Xen devel - Apr 2009 - [PATCH] add long interrupt measurement capability

Hi Keir --

Since I see you are still taking boot-option patches
for 3.4, I thought I''d retry this one which optionally
measures cycles spent in each irq.  I thought
you had just decided it was too late for 3.4... if
you rejected it for another reason, please let me
know.

Thanks,
Dan
> From: Dan Magenheimer 
> Sent: Tuesday, April 14, 2009 9:39 AM
> To: Keir Fraser; Tian, Kevin; Xen-Devel (E-mail)
> Subject: RE: [Xen-devel] million cycle interrupt
>
> But first, Keir, would you take this patch so that this
> kind of issue can be more easily diagnosed in the future?
> If you''re worried about the extra two global variable
> reads, you could wrap an "#ifndef NDEBUG" around it
> (with #else #define opt_measure_irqs 0), but it might
> be nice to have this diagnostic capability in released
> code.
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 29/04/2009 17:50, "Dan Magenheimer"
<dan.magenheimer@oracle.com> wrote:
> Since I see you are still taking boot-option patches
> for 3.4, I thought I''d retry this one which optionally
> measures cycles spent in each irq.  I thought
> you had just decided it was too late for 3.4... if
> you rejected it for another reason, please let me
> know.
I don''t think it has sufficient general utility to be applied. Anyone
who
would actually have the want and wit to run this option can apply the patch
themselves.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Dan, Couldn''t you add this to xentrace ?

-dulloor

On Wed, Apr 29, 2009 at 2:03 PM, Keir Fraser
<keir.fraser@eu.citrix.com>wrote:
> On 29/04/2009 17:50, "Dan Magenheimer"
<dan.magenheimer@oracle.com> wrote:
>
> > Since I see you are still taking boot-option patches
> > for 3.4, I thought I''d retry this one which optionally
> > measures cycles spent in each irq.  I thought
> > you had just decided it was too late for 3.4... if
> > you rejected it for another reason, please let me
> > know.
>
> I don''t think it has sufficient general utility to be applied.
Anyone who
> would actually have the want and wit to run this option can apply the patch
> themselves.
>
>  -- Keir
>
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Perhaps.  However measuring cycles is important and, more
specifically, measuring MAX cycles spent across a set of
interrupts.  As a result, I suspect any code that measures
this (regardless of whether the result is reported by
xentrace or debug-key) would likely encounter the
same objection from Keir.

-----Original Message-----
From: Dulloor [mailto:dulloor@gmail.com]
Sent: Thursday, May 07, 2009 6:14 PM
To: Keir Fraser
Cc: Dan Magenheimer; Xen-Devel (E-mail)
Subject: Re: [Xen-devel] Re: [PATCH] add long interrupt measurement capability


Dan, Couldn''t you add this to xentrace ?

-dulloor


On Wed, Apr 29, 2009 at 2:03 PM, Keir Fraser <keir.fraser@eu.citrix.com>
wrote:

On 29/04/2009 17:50, "Dan Magenheimer"
<dan.magenheimer@oracle.com> wrote:
> Since I see you are still taking boot-option patches
> for 3.4, I thought I''d retry this one which optionally
> measures cycles spent in each irq.  I thought
> you had just decided it was too late for 3.4... if
> you rejected it for another reason, please let me
> know.

I don''t think it has sufficient general utility to be applied. Anyone
who
would actually have the want and wit to run this option can apply the patch
themselves.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> Perhaps.  However measuring cycles is important and, more
> specifically, measuring MAX cycles spent across a set of
> interrupts.  As a result, I suspect any code that measures
> this (regardless of whether the result is reported by
> xentrace or debug-key) would likely encounter the
> same objection from Keir.
I can''t imagine there''d be any objection to adding trace
macros to record this. The xentrace log processing tool can then be updated to
generate max or histogram values.

