Hi Nicolas,

On Tue, Jan 18, 2011 at 2:26 PM, Nicolas Pitre [off-list ref] wrote:

On Tue, 18 Jan 2011, Grant Likely wrote:

quoted

The dtb is passed to the kernel via register r2, which is the same
method that is used to pass an atags pointer.  This patch modifies
__vet_atags to not clear r2 when it encounters a dtb image.

Signed-off-by: Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>
---
 arch/arm/kernel/head-common.S |   19 +++++++++++++------
 arch/arm/kernel/head.S        |    8 ++++----
 2 files changed, 17 insertions(+), 10 deletions(-)

diff --git a/arch/arm/kernel/head-common.S b/arch/arm/kernel/head-common.S
index 8f57515..d9a9105 100644
--- a/arch/arm/kernel/head-common.S
+++ b/arch/arm/kernel/head-common.S

@@ -14,6 +14,7 @@

 #define ATAG_CORE 0x54410001
 #define ATAG_CORE_SIZE ((2*4 + 3*4) >> 2)
 #define ATAG_CORE_SIZE_EMPTY ((2*4) >> 2)
+#define OF_DT_MAGIC 0xedfe0dd0 /* 0xd00dfeed in big-endian */

ARM can be big endian too.  Would be nice to make this endian
independent, or at least list this limitation in the commit log.

How about:
#ifdef CONFIG_CPU_BIG_ENDIAN
#define OF_DT_MAGIC 0xd00dfeed
#else
#define OF_DT_MAGIC 0xedfe0dd0 /* 0xd00dfeed in big-endian */
#endif

Or is there a better way for loading and comparing big endian values in ARM?

quoted

 /*
  * Exception handling.  Something went wrong and we can't proceed.  We

@@ -105,22 +106,28 @@ __lookup_machine_type_data:

 /* Determine validity of the r2 atags pointer.  The heuristic requires
  * that the pointer be aligned, in the first 16k of physical RAM and
- * that the ATAG_CORE marker is first and present.  Future revisions
+ * that the ATAG_CORE marker is first and present.  If CONFIG_OF_FLATTREE
+ * is selected, then it will also accept a dtb pointer.  Future revisions
  * of this function may be more lenient with the physical address and
  * may also be able to move the ATAGS block if necessary.
  *
  * r8  = machinfo
  *
  * Returns:
- *  r2 either valid atags pointer, or zero
+ *  r2 either valid atags pointer, valid dtb pointer, or zero
  *  r5, r6 corrupted
  */
 __vet_atags:
      tst     r2, #0x3                        @ aligned?
      bne     1f

-     ldr     r5, [r2, #0]                    @ is first tag ATAG_CORE?
-     cmp     r5, #ATAG_CORE_SIZE
+#ifdef CONFIG_OF_FLATTREE
+     ldr     r5, [r2, #0]                    @ is it a DTB?
+     ldr     r6, =OF_DT_MAGIC
+     cmp     r5, r6
+     beq     2f
+#endif
+     cmp     r5, #ATAG_CORE_SIZE             @ is first tag ATAG_CORE?

The "ldr r5 ..." is now done only within #ifdef CONFIG_OF_FLATTREE.  So
if CONFIG_OF_FLATTREE is undefined then r5 will contain garbage.

Oops, that was a silly oversight.  Fixed now.

g.

Nicolas

-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.