This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: Fix assembly of Thumb pcrel LDRs against global symbols
- From: Richard Sandiford <richard dot sandiford at linaro dot org>
- To: Richard Earnshaw <rearnsha at arm dot com>
- Cc: binutils at sourceware dot org, patches at linaro dot org
- Date: Wed, 02 Mar 2011 16:35:57 +0000
- Subject: Re: Fix assembly of Thumb pcrel LDRs against global symbols
- References: <g4tyfoxqlj.fsf@linaro.org> <1299079089.24968.14.camel@e102346-lin.cambridge.arm.com>
Richard Earnshaw <rearnsha@arm.com> writes:
> On Mon, 2011-02-28 at 12:59 +0000, Richard Sandiford wrote:
>> The PC-relative LDR instructions have no associated relocation,
>> so even LDRs for global symbols should be resolved by the assembler.
>> We currently handle this correctly for single-register ARM loads,
>> but we're missing the associated relocation types for LDRD and Thumb.
>> This leads to errors like:
>
> I'm not sure I agree with this. If I write
>
> .global foo
>
> ...
> ldr r0, foo
>
> ...
>
> foo:
> ...
>
> but then at link/load time pre-empt foo with some other definition, that
> will silently leave me with the wrong answer.
Yeah, I realise we normally try to do that. The point was that we
already resolve these references at assembly time for ARM LDRs --
.syntax unified
ldr r3,foo
nop
.globl foo
foo:
.word 0x1234
-- just not for Thumb LDRs or for (any) LDRDs. I think I'd wrongly assumed
that the current ARM LDR behaviour was by design, and that we should make
the other cases match. However, looking at the ABI again, I see there is
a reloc (R_ARM_THM_PC12) that we can use here, but that we aren't using.
So if the current ARM behaviour is wrong, then I suppose there are
three bugs:
- We're using a generic ARM_OFFSET_IMM fixup for ARM LDRs, which is
always getting resolved by the assembler. We should generate a
reloc instead for this case (but not for some other ARM_OFFSET_IMM cases).
- We're using a generic T32_OFFSET_IMM fixup for Thumb LDRs, which is
always triggering an error (one of the two in my message). We should
generate a reloc instead (but not for some other T32_OFFSET_IMM cases).
- We generate the internal error for things like:
.syntax unified
.thumb_func
ldr r3,[r12,#foo-.]
nop
.globl foo
foo:
.word 0x1234
which AFAICS couldn't be handled by relocations. (The patch fixes
this too.)
That's a bit more work than I'd anticipated, and the first change might
annoy some people, so I think I'd better put it to one side for now. :-)
Richard