[ofw] Reference count debugging mechanism proposal

Tzachi Dar tzachid at mellanox.co.il
Tue Dec 7 21:42:24 PST 2010


Hi fab,

The scenarios that we are facing are trying to capture licks that take a very long time to happen. This means that we need to have the listing of wpp all the time (from boot) and for very long time. We also need to have the captures for different catagories.

>From what we saw in the past WPP was almost impossible to get working, very hard to connect to a kernel debugger, and not very reliable if to say the list.

More than that wpp will also have higher influence on performance.

Thanks
Tzachi

From: ofw-bounces at lists.openfabrics.org [mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of Fab Tillier
Sent: Tuesday, December 07, 2010 11:35 PM
To: Uri Habusha; Irena Gannon; 'ofw at lists.openfabrics.org'
Subject: Re: [ofw] Reference count debugging mechanism proposal


-          There's nothing to prevent WPP from being run for debug builds too.  That's an OpenFabrics decision, not a WPP limitation.

-          You can have ETW events log to the debugger.  A WPP trace session is managed explicitly, so I don't understand what you mean by your second point.

-          Filtering data, if you structure your trace messages in a logical way, should be trivial.

-          Right, you need to write more new code to filter the data, code that isn't necessarily portable to things that don't use cl_obj or al_obj or whatever_obj you are using at the time.  Using WPP tracing is generic, independent of any underlying code structure.


From: Uri Habusha [mailto:urih at mellanox.co.il]
Sent: Tuesday, December 07, 2010 4:23 AM
To: Fab Tillier; Irena Gannon; 'ofw at lists.openfabrics.org'
Subject: RE: [ofw] Reference count debugging mechanism proposal

 Because of following reasons:


*         WPP is only run on free not in debug.

*         WPP should explicit to be operate. that means when we need it, it will not be.

*         Using WPP is more complicate. You need to take the binaries and format it. Afterward filtering of specific data is complicated.

*         Using internal data allows us to right a debugger extension and filter the data easily


From: ofw-bounces at lists.openfabrics.org [mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of Fab Tillier
Sent: Tuesday, December 07, 2010 1:48 AM
To: Irena Gannon; 'ofw at lists.openfabrics.org'
Subject: Re: [ofw] Reference count debugging mechanism proposal

Why not just use WPP/ETW tracing for this, rather than re-implement the logging?  The WPP/ETW tracing has the advantage that the logs could be gathered in one place and analyzed in another.  You can trace all the same information as you propose.

-Fab

From: ofw-bounces at lists.openfabrics.org [mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of Irena Gannon
Sent: Monday, November 29, 2010 8:46 AM
To: 'ofw at lists.openfabrics.org'
Subject: [ofw] Reference count debugging mechanism proposal

Hello,

Please review the attached proposal for adding debugging information to the reference count mechanism.

Thank you,
Irena
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20101208/798b22b7/attachment.html>


More information about the ofw mailing list