mdw vs mdb

Development Boards MicroMod
1

I have the following defined:

extern unsigned char macAddr[6];
macAddr[0] = 0x00;
macAddr[1] = 0x0D;
macAddr[2] = 0x77;
macAddr[3] = 0x0A;
macAddr[4] = 0x0B;
macAddr[5] = 0x0C;

The Eclipse debugger tells me that macAddr is at 0x40000045

When I look at memory using the mdb command it shows this:

> mdb 0x40000045 0x7
0x40000045: 00 0d 77 0a 0b 0c 00

Just as I expect t it to be.

But when I use mdw it shows this:

> mdw 0x40000040 0xff
0x40000040: 40000f04 770d0041 000c0b0a 00001000 0000196e abcd0000 ffffffff 0000ffff
                     ^^^^^^^^^^^^^^^^^

Is this a bug, or am I ignorant about how memory works?

2

OK, I see what is happening. The mdw command shows the words “byte-swoped”. Flipping the bytes of each word puts the bytes of the array in the “correct” order.

However, when I look at a 32bit register stored on the stack, the word is displayed in the “correct” order. In other words, if R14 (LR) is 0x003654C and I look at the stack:

0x40007800: e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5
0x40007820: e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 e5e5e5e5 40005618
0x40007840: 4000786c 40007850 0003255c 000366a8 00000000 00000000 00000000 00000000
0x40007860: 40007884 40007870 000379b4 00032228 400055e8 4000789c 40007884 00034008
0x40007880: 00000000 400055d4 40004304 4000430c 40004318 400055c0 400078a4 0003654c
                                                                           ^^^^^^^^
0x400078a0: 000334e0 40000080 0000ea60 400078c8 400078bc 00036950 00036498 400078e4
0x400078c0: 400078cc 00032198 0003694c 00000000 00000000 00000001 400078fc 400078e8
0x400078e0: 00003190 0003212c 00000001 00000000 ffffffff 40007900 000010e0 00003100

There, the bytes are displayed in the “correct” order.

I don’t understand. Why is byte order of data in memory swoped in some cases but not in others?

3

The Eclipse debugger tells me that macAddr is at 0x40000045

Using word accesses for unaligned locations causes the data loaded to appear byte shifted or swapped on ARMv4/5. This is fairly typical for RISC processors. Cortex allows unaligned loads and stores at the cost of an extra bus cycle.
4

If you are using GCC you can use this macro :-

#define AL(in) attribute ((aligned (in)))

Then define your variable :-

AL(4) unsigned char ByteAreay[42];

It should be better to view in the debugger then !

If you are worried about ram wastage in the finally application you can do this

#ifdef DEBUG

#define AL(in) attribute ((aligned (in)))

#else

#define AL(in)

#endif

Then the wasted ram can be removed in your release version !