>
>Is it ready in your opinion?
>

Yes.  I've been using and tweaking it for about 3 weeks.  The other day I
stopped making changes so as not to introduce a bug in the 11th hour (like my
last simple patch).  I'll let you know if I find a problem, but I'm confident
that the current patch improves on the original code.  Don't mean to disparage
that code... the fundamentals were there, but it was very rough around the
edges.  xscale_analyze_trace() looks like it may have evolved into a
unmaintainable hack.  And clearly it was never used with a trace collected in
'wrap' mode, which seems to me to be the more useful mode since it lets you
examine the path of execution leading up to a breakpoint, watchpoint, or
vector trap.

If anyone has suggestions on how trace functionality should work, I'd be
interested, and maybe could implement it, since I've had my nose in this for a
while now.  Never got around to taking a look at how generic arm ETM works. 
One thing I've been wondering is why there's no option to read the
instructions directly from target memory, as opposed to relying on a
separately loaded image file.  Only thing I could think of is that the memory
mapping may have changed in between collecting the trace and running the
analysis (with some work, this issue can be addressed).  Also annoyed that
trace is always disabled on a debug break.  Don't see why it can't remain
enabled in 'wrap' mode.  I haven't seen anything in the manual that suggests
it incurs a run-time performance penalty.

Thanks,
Mike

----

_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to