byteorder problem

byteorder problem

[Antony Pavlov]

Hi!

Yesterday I have investigated strange clocksource behaviour of JZ4755
MIPS CPU: the function cyc2ns() works wrong.

The problem was in realisation of the '>>' u64 operation (see __lshrdi3()).

In include/linux/byteorder/generic.h we have

#ifndef __LITTLE_ENDIAN
#define __LITTLE_ENDIAN        1234
#endif
#ifndef __BIG_ENDIAN
#define __BIG_ENDIAN   4321
#endif

so after including this file we will have __BIG_ENDIAN and
__LITTLE_ENDIAN both defined.

On the other hand in arch/mips/lib/libgcc.h we have

#ifdef __BIG_ENDIAN
struct DWstruct {
        int high, low;
};
#elif defined(__LITTLE_ENDIAN)
struct DWstruct {
        int low, high;
};
#else
#error I feel sick.
#endif

In any case we will got big-endian DWstruct, so on little-endian MIPS
machines we will have wrong result of the '>>' u64 operation (the low
and high parts of 64-bit word will be swapped).

The problem can be solved by this patch:

--- a/include/linux/byteorder/generic.h
+++ b/include/linux/byteorder/generic.h
@@ -78,13 +78,6 @@
  *
  */

-#ifndef __LITTLE_ENDIAN
-#define __LITTLE_ENDIAN        1234
-#endif
-#ifndef __BIG_ENDIAN
-#define __BIG_ENDIAN   4321
-#endif
-

But in several places we have something like this:

#if __BYTE_ORDER == __LITTLE_ENDIAN
...
#elif __BYTE_ORDER == __BIG_ENDIAN

and we will get the warnings from compiler (e.g. warning:
"__BIG_ENDIAN" is not defined). So we need addition changes like this

-#if __BYTE_ORDER == __LITTLE_ENDIAN
+#ifdef __LITTLE_ENDIAN
...
-#elif __BYTE_ORDER == __BIG_ENDIAN
+#elif defined(__BIG_ENDIAN)

There is another solution. We can change arch/mips/lib/libgcc.h:

-#ifdef __LITTLE_ENDIAN
+#if __BYTE_ORDER == __LITTLE_ENDIAN
...
-#elif defined(__BIG_ENDIAN)
+#elif __BYTE_ORDER == __BIG_ENDIAN

But this solution is not linux-compatible because __LITTLE_ENDIAN and
__BIG_ENDIAN can be defined simultaneously.

Please comment this message.

-- 
Best regards,
  Antony Pavlov

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox