XScale: setting hw breakpoint in Linux kernel possible?

Development Boards MicroMod
1

Hi folks,

I’d like to set a breakpoint in a kernel I don’t have source code for, as I am just trying to see how it initializes some hardware… ie pci.

I have an Olimex Arm-Usb-OCD, and I have used OpenOCD v728 for most of my time with OpenOCD. Yesterday I pulled rev 1192 from svn. Whew! It has changed alot! Xscale support has improved. I’ll just name 2 items… but there are more.

xscale debug_handler

xscale vector_catch

I hope those 2 new additions can help me do what I need to do. I believe I have it setup correctly. It detects my flash chip, and I have working_area setup to speed things up a bit. :slight_smile:

When I get the target halted, with pc=0, and then do a resume, it will proceed for a bit but inevitably gets stuck with pc=0xffff0018. At that point the mmu, D-Cache, I-Cache are also enabled by that kernel.

Documents I have found that discuss what may be happening here.

www.arium.com/support/pdf/XScale_app_note.pdf

http://www.webetools.com/ultsol/users/k … d=39&sid2=

Dominic Rath’s master_thesis.pdf ‘out there’ mentions some of the inabilities of open source debuggers and mmu units. (I hope what he says there is still not the case!)

In openocd/src/target/xscale.c line 1735 it appears to use some hard coded locations… at 0x0 and 0xffff0000 in calls to xscale_load_ic but that may be insignificant.

I guess I should also mention this is an older 2.4 kernel.

I hope that I can get around this, even if I have to set some memory watches at the vector tables and do some fudging to make sure the mini ic is in agreement with the system.

2

well you learn something new everyday…

I had previously tried to set a wp point up at 0xffff0008, and then at 0xffff0012. Apparently I didn’t go high enough.

wp 0xFFFF001C 4 w

was the command I needed. Thank You!

As I was instructed:

"…you need to place a hw breakpoint at the point where the exception vectors get re-written to give the openocd a chance to re-read the new vectors otherwise your xscale is going to use stale data from the mini i-cache.

Breaking and resuming at some point after the new vectors have been written, but before the first exception occurs, solves this problem."

Again that’s:

mw 0xffff001c 4 w