Re: [patch 4/4] KVM-trace port to tracepoints

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Avi Kivity (avi@qumranet.com) wrote:
> Peter Zijlstra wrote:
>> On Tue, 2008-07-22 at 21:46 +0300, Avi Kivity wrote:
>>   
>>> Jan Kiszka wrote:
>>>     
>>>> That's true - as long as you don't have to add/remove/modify
>>>> tracepoints. I had to do this job in the past (not for KVM). Having 1
>>>> spot in 1 file (based on generic probes) would be handier in that case
>>>> than 5 spots in 3 files. But if the KVM tracepoints are considered
>>>> stable in their number and structure, that shouldn't be an issue here.
>>>>
>>>>         
>>> Tracepoints aren't stable; they are artefacts of the implementation.
>>>     
>>
>> Which IMHO makes it unsuitable for trace_mark() as that will be exported
>> to user-space, and every time you change your tracepoints you'll change
>> user visible things - not nice.
>>   
>
> They are used for debugging (mostly performance related), so changes are 
> expected.
>
> What uses of trace_mark() depend on a stable interface? blktrace?
>

Actually, LTTng likes to have the { marker name, field name } pairs
unchanged for the markers it looks for, but that's about it. If a
userspace analysis plugin fails to see a marker (because it is disabled
or changed), it just does not apply its particular analysis on the
trace.

Since the markers and marker types are self-described in the trace,
userspace can detect any change the the present markers, so there is no
need to rely on "version numbers" because we are able to proceed to a
complete marker list verification (names, field names, types) before
starting the trace analysis.

Mathieu

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68