[OpenOCD-user] Cortex M7/ATSAME70, breakpoints and cache

[Thomas D. <tho...@em...>]

Hello,

I came around a serious problem regarding openOCD 0.10.0, an ATSAME70
(Cortex M7) board, gdb and breakpoints (incl. single-stepping).

The symptoms are, that when I hit a breakpoint in my application, I
can't proceed execution of my code, no matter whether I leave the
breakpoint active or I remove the breakpoint. Funny enough, in my
startup code, breakpoints work rather fine (the core stops, and with a
"resume" it continues execution).

I finally found out, that openOCD modifies the code in RAM to insert a
"breakpoint" instruction, but does not ensure, that any entries in the
instruction cache are invalidated. So when the core hits the breakpoint:

- it pulls the opcode from external RAM into the instruction cache
- it tries executing this instruction
- the core stops
- openOCD gains control again
- openOCD writes the original opcode into external RAM
- after a "resume" command, openOCD lets the core execute code again
- BUT: The core still finds the breakpoint opcode in the instruction
cache and therefore stops again instead of continueing code execution-


So my questions are:

- has anybody seen the same bahvior?
- can anybody verify my thoughts?
- is there a known fix for this behavior?
- any ideas how to proceed?

Kind regards,

Thomas.