Re: ARM linker stub changes - placement, PIC

[Christophe LYON <christophe dot lyon at st dot com>]

On 24.02.2009 04:09, Daniel Jacobowitz wrote:

> hardware interrupt controller.  So this should stop dropping stubs on
> top of the reset vector.  No change in testsuite results, but a lot
> of churn in the expected output.

Indeed, thanks for the effort!

One comment, not linked to this patch: on several occasions, I can see 
that andeq r0,r0,r0 (arm) / lsls r0,r0,r0 (thumb) is used as padding 
(opcode=0). Shouldn't there be proper "nop"s instead? (see for instance 
farcall-mix2 test)


> The second changes "add pc, pc, ip" to "add ip, pc, ip; bx ip" when
> the target is Thumb.  It turns out that add pc does not mode switch on
> ARM v6, but does on ARM v7 (sometimes, under some complicated set of
> conditions).  Clearly better to avoid it.  This fixes three crashes
> when I locally converted the linker dump tests to execution tests.
>
> Comments welcome!  Credit to Joseph for the code in the first patch.

Thanks, both look OK to me.

Christophe.