User-space tracing on ARM

User-space tracing on ARM

[Turgis, Frederic]

Hi

After using systemtap a lot to probe kernel, we would like to move on to user-space probing on ARM. This requires uprobes/utrace so I asked a few questions on utrace-devel mailing list. However this forum is also concerned:

- Is anyone probing user-space on ARM with systemtap ? I asked the question in the context of Linaro organization and I did not get anwers :-(

- how does systemtap compare to other tools for user-space ? I understand it may be the only front-end tool to dynamically instrument process functions through uprobes (or maybe LTTng with gdb tracepoints). In terms of perf, could be less good than tools focusing only on user-space but that is also a quality to focus on both kernel and user-space. And with regards to our amount of trace data, perf is not a key criteria.


Thanks for your inputs.


Regards
Fred

Frederic Turgis
OMAP Platform Business Unit - OMAP System Engineering - Platform Enablement


Texas Instruments France SA, 821 Avenue Jack Kilby, 06270 Villeneuve Loubet. 036 420 040 R.C.S Antibes. Capital de EUR 753.920

Re: User-space tracing on ARM

[Frank Ch. Eigler]

Hi, Fred -

> [...]  Is anyone probing user-space on ARM with systemtap ? I asked
> the question in the context of Linaro organization and I did not get
> anwers :-(

Due to the unfortunate lack of a utrace | uprobes port for ARM,
systemtap's support for userspace on your platform is still very
limited.

The medium-term way forward would appear to be porting the proposed
ibm uprobes lkml patches to ARM, and waiting/helping till systemtap is
able to make use of those.

> - how does systemtap compare to other tools for user-space ? 
> [...]

For joint kernel+userspace dynamic+static tracing+probing, systemtap
has been the only available alternative.  For different subsets of
functionality, with different performance/portability tradeoffs, the
other tools may suit well.  Perhaps we should add a few more
rows/columns to the (increasingly misnamed) comparison page [1].

[1] http://sourceware.org/systemtap/wiki/SystemtapDtraceComparison

- FChE

RE: User-space tracing on ARM

[Turgis, Frederic]

> For joint kernel+userspace dynamic+static tracing+probing, systemtap has been the only available alternative.  For different
> subsets of functionality, with different performance/portability tradeoffs, the other tools may suit well.
> Perhaps we should add a few more rows/columns to the (increasingly misnamed) comparison page [1].

I think even just for user-space dynamic instrumentation, I did not see anything as versatile as systemtap through its scripting language (be it perf, lttng, dyninst, ...). But as you mention, there are always trade-offs and I can probably live without all versatility I am dreaming of !

Thanks

Regard
Fred
Texas Instruments France SA, 821 Avenue Jack Kilby, 06270 Villeneuve Loubet. 036 420 040 R.C.S Antibes. Capital de EUR 753.920