[PATCH 1/4] ARM: kprobes: fix instruction fetch order with <asm/opcodes.h>

[Dave.Martin@arm.com (Dave Martin)]

On Wed, Jul 31, 2013 at 08:43:07PM +0100, Ben Dooks wrote:
> On 29/07/13 09:01, Jon Medhurst (Tixy) wrote:
> >On Thu, 2013-07-25 at 22:08 +0100, Ben Dooks wrote:
> >>If we are running BE8, the data and instruction endian-ness do not
> >>match, so use<asm/opcodes.h>  to correctly translate memory accesses
> >>into ARM instructions.
> >>
> >>Signed-off-by: Ben Dooks<ben.dooks@codethink.co.uk>y
> >>---
> >>  arch/arm/kernel/kprobes-common.c |   14 ++++++++------
> >>  arch/arm/kernel/kprobes.c        |    9 +++++----
> >>  2 files changed, 13 insertions(+), 10 deletions(-)
> >>
> >>diff --git a/arch/arm/kernel/kprobes-common.c b/arch/arm/kernel/kprobes-common.c
> >>index 18a7628..c0b202e 100644
> >>--- a/arch/arm/kernel/kprobes-common.c
> >>+++ b/arch/arm/kernel/kprobes-common.c
> >>@@ -14,6 +14,7 @@
> >>  #include<linux/kernel.h>
> >>  #include<linux/kprobes.h>
> >>  #include<asm/system_info.h>
> >>+#include<asm/opcodes.h>
> >>
> >>  #include "kprobes.h"
> >>
> >>@@ -305,7 +306,8 @@ kprobe_decode_ldmstm(kprobe_opcode_t insn, struct arch_specific_insn *asi)
> >>
> >>  	if (handler) {
> >>  		/* We can emulate the instruction in (possibly) modified form */
> >>-		asi->insn[0] = (insn&  0xfff00000) | (rn<<  16) | reglist;
> >>+		asi->insn[0] = __opcode_to_mem_arm((insn&  0xfff00000) |
> >>+						   (rn<<  16) | reglist);
> >>  		asi->insn_handler = handler;
> >>  		return INSN_GOOD;
> >>  	}
> >>@@ -338,9 +340,9 @@ prepare_emulated_insn(kprobe_opcode_t insn, struct arch_specific_insn *asi,
> >>  		thumb_insn[2] = 0x4770; /* Thumb bx lr */
> >
> >The line above and the one before it not in this diff also need fixing
> >by changing to __opcode_to_mem_thumb16(0x4770).
> 
> IIRC, these are then fixed by the set_emulated instruction, which is
> also why the insn return is not swapped.

Ah, I see.

I think it would be better to have consistent behaviour between the ARM
and Thumb cases here: in its current form, this patch makes
set_emulated_insn() responsible for the swabbing in the Thumb case and
not in the ARM case.

Doing it in the same place for both cases would be preferable ... or was
there a reason why this doesn't work?

Cheers
---Dave

> >>  		return insn;
> >>  	}
> >>-	asi->insn[1] = 0xe12fff1e; /* ARM bx lr */
> >>+	asi->insn[1] = __opcode_to_mem_arm(0xe12fff1e); /* ARM bx lr */
> >>  #else
> >>-	asi->insn[1] = 0xe1a0f00e; /* mov pc, lr */
> >>+	asi->insn[1] = __opcode_to_mem_arm(0xe1a0f00e); /* mov pc, lr */
> >>  #endif
> >>  	/* Make an ARM instruction unconditional */
> >>  	if (insn<  0xe0000000)
> >>@@ -360,12 +362,12 @@ set_emulated_insn(kprobe_opcode_t insn, struct arch_specific_insn *asi,
> >>  	if (thumb) {
> >>  		u16 *ip = (u16 *)asi->insn;
> >>  		if (is_wide_instruction(insn))
> >>-			*ip++ = insn>>  16;
> >>-		*ip++ = insn;
> >>+			*ip++ = ___asm_opcode_to_mem_thumb16(insn>>  16);
> >>+		*ip++ = ___asm_opcode_to_mem_thumb16(insn);
> >

[...]