HW watchpoint on Linux/ARM

HW watchpoint on Linux/ARM

[Christian Gagneraud]

Hi there,

Does anyone know if it is possible to use HW watchpoint on ARM platform?

I have tried this feature on a Freescale i.mx6 CPU and a TI Sitara
(both ARM cortex), with GDB 7.12 and was unsuccessful so far.

I did some research, and it seems to me that this feature is supported
by both GDB for ARM and Linux for ARM

Support in GDB 7.12.1:
https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=blob;f=gdb/gdbserver/linux-arm-low.c;h=b2b22581634489bc68130620428cda25f6ae43a0;hb=59b102452f0bdab7a9b19d9fdbb8ce5253f0ce9f

Support in Linux 4.9:
https://github.com/torvalds/linux/blob/v4.9/arch/arm/kernel/hw_breakpoint.c

When I tried with my i.mx6 system, I thought that my system might be a
bit old, so I decided to try with my BeagleBone system, which run a
much more recent software stack (Linux 4.9.37, GDB 7.12.1).

At this point i don't know which of GDB or Linux (or me) is to blame.
Has anyone had any success with hardware watchpoint on ARM (Cortex) platform?

Any sort of feedback appreciated.

Update:
On BeagleBone, the blame goes to Linux, it doesn't support
ARM_DEBUG_ARCH_V7_MM (memory-mapped interface), the kernel says
"hw-breakpoint: debug architecture 0x4 unsupported."
But on i.mx6, it does. From memory, the kernel told me that I have
access to 5 HW breakpoints and 1 HW watchpoint (I don't have the board
at hand, but will post details on Monday).


Thanks,
Chris

Re: HW watchpoint on Linux/ARM

[Christian Gagneraud]

On 16 July 2017 at 20:52, Peter Schauer <peterschauer@gmx.net> wrote:
>> But on i.mx6, it does. From memory, the kernel told me that I have
>> access to 5 HW breakpoints and 1 HW watchpoint (I don't have the board
>> at hand, but will post details on Monday).
>
> I had the same problem with an i.mx6 board three years ago.
> I don't remember all the details, but it had to do with the
> ARM secure debug mode, which is active when the SPIDEN pin is pulled low
> in hardware.
>
> I had a kernel hack back then, which enabled the hardware break mechanism
> even if the SPIDEN pin is pulled low, but it disables the secure debug mode,
> and worked only for my arm architecture (I couldn't test others), so
> I didn't post it to the kernel group.
>
> If your problem is caused by SPIDEN low, and you are interested in
> the kernel hack, please let me know.
> I could also send you a detailed analysis of the problem in german,
> in case you are interested.

Hi Peter,

Yes please, I'm interested by your hacky patch, and i'll take the
german analysis too! ;)

Thanks,
Chris

>
> --
> Peter Schauer                   Peter.Schauer@mytum.de