Ian 
> -----Original Message-----
> From: Dulloor [mailto:dulloor@gmail.com]
> Sent: Thursday, May 07, 2009 6:14 PM
> To: Keir Fraser
> Cc: Dan Magenheimer; Xen-Devel (E-mail)
> Subject: Re: [Xen-devel] Re: [PATCH] add long interrupt measurement
> capability
> 
> 
> Dan, Couldn''t you add this to xentrace ?
> 
> -dulloor
> 
> 
> On Wed, Apr 29, 2009 at 2:03 PM, Keir Fraser
> <keir.fraser@eu.citrix.com> wrote:
> 
> On 29/04/2009 17:50, "Dan Magenheimer"
<dan.magenheimer@oracle.com>
> wrote:
> 
> > Since I see you are still taking boot-option patches
> > for 3.4, I thought I''d retry this one which optionally
> > measures cycles spent in each irq.  I thought
> > you had just decided it was too late for 3.4... if
> > you rejected it for another reason, please let me
> > know.
> 
> 
> I don''t think it has sufficient general utility to be applied.
Anyone
> who
> would actually have the want and wit to run this option can apply the
> patch
> themselves.
> 
>  -- Keir
> 
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 08/05/2009 22:53, "Ian Pratt" <Ian.Pratt@eu.citrix.com>
wrote:
>> Perhaps.  However measuring cycles is important and, more
>> specifically, measuring MAX cycles spent across a set of
>> interrupts.  As a result, I suspect any code that measures
>> this (regardless of whether the result is reported by
>> xentrace or debug-key) would likely encounter the
>> same objection from Keir.
> 
> I can''t imagine there''d be any objection to adding trace
macros to record
> this. The xentrace log processing tool can then be updated to generate max
or
> histogram values.
Yes, xentrace records would be okay. It''s adding another debug key and
printing to Xen console for this purpose which I do not think is worthwhile.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Here is a xentrace patch.

- Should irq_desc_measure_t move to some .h file ?
- I have defined the new trace event in general class. Is it fine ?

- I have defined tsc_in as volatile to avoid initializing it in the main
code path.

thanks
dulloor

On Sat, May 9, 2009 at 3:56 AM, Keir Fraser
<keir.fraser@eu.citrix.com>wrote:
> On 08/05/2009 22:53, "Ian Pratt" <Ian.Pratt@eu.citrix.com>
wrote:
>
> >> Perhaps.  However measuring cycles is important and, more
> >> specifically, measuring MAX cycles spent across a set of
> >> interrupts.  As a result, I suspect any code that measures
> >> this (regardless of whether the result is reported by
> >> xentrace or debug-key) would likely encounter the
> >> same objection from Keir.
> >
> > I can''t imagine there''d be any objection to adding
trace macros to record
> > this. The xentrace log processing tool can then be updated to generate
> max or
> > histogram values.
>
> Yes, xentrace records would be okay. It''s adding another debug key
and
> printing to Xen console for this purpose which I do not think is
> worthwhile.
>
>  -- Keir
>
>
>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> > I can''t imagine there''d be any objection to adding
trace
> macros to record
> > this. The xentrace log processing tool can then be updated 
> to generate max or
> > histogram values.
> 
> Yes, xentrace records would be okay. It''s adding another debug key
and
> printing to Xen console for this purpose which I do not think 
> is worthwhile.
Well, for the record, the original patch does NOT add
another debug key, it just adds data to an existing one.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

I guess you can handle it much simpler by stamping a record in both irq_enter
and irq_exit. All the statistics jobs are left for xentrace script to digest.

Also I'd call it as trace_irq instead of trace_guest_irq. :-)

Thanks,
Kevin

________________________________
From: xen-devel-bounces@lists.xensource.com
[mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Dulloor
Sent: 2009年5月9日 17:12
To: Keir Fraser
Cc: Ian Pratt; Xen-Devel (E-mail); Dan Magenheimer
Subject: Re: [Xen-devel] Re: [PATCH] add long interrupt measurement capability

Here is a xentrace patch.

- Should irq_desc_measure_t move to some .h file ?
- I have defined the new trace event in general class. Is it fine ?

- I have defined tsc_in as volatile to avoid initializing it in the main code
path.

thanks
dulloor

On Sat, May 9, 2009 at 3:56 AM, Keir Fraser
<keir.fraser@eu.citrix.com<mailto:keir.fraser@eu.citrix.com>> wrote:
On 08/05/2009 22:53, "Ian Pratt"
<Ian.Pratt@eu.citrix.com<mailto:Ian.Pratt@eu.citrix.com>> wrote:
>> Perhaps.  However measuring cycles is important and, more
>> specifically, measuring MAX cycles spent across a set of
>> interrupts.  As a result, I suspect any code that measures
>> this (regardless of whether the result is reported by
>> xentrace or debug-key) would likely encounter the
>> same objection from Keir.
>
> I can't imagine there'd be any objection to adding trace macros to
record
> this. The xentrace log processing tool can then be updated to generate max
or
> histogram values.
Yes, xentrace records would be okay. It's adding another debug key and
printing to Xen console for this purpose which I do not think is worthwhile.

