arm64: fix early_io_map for 64K pages

Commit Message

Rob Herring Feb. 5, 2014, 2:50 p.m. UTC

From: Rob Herring <[e-mail]>

earlyprintk is broken with 64KB pages. The problem is pgprot_default
is not yet initialized when early_io_map is called, so the pte does not
get marked as valid. Set the necessary page and access permission bits.

Cc: Mark Salter <[e-mail]>
Cc: Catalin Marinas <[e-mail]>
Cc: Will Deacon <[e-mail]>
Signed-off-by: Rob Herring <[e-mail]>
---

I'm not sure this is really the best fix. Perhaps pgprot_default can be
statically initialized to something useful instead? 

Rob

 arch/arm64/mm/mmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Will Deacon Feb. 5, 2014, 2:52 p.m. UTC | #1

Hi Rob,

On Wed, Feb 05, 2014 at 02:50:29PM +0000, Rob Herring wrote:
> From: Rob Herring <[e-mail]>
> 
> earlyprintk is broken with 64KB pages. The problem is pgprot_default
> is not yet initialized when early_io_map is called, so the pte does not
> get marked as valid. Set the necessary page and access permission bits.
> 
> Cc: Mark Salter <[e-mail]>
> Cc: Catalin Marinas <[e-mail]>
> Cc: Will Deacon <[e-mail]>
> Signed-off-by: Rob Herring <[e-mail]>
> ---

I thought we fixed this in 4ce00dfcf19c by reverting the use of
pgprot_default?

Will

Catalin Marinas Feb. 5, 2014, 2:55 p.m. UTC | #2

On Wed, Feb 05, 2014 at 02:52:51PM +0000, Will Deacon wrote:
> On Wed, Feb 05, 2014 at 02:50:29PM +0000, Rob Herring wrote:
> > From: Rob Herring <[e-mail]>
> > 
> > earlyprintk is broken with 64KB pages. The problem is pgprot_default
> > is not yet initialized when early_io_map is called, so the pte does not
> > get marked as valid. Set the necessary page and access permission bits.
> > 
> > Cc: Mark Salter <[e-mail]>
> > Cc: Catalin Marinas <[e-mail]>
> > Cc: Will Deacon <[e-mail]>
> > Signed-off-by: Rob Herring <[e-mail]>
> > ---
> 
> I thought we fixed this in 4ce00dfcf19c by reverting the use of
> pgprot_default?

Indeed. We still need to fix the ioremap_cache() case where the
shareability isn't set but I haven't yet decided what we do for
SMP-on-UP in arm64 (if we always assume SMP, I would just get rid of
pgprot_default entirely).

Mark Salter Feb. 5, 2014, 3:04 p.m. UTC | #3

On Wed, 2014-02-05 at 08:50 -0600, Rob Herring wrote:
> From: Rob Herring <[e-mail]>
> 
> earlyprintk is broken with 64KB pages. The problem is pgprot_default
> is not yet initialized when early_io_map is called, so the pte does not
> get marked as valid. Set the necessary page and access permission bits.
> 
> Cc: Mark Salter <[e-mail]>
> Cc: Catalin Marinas <[e-mail]>
> Cc: Will Deacon <[e-mail]>
> Signed-off-by: Rob Herring <[e-mail]>
> ---
> 

FWIW, I had a patch in my early_ioremap series which initialized
pgprot_default earlier in boot:

http://www.spinics.net/lists/arm-kernel/msg299656.html

(which reminds me to get v4 of that series out this week)

Rob Herring Feb. 5, 2014, 3:05 p.m. UTC | #4

On Wed, Feb 5, 2014 at 8:52 AM, Will Deacon <[e-mail]> wrote:
> Hi Rob,
>
> On Wed, Feb 05, 2014 at 02:50:29PM +0000, Rob Herring wrote:
>> From: Rob Herring <[e-mail]>
>>
>> earlyprintk is broken with 64KB pages. The problem is pgprot_default
>> is not yet initialized when early_io_map is called, so the pte does not
>> get marked as valid. Set the necessary page and access permission bits.
>>
>> Cc: Mark Salter <[e-mail]>
>> Cc: Catalin Marinas <[e-mail]>
>> Cc: Will Deacon <[e-mail]>
>> Signed-off-by: Rob Herring <[e-mail]>
>> ---
>
> I thought we fixed this in 4ce00dfcf19c by reverting the use of
> pgprot_default?

Ah crap. I only checked the history on mmu.c and missed that fix. And
I was so happy I found a kernel issue with qemu.

Rob

Patch

diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index f557ebb..c41daa6 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -284,7 +284,7 @@  void __iomem * __init early_io_map(phys_addr_t phys, unsigned long virt)
 		if (pmd_none(*pmd))
 			return NULL;
 		pte = pte_offset_kernel(pmd, virt);
-		set_pte(pte, __pte((phys & mask) | PROT_DEVICE_nGnRE));
+		set_pte(pte, __pte((phys & mask) | PTE_TYPE_PAGE | PTE_AF | PROT_DEVICE_nGnRE));
 	} else {
 		set_pmd(pmd, __pmd((phys & mask) | PROT_SECT_DEVICE_nGnRE));
 	}