[arm] force gdb into disassembling in thumb (arm) mode

[arm] force gdb into disassembling in thumb (arm) mode

[Adrian von Bidder]

Heyho!

[May have been asked before - just point me in the general direction as
I couldn't find it with google or in the mailing lsit archive]

How can I force gdb to disassemble a memory region in thumb?

gdb 5.1 gets it right if there is symbol information, but just
disassembles to arm instructions when no symbol information is present
(apparently). Can I change this default to thumb (runtime), or is it
possible with a argument to the x/i command?

The sensible way (imho) would be to interpret the (h)alfword/(w)ord
format letters, so that x/ih and x/iw would force to interpret 16/32 bit
instructions.

x/i on an odd address currently does switch to thumb mode - but it seems
to interpret the instructions at the odd address (it doesn't strip the
lsb when loading the data):

(gdb) x/20i 0xbe1
0xbe1:	ldrsh	r5, [r6, r6]
0xbe3:	strb	r6, [r0, r5]
0xbe5:	ldr	r4, [pc, #280]	(0xd00)
0xbe7:	add	sp, #280
0xbe9:	undefined instruction 0x47b4
0xbeb:	strh	r6, [r0, #2]

but the halfword at 0xbe0 does (in this example) disassemble to 
ldr r7,[sp, #0xb5].

greets from Zürich
-- vbi

Re: [arm] force gdb into disassembling in thumb (arm) mode

[Richard Earnshaw]

> The sensible way (imho) would be to interpret the (h)alfword/(w)ord
> format letters, so that x/ih and x/iw would force to interpret 16/32 bit
> instructions.

Yes, something like this would be nice.

> x/i on an odd address currently does switch to thumb mode - but it seems
> to interpret the instructions at the odd address (it doesn't strip the
> lsb when loading the data):
> 
> (gdb) x/20i 0xbe1
> 0xbe1:	ldrsh	r5, [r6, r6]
> 0xbe3:	strb	r6, [r0, r5]
> 0xbe5:	ldr	r4, [pc, #280]	(0xd00)
> 0xbe7:	add	sp, #280
> 0xbe9:	undefined instruction 0x47b4
> 0xbeb:	strh	r6, [r0, #2]

Hmm, a quick test on the CVS sources shows that using an odd address will 
DTRT (ie force thumb mode but then round the address down); I can't see 
anything from the commit log that might have changed.  The key is this 
line of code:

gdb/arm-tdep.c/gdb_print_insn_arm()
      memaddr = UNMAKE_THUMB_ADDR (memaddr);

Which strips off the thumb bit before actually loading the half-word.

R.

Re: [arm] force gdb into disassembling in thumb (arm) mode

[Adrian von Bidder]

On Wed, 2002-01-23 at 11:44, Richard Earnshaw wrote:

> Hmm, a quick test on the CVS sources shows that using an odd address will 
> DTRT (ie force thumb mode but then round the address down); I can't see 
> anything from the commit log that might have changed.  The key is this 
> line of code:
> 
> gdb/arm-tdep.c/gdb_print_insn_arm()
>       memaddr = UNMAKE_THUMB_ADDR (memaddr);
> 
> Which strips off the thumb bit before actually loading the half-word.

Sorry, my fault, mostly - Memory didn't contain what I thought it did,
which did confuse me, together with the memory addresses on the
disassembler output having the lsb set. But you're right, the data
actually displayed is at the right addresses.

But then I should be able to do "display/i $pc+1" but this has no effect
(neither does disp/i $pc|1):
(gdb) disp/i $pc+1
4: x/i $pc + 1  0xf54:	undefined

(is the same as plain disp/i $pc - it tries to decode arm).

while 
(gdb) x/i $pc+1
0xf55:	push	{r4, r5, r7, lr}

shows the thumb instruction really at 0xf54.

greets from Zürich
-- vbi