 -- Keir





_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

- Changed the name to trace_irq :)

- trace_irq dumps just tsc_in and tsc_out values in a single record. I guess
there is no need to write two records (wasting trace-buf mem, more
processing, additional logic in xentrace and/or xentrace_format).

- xentrace_format does what Dan wanted.

-dulloor

2009/5/11 Tian, Kevin <kevin.tian@intel.com>
>  I guess you can handle it much simpler by stamping a record in both
> irq_enter and irq_exit. All the statistics jobs are left for xentrace
script
> to digest.
>
> Also I''d call it as trace_irq instead of trace_guest_irq. :-)
>
> Thanks,
> Kevin
>
>  ------------------------------
> *From:* xen-devel-bounces@lists.xensource.com [mailto:
> xen-devel-bounces@lists.xensource.com] *On Behalf Of *Dulloor
> *Sent:* 2009年5月9日 17:12
> *To:* Keir Fraser
> *Cc:* Ian Pratt; Xen-Devel (E-mail); Dan Magenheimer
> *Subject:* Re: [Xen-devel] Re: [PATCH] add long interrupt measurement
> capability
>
> Here is a xentrace patch.
>
> - Should irq_desc_measure_t move to some .h file ?
> - I have defined the new trace event in general class. Is it fine ?
>
> - I have defined tsc_in as volatile to avoid initializing it in the main
> code path.
>
> thanks
> dulloor
>
> On Sat, May 9, 2009 at 3:56 AM, Keir Fraser
<keir.fraser@eu.citrix.com>wrote:
>
>> On 08/05/2009 22:53, "Ian Pratt"
<Ian.Pratt@eu.citrix.com> wrote:
>>
>> >> Perhaps.  However measuring cycles is important and, more
>> >> specifically, measuring MAX cycles spent across a set of
>> >> interrupts.  As a result, I suspect any code that measures
>> >> this (regardless of whether the result is reported by
>> >> xentrace or debug-key) would likely encounter the
>> >> same objection from Keir.
>> >
>> > I can''t imagine there''d be any objection to
adding trace macros to
>> record
>> > this. The xentrace log processing tool can then be updated to
generate
>> max or
>> > histogram values.
>>
>> Yes, xentrace records would be okay. It''s adding another debug
key and
>> printing to Xen console for this purpose which I do not think is
>> worthwhile.
>>
>>  -- Keir
>>
>>
>>
>


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi Keir,

Can you consider this for check-in.

-dulloor

2009/5/11 Dulloor <dulloor@gmail.com>
> - Changed the name to trace_irq :)
>
> - trace_irq dumps just tsc_in and tsc_out values in a single record. I
> guess there is no need to write two records (wasting trace-buf mem, more
> processing, additional logic in xentrace and/or xentrace_format).
>
> - xentrace_format does what Dan wanted.
>
> -dulloor
>
> 2009/5/11 Tian, Kevin <kevin.tian@intel.com>
>
>  I guess you can handle it much simpler by stamping a record in both
>> irq_enter and irq_exit. All the statistics jobs are left for xentrace
script
>> to digest.
>>
>> Also I''d call it as trace_irq instead of trace_guest_irq. :-)
>>
>> Thanks,
>> Kevin
>>
>>  ------------------------------
>> *From:* xen-devel-bounces@lists.xensource.com [mailto:
>> xen-devel-bounces@lists.xensource.com] *On Behalf Of *Dulloor
>> *Sent:* 2009年5月9日 17:12
>> *To:* Keir Fraser
>> *Cc:* Ian Pratt; Xen-Devel (E-mail); Dan Magenheimer
>> *Subject:* Re: [Xen-devel] Re: [PATCH] add long interrupt measurement
>> capability
>>
>> Here is a xentrace patch.
>>
>> - Should irq_desc_measure_t move to some .h file ?
>> - I have defined the new trace event in general class. Is it fine ?
>>
>> - I have defined tsc_in as volatile to avoid initializing it in the
main
>> code path.
>>
>> thanks
>> dulloor
>>
>> On Sat, May 9, 2009 at 3:56 AM, Keir Fraser
<keir.fraser@eu.citrix.com>wrote:
>>
>>> On 08/05/2009 22:53, "Ian Pratt"
<Ian.Pratt@eu.citrix.com> wrote:
>>>
>>> >> Perhaps.  However measuring cycles is important and, more
>>> >> specifically, measuring MAX cycles spent across a set of
>>> >> interrupts.  As a result, I suspect any code that measures
>>> >> this (regardless of whether the result is reported by
>>> >> xentrace or debug-key) would likely encounter the
>>> >> same objection from Keir.
>>> >
>>> > I can''t imagine there''d be any objection to
adding trace macros to
>>> record
>>> > this. The xentrace log processing tool can then be updated to
generate
>>> max or
>>> > histogram values.
>>>
>>> Yes, xentrace records would be okay. It''s adding another
debug key and
>>> printing to Xen console for this purpose which I do not think is
>>> worthwhile.
>>>
>>>  -- Keir
>>>
>>>
>>>
>>
>

