Re: [PATCH] Fix non-LPAE boot regression.

[Vasily Khoruzhick <anarsoul@gmail.com>]

On Monday 15 August 2011 15:09:14 Catalin Marinas wrote:
> Hi Vasily,
> 
> On Sat, Aug 13, 2011 at 01:58:19PM +0100, Vasily Khoruzhick wrote:
> > It was introduced by  407f8b4cb07cbc5c1c7cc386f231224e2524ccea
> > ARM: LPAE: MMU setup for the 3-level page table format
> > 
> > Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
> > ---
> > 
> >  arch/arm/kernel/head.S |    4 ++--
> >  1 files changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/arm/kernel/head.S b/arch/arm/kernel/head.S
> > index 0bdafc4..5add5f5 100644
> > --- a/arch/arm/kernel/head.S
> > +++ b/arch/arm/kernel/head.S
> > 
> > @@ -206,7 +206,7 @@ __create_page_tables:
> >  1:	orr	r3, r7, r5, lsl #SECTION_SHIFT	@ flags + kernel base
> >  
> >  	str	r3, [r4, r5, lsl #PMD_ORDER]	@ identity mapping
> >  	cmp	r5, r6
> > 
> > -	addlo	r5, r5, #SECTION_SHIFT >> 20	@ next section
> > +	addlo	r5, r5, #1			@ next section
> > 
> >  	blo	1b
> 
> That's correct.
> 
> >  	/*
> > 
> > @@ -217,7 +217,7 @@ __create_page_tables:
> >  	mov	r3, r3, lsr #SECTION_SHIFT
> >  	orr	r3, r7, r3, lsl #SECTION_SHIFT
> >  	add	r0, r4,  #(KERNEL_START & 0xff000000) >> (SECTION_SHIFT -
> >  	PMD_ORDER)
> > 
> > -	str	r3, [r0, #(KERNEL_START & 0x00e00000) >> (SECTION_SHIFT -
> > PMD_ORDER)]! +	str	r3, [r0, #(KERNEL_START & 0x00f00000) >>
> > (SECTION_SHIFT - PMD_ORDER)]!
> 
> The reason for this was that the sections are 2MB with LPAE and a page
> table entry is 64-bit wide. We always shift that value by 18 but with
> LPAE we don't want to write in the middle of a page table entry if
> KERNEL_START is not 2MB aligned.
> 
> But if KERNEL_START is not 2MB aligned, I think we get the wrong
> physical address by 1MB (with the classic page table format).

Yep, for my case KERNEL_START is not 2MB aligned (TEXT_OFFSET is 0x00108000). 
(CONFIG_PM_H1940 is set)

> There are a few alternatives to fixing this:
> 
> 1. Different KERNEL_START masking for classic or LPAE page tables.
> 2. Always force 2MB section and the code above moving the phys addr into
>    r3 would need to take this into account.
> 
> I would go for 1 with some shifting like below:
> 
> +	str	r3, [r0, #((KERNEL_START & 0x00f00000) >> SECTION_SHIFT) <<
> PMD_ORDER]!
> 
> This should give us 0x00f00000 with classic page tables and 0x00e00000
> with LPAE.
> 
> Thanks.

Ok, I'll test this change when I get home.

Regards
Vasily