Re: [Qemu-devel] [PATCH v3 00/14] trace: Add static tracing to QEMU

[Stefan Hajnoczi]

On Mon, Sep 6, 2010 at 6:12 PM, Anthony Liguori <address@hidden> wrote:
> On 09/06/2010 11:51 AM, Daniel P. Berrange wrote:
>>
>> On Mon, Sep 06, 2010 at 04:13:57PM +0100, Stefan Hajnoczi wrote:
>>
>>>
>>> This patch series adds static tracing to QEMU.  It can be used to
>>> instrument
>>> QEMU code by means of lightweight logging called trace events.
>>>
>>> Prerna and I are now posting the entire patch series with a serious eye
>>> towards
>>> checking we meet users' and developers' tracing needs and with the goal
>>> of
>>> getting this functionality merged into qemu.git.
>>>
>>
>> The main question would be why create a tracing framework and probe
>> markup macros specific to QEMU ? It looks like quite a few major
>> open source projects (PostgreSQL, Python, TCL, OpenJDK) are using
>> DTrace static probe markers for code instrumentation. IIUC this
>> is accessible on Solaris, (Free/Net?)-BSD, OS-X and also Linux via
>> SystemTAP's DTrace compat layer. Is this QEMU specific probe markup
>> flexible enough to make it possible to also support DTrace/SystemTAP
>> without having to add a second set of source code markers to every
>> probe point ?
>>
>
> Yes, there's a simple generator which converts are marker format to any type
> of backend.  It can be  LTTng, dtrace, or something simpler.
>
> If you look at some of the earlier threads, the basic problem is that no
> single trace point infrastructure seems to be sufficiently mature today so
> using an intermediary to delay the decision of which backend should be used
> seemed like the prudent thing to do.

When the dust settles and userspace tracing works out-of-the-box.
Right now tracers are just not there yet.  They are not upstream, they
are not packaged, or they require other admin setup.  That is not an
issue for developers with sufficient time, but it is blocker for
someone who's seeing a problem on a production system.

LTTng UST support is included in this patchset.  Adding DTrace static
probes should be easy too since QEMU's trace events are independent of
the trace backend.

Stefan