RE: meeting minutes 20060504

RE: meeting minutes 20060504

[Stone, Joshua I]

On Thursday, May 04, 2006 2:09 PM, Hien Nguyen wrote:
> BZ #2639 - Probe on "__kprobes" function ... it appears to be closed,
> Jim will follow up this up.

I just opened this yesterday, and no-one else has commented on it, so
how is it closed?  There was #2506, which mentioned __exit and
__kprobes.  I found a solution for __exit, but probes on __kprobes have
more issues, so I split it into a different bug (2639).

> OLS paper - Gray Brian ?? withdrew his OLS paper (Instrumentation
> using systemtap) due to some concerns with the data that he got
> (kprobes related issues)

Ray Bryant.  And I would still like more details about what he did,
especially if it means dealing with the concerns that he had.  If we
don't leverage all information, I'm likely to have the same problems
without realizing it.


Josh

RE: meeting minutes 20060504

[Hien Nguyen]

On Thu, 2006-05-04 at 14:45 -0700, Stone, Joshua I wrote:
> On Thursday, May 04, 2006 2:09 PM, Hien Nguyen wrote:
> > BZ #2639 - Probe on "__kprobes" function ... it appears to be closed,
> > Jim will follow up this up.
> 
> I just opened this yesterday, and no-one else has commented on it, so
> how is it closed?  
My mistake. BZ#2639 is NEW.
> There was #2506, which mentioned __exit and
> __kprobes.  I found a solution for __exit, but probes on __kprobes have
> more issues, so I split it into a different bug (2639).
> 

RE: meeting minutes 20060504

[Richard J Moore]

systemtap-owner@sourceware.org wrote on 04/05/2006 22:45:30:

> On Thursday, May 04, 2006 2:09 PM, Hien Nguyen wrote:
> > BZ #2639 - Probe on "__kprobes" function ... it appears to be closed,
> > Jim will follow up this up.
>
> I just opened this yesterday, and no-one else has commented on it, so
> how is it closed?  There was #2506, which mentioned __exit and
> __kprobes.  I found a solution for __exit, but probes on __kprobes have
> more issues, so I split it into a different bug (2639).
>
> > OLS paper - Gray Brian ?? withdrew his OLS paper (Instrumentation
> > using systemtap) due to some concerns with the data that he got
> > (kprobes related issues)
>
> Ray Bryant.  And I would still like more details about what he did,
> especially if it means dealing with the concerns that he had.  If we
> don't leverage all information, I'm likely to have the same problems
> without realizing it.
>

Ray simply told me he couldn't reproduce his results. When I asked him
whether that was due to the instrumentation method he said the problem was
more fundamental. I assume he meant he was looking for a problem that
didn't exist.

As for the instrumentation: that was achieved by putting a few probes in
the VM code - not via systemtap, but via handcrafted kprobes modules.

Richard

>
> Josh

meeting minutes 20060504

[Hien Nguyen]

kprobes portion
---------------
Attendees -
IBM - Jim, Hien, Ryan, Jeff, Henry, Prasanna, Richard
Intel - Anil
Hitachi - Satoshi

Userspace probes -
Prasanna posted the patch on the internal mailing list for review, he
addressed most concerns. The current implementation is not different
from the original implementation.
Jim had userspace return probes working on i386.

BZ #2637 skipped probes on FC6 - Prasanna will investigate/work on the
bug.
BZ #2639 - Probe on "__kprobes" function ... it appears to be closed,
Jim will follow up this up.

Nofity pagefault patch -
Anil proposed to backpor this patch to RHEL4U4. This patch sat on mm
tree for a while without any objection, the chance it gets into the
mainline is high.

OLS paper - Gray Brian ?? withdrew his OLS paper (Instrumentation using
systemtap) due to some concerns with the data that he got (kprobes
related issues)

systemtap portion
-----------------
Attendees - All of the above and ..
IBM - Vara
Redhat - Martin, Will, David, Elena
Intel - Josh, Thang

RH summit: Frank will travel May 10/11/12.
Summit itself May 30 through June 2.

Kernel bug with page fault:
	BZ# 2637 see above
	Notify pagefault patch - see above

roundtable
Martin - worked on syscalls tapset tests.
Will - added probes profiling in the translator to measure probe
handlers performance (option -t) also update the man page. Skipped the
detail discussion about presentation that he attended two week ago :-)
David - talked about filtering stpd as a default mode.

User space probes: 
See above.

f2f:
Several options
1. All day conference calls
2. Meet at OLS
3. Meet in UK in the falls.