[PATCH v20 08/11] ARM: kprobes: enable OPTPROBES for ARM 32

[linux@arm.linux.org.uk (Russell King - ARM Linux)]

On Fri, Jan 09, 2015 at 05:28:22PM +0000, Jon Medhurst (Tixy) wrote:
> Using objdump I can see that the BX instruction does indeed end up in
> the code, it hasn't been auto-magically turned into a MOV PC,R2.
> 
> Adding in a ".code 16" to the assembler produces "Error: selected
> processor does not support THUMB opcodes", so at least it's got that
> right. 
> 
> I have "gcc version 4.9.1 (Ubuntu/Linaro 4.9.1-16ubuntu6)"

Remember that it's binutils which issues the errors about the assembly.

> Interestingly...
> 
> $ echo 'asm ("bx r2\n");' | arm-linux-gnueabihf-gcc -x c -S -march=armv4 -
> <stdin>:1:0: warning: target CPU does not support THUMB instructions
> $

Mine doesn't do that.

> but adding -marm gets rid of that error.
> 
> $ echo 'asm ("bx r2\n");' | arm-linux-gnueabihf-gcc -x c -S -marm -march=armv4 -
> $

Yes - but check the -.s file for the output... this won't run the assembler
so the assembler won't check that the instruction is legal.

For me:

$ echo 'asm ("bx r2\n");' | arm-linux-gcc -x c -c -marm -march=armv4 -v - -o o.o

calls the assembler thusly:

/usr/local/lib/gcc/arm-linux-gnueabi/4.7.4/../../../../arm-linux-gnueabi/bin/as \
-v -march=armv4 -meabi=5 --fix-v4bx -o o.o /tmp/ccB0cZgO.s

Sure enough, the object file contains:

00000000 <.text>:
   0:   e12fff12        bx      r2
                        0: R_ARM_V4BX   *ABS*

so it looks like it's been told...  Then if you do:

$ arm-linux-ld --fix-v4bx -o o1.o o.o
$ arm-linux-objdump -dr o1.o

you get:

    8074:       e1a0f002        mov     pc, r2

Hmm, I wonder if this means we should have the kernel linker deal with
V4BX relocations on ARMv4, converting them to their mov pc, X variant.

Also, do we need --fix-v4bx for the link of vmlinux?

-- 
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.