Re: MIPS RT debug support

[john cooper <john.cooper@third-harmonic.com>]

Steven Rostedt wrote:
> On Thu, 15 Nov 2007, Tim Bird wrote:
>
> > john cooper wrote:
> > > The more daunting problem stems from limitations in the MIPS
> > > ABI which makes the latency trace support problematic.
> > > Rather than rehash the issue:
> > >
> > >     http://lists.linuxcoding.com/kernel/2005-q4/msg10163.html
> > >
> > > Until we have a usable instrumentation solution in place,
> > > characterization, debug, and support of PREEMPT_RT for MIPS
> > > is going to be a challenge.
> > Agreed.  I have been using KFT (Kernel Function Trace)
> > on MIPS, and it has decent support for function traceback
> > reporting, but it's not currently integrated with latency-trace
> > at all.   We should discuss if this could possibly be
> > used to debug RT-preempt.  It is much heavier weight than
> > the mcount stuff, but uses similar (but not identical)
> > gcc profiling instrumentation.  I'm not sure if the
> > two can be turned on together, or how hard it would
> > be to move latency-trace onto -finstrument_functions.
>
> I'm not familiar with the KFT but I'm sure it would be easy to port
> latency_trace to it. Really, all the mcount does is make a wrapper to pass
> to the trace calls.

It isn't an issue of getting a hook into the FUNCTION_PROLOGUE
(mcount() here) but rather of emulating the CALLER_ADDR[0123]
defs which map onto the gcc internal __builtin_return_address().
Doing so using the affectionately dubbed "Three Stooges Algorithm"
called out in the MIPS ABI is both complex and nondeterministic.

The entry FUNCTION_PROLOGUE hook provides a means to log the path
traveled thus far.  __builtin_return_address() gives a way to
snapshot a stub of the stack invocation. Together they provide
somewhat complimentary but useful debug information.

We could log the stack invocation progress in the latency
instrumentation itself by noting a new invocation in the
FUNCTION_PROLOGUE and unwind of the same in the currently
unused FUNCTION_EPILOGUE hook.  So we're just shadowing the
actual runtime stack in effect with a data structure which
can be traversed far more easily.

-john

--
john.cooper@third-harmonic.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/