how to set endianness in ARM Cortex-A8

Hi,your saying that we can't run big endian code on rvds or ds-5

On ARM-v7 there is no such thing as big endian storage of code. All memory accesses made by the instruction-side memory system are always little endian. It's nothing to do with the tool chain; the ARM Architecture physically restricts instruction-side accesses to being little endian and this cannot be changed.

Data accesses, for example reading data off a ethernet chipset which is normally receive a bigendian byte stream, can be either big or little endian. The data-side memory system can be configured dynamically as such at runtime. It is entirely valid code code stored as little endian to make a data access to a big endian data resource.

Hi ranjithkumards,

could you tell me how you generate a big endian code for my future reference?

Are there such compiler as generate big endian code?

Best regards,

Yasuhiko Koumoto.

no.actually only data access can be made as big endian remaining almost same for bigendian

If you use --bigend for building it will build the library in big endian format for RVDS, but i need to know which command  i need to give in linux to build the code to big endian.

In linux how can i give the endian option acutally -mbig-endian is not supporting,so please tell the correct option in linux

I believe that big-endian support is an optional compile time build option for the GCC compiler itself, and not many pre-built versions for ARM support it as it is so rarely used, so you may need to re-build your own compiler with big-endian support.

Hello ranjithkumards,

do you ask how to set EE bit of SCTLR?

If SCTLR.EE=1, the data access is performed by big-endian although instruction access is little-endian..

MRC p15, 0, r0, c1, c0, 0    @ Read SCTLR into R0
ORR r0, r0, (1<<25)          @ Set EE bit  position
MCR p15, 0, r0, c1, c0, 0    @  Write R0 to SCTLR

The sequence will set EE bit.

Is this your answer?

Best regards,

Yasuhiko Koumoto.

hi what is the meaning of "you  may need to re-build your own compiler with big-endian support"?.

You will have to build a custom build of GCC, configured to include big-endian support.

This generally doesn't work unless you write the whole code sequence with "other endianness" in assembler. The compiler will make implicit memory accesses (such as stacking, and literal pool access) and therefore needs to have visibility / control of what the current endianness is; i.e. overriding directly via assembler is only going to work if you change back to "compiler endianness" before returning to C code.

Oh, thanks Harris

Hello Peter,

your comment is true. But I think we can dynamically access the data area by making special functions as following (for example).

Of course, these functions should be executed in the privileged mode.

How do you think about the idea?

/* short GetHalf(long *addr) */
GetHalf:
    PUSH {r10}
    MRC p15, 0, r10, c1, c0, 0  
    ORR r10, r10, (1<<25) 
    MCR p15, 0, r10, c1, c0, 0
    LDRH r0, [r0]
    MRC p15, 0, r10, c1, c0, 0  
    ORR r10, r10, (1<<25) 
    EOR r10, r10, (1<<25) 
    MCR p15, 0, r10, c1, c0, 0
    POP  {r10}
    BX  lr
/* void PutHalf(long *addr, long data) */
PutHalf:
    PUSH {r10}
    MRC p15, 0, r10, c1, c0, 0  
    ORR r10, r10, (1<<25) 
    MCR p15, 0, r10, c1, c0, 0
    STRH r1, [r0]
    MRC p15, 0, r10, c1, c0, 0  
    ORR r10, r10, (1<<25) 
    EOR r10, r10, (1<<25) 
    MCR p15, 0, r10, c1, c0, 0
    POP  {r10}
    BX  lr

Best regards,

Yasuhiko Koumoto.

How do you think about the idea?

It's a very expensive way of doing it. If you are going to do functions to load single fields from memory then just use little-endian loads and use the "rev" instruction to reverse the result - it's only one additional instruction and much faster (and small enough to be inlined so you avoid the function overhead too).

It's only worth switching the whole processor data endianness if you are going to load a large block of data, and want to avoid executing hundreds of instructions to reverse the data.

Pete