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

I applied it earlier today.


On 18/05/2009 21:41, "Dulloor" <dulloor@gmail.com> wrote:
> Hi Keir,
> 
> Can you consider this for check-in.
> 
> -dulloor
> 
> 2009/5/11 Dulloor <dulloor@gmail.com>
>> - Changed the name to trace_irq :)
>> 
>> - trace_irq dumps just tsc_in and tsc_out values in a single record. I
guess
>> there is no need to write two records (wasting trace-buf mem, more
>> processing, additional logic in xentrace and/or xentrace_format).
>> 
>> - xentrace_format does what Dan wanted.
>> 
>> -dulloor
>> 
>> 2009/5/11 Tian, Kevin <kevin.tian@intel.com>
>> 
>>> I guess you can handle it much simpler by stamping a record in both
>>> irq_enter and irq_exit. All the statistics jobs are left for
xentrace script
>>> to digest.
>>>  
>>> Also I''d call it as trace_irq instead of trace_guest_irq.
:-)
>>>  
>>> Thanks,
>>> Kevin
>>> 
>>>>  
>>>>  
>>>> 
>>>>  From: xen-devel-bounces@lists.xensource.com
>>>> [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of 
Dulloor
>>>> Sent: 2009年5月9日 17:12
>>>> To: Keir  Fraser
>>>> Cc: Ian Pratt; Xen-Devel (E-mail); Dan  Magenheimer
>>>> 
>>>> Subject: Re: [Xen-devel] Re: [PATCH] add long interrupt 
measurement
>>>> capability
>>>> 
>>>>  
>>>> Here is a xentrace patch.
>>>> 
>>>> - Should irq_desc_measure_t move  to some .h file ?
>>>> - I have defined the new trace event in general class. Is  it
fine ?
>>>> 
>>>> - I have defined tsc_in as volatile to avoid initializing it 
in the main
>>>> code path.
>>>> 
>>>> thanks
>>>> dulloor
>>>> 
>>>>  
>>>> On Sat, May 9, 2009 at 3:56 AM, Keir Fraser
<keir.fraser@eu.citrix.com>
>>>> wrote:
>>>>  
>>>>>  
>>>>> On 08/05/2009 22:53, "Ian Pratt"
<Ian.Pratt@eu.citrix.com> wrote:
>>>>> 
>>>>>>>  Perhaps.  However measuring cycles is important
and, more
>>>>>>>  specifically, measuring MAX cycles spent across a
set of
>>>>>>>  interrupts.  As a result, I suspect any code that
measures
>>>>>>>  this (regardless of whether the result is reported
by
>>>>>>> xentrace  or debug-key) would likely encounter the
>>>>>>> same objection from  Keir.
>>>>>> 
>>>>>> I can''t imagine there''d be any
objection to adding  trace macros to
>>>>>> record
>>>>>> this. The xentrace log processing tool can  then be
updated to generate
>>>>>> max or
>>>>>> histogram  values.
>>>>> 
>>>>> Yes, xentrace records would be okay. It''s adding 
another debug key and
>>>>> printing to Xen console for this purpose which I do  not
think is
>>>>> worthwhile.
>>>>> 
>>>>>  --  Keir
>>>>> 
>>>>> 
>>>> 
>> 
> 
> 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel