Re: [PATCH tracing/kprobes 0/7] tracing/kprobes: kprobe-based event tracer update and perf support

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Sep 11, 2009 at 03:03:35PM -0400, Frank Ch. Eigler wrote:
> Frederic Weisbecker <fweisbec@gmail.com> writes:
> 
> > [...]  I'm really looking forward seeing this C expression-like
> > kprobe creation tool.  It seems powerful enough to replace printk +
> > kernel rebuild.  No need anymore to write some printk to debug,
> > worrying, [...]
> 
> To a large extent, systemtap had delivered this already some years
> ago, including the cushy ponies dancing in the sunlight.  While such
> low-level machinery is fine, some of our experience indicates that it
> is dramatically easier to use if high-level, symbolic, debugging data
> is used to compute probe locations and variable names/types/locations.
> 
> It is also too easy to stress the low-level machinery beyond its
> humble origins, in this case meaning putting probes in all kinds of
> tender spots that go "ouch".  The kprobes robustness patches coming in
> are great and will benefit all of our efforts, but it will be awhile
> until the kernel can survive a fuzz/crashme type stress test on that
> subsystem.  So expect ongoing effort there.


Fully agreed! The more I see corner recursivity cases, the more I think
we'll never fix every potential cases. But yeah it's worth trying to
fix all of them that are reported/anticipated, the more such case
are covered, the more it's usable.