
> > page:f666b000 flags:00840000 count:0 mapcount:0 mapping:(null) index:0
> > Pid: 1, comm: swapper Not tainted 2.6.30-rc3-tip-01696-g9b182a3-dirty #37518
> > Call Trace:
> >  [<c08730d0>] ? printk+0x15/0x1d
> >  [<c01aa0bb>] bad_page+0xdb/0x150
> >  [<c01abba7>] free_hot_cold_page+0x1b7/0x2f0
> >  [<c01abd4f>] free_hot_page+0xf/0x20
> >  [<c01abd97>] __free_pages+0x37/0x40
> >  [<c01abddf>] free_pages+0x3f/0x50
> >  [<c011b8c0>] free_init_pages+0xb0/0xd0
> >  [<c011b8fc>] free_initmem+0x1c/0x20
> >  [<c01011d2>] init_post+0x12/0x170
> >  [<c0c142f3>] kernel_init+0x14b/0x161
> >  [<c0c141a8>] ? kernel_init+0x0/0x161
> >  [<c0103be7>] kernel_thread_helper+0x7/0x10
> > Disabling lock debugging due to kernel taint
> > Write protecting the kernel text: 7660k
> > Testing CPA: Reverting c0100000-c087b000
> > 
> > more details attached. (sendig the System.map separately, it's too 
> > large) Commit used was v2.6.30-rc3-1696-g9b182a3.
> 
> OK - I think I know what is wrong.

that's good - because i tried to bisect it and it's not bisectable. 
(went away when i moved to a pure x86/kbuild tree - i.e. the bug is 
possibly dependent on specific symbol placement, size and alignment)

> You have:
> 
> CONFIG_RELOCATABLE=y
> 
> And I moved a few symbols out of an output section.
> But they were there for a purpose as the comment in the top of
> the file says.
> 
> I am on my way out of the door.
> Patch follows tonight.

hey this description is enough for me to fix it ;-) I suspect the 
main one that hit us here is:

        __init_begin = .; /* paired with __init_end */

	...

        __init_end = .;

Havent found any other absolute symbols.

	Ingo