Re: [Xen-devel] [PATCH] tools: delete gtraceview and gtracestat

[George Dunlap <george.dunlap@xxxxxxxxxx>]

On 02/09/16 12:07, Wei Liu wrote:
> On Fri, Sep 02, 2016 at 12:05:49PM +0100, George Dunlap wrote:
>> On 30/08/16 13:50, Ian Jackson wrote:
>>> Wei Liu writes ("[PATCH] tools: delete gtraceview and gtracestat"):
>>>> There has not been any substantial update to them since 2011. My quick
>>>> check shows that they don't work.
>>>>
>>>> Just delete them. It would be easy to resurrect them from git log should
>>>> people still need them.
>>>
>>> I'm not sure what these are.  git log is not particularly
>>> illuminating.  gtraceview.c and some of the commit messages talk about
>>> "Cx events" or "x86 Cx" and there is an Intel copyright notice, but
>>> not Intel email addresses.
>>>
>>> There is also some discussion of xentrace in the source code.
>>>
>>> Are these the only in-tool consumer of some kind of hypervisor
>>> -generated or -mediated data ?  If so it should be retained IMO, or
>>> the corresponding hypervisor code removed too.
>>>
>>> I have added George to the CC in the hope that he may know more...
>>
>> Well I'd never looked at it before, but it does indeed seem to be
>> something about skimming through xentrace records.
>>
>> The help for both functions says to run "xentrace -e 0x80f000", which
>> will capture all events of type TRC_HW, which includes TRC_PM.  The code
>> (again for both) then seems to parse events of type TRC_PM_IDLE_ENTRY,
>> TRC_PM_IDLE_EXIT, and TRC_PM_FREQ_CHANGE.
>>
>> gtraceview seems to be an ncurses-based browser; gtracestat seems to be
>> a xenalyze-style statistical analysis thing.
>>
>> So it would seem to be the case that they consume hypervisor-generated
>> data, but they are not the *only* consumers of that data
>> (xentrace_format and xenalyze being the other ones).
>>
>> When you say they "don't work", what do you mean?
>>
> 
> Tried to run it. It failed to produce anything useful. It is bit-rotten.

You took a trace and passed the resulting binary to it then?

If so, it's probably just the case that the trace record format has
changed somewhat and these weren't updated.  It would probably be fairly
straightforward to fix; but if nobody's using it, I'd just as soon
delete it.  The "gtracestat" functionality would probably better be
added to xenalyze in any case.

 -George


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel