03-16-2006 meeting minutes

[Hien Nguyen <hien at us dot ibm dot com>]

Attendees
IBM – Ananth, Prasanna, Jim, Hien, Richard, Tom, Vara
Redhat – Will, Frank
Intel – Anil, Josh
Hitachi – Satoshi

User space probes –
Per Prasanna’s request, Anil and Jim is going to review the patch. 
Prasanna would like to post the patch early next week.
Prasanna will update the user space probes to-do list.

Regression test –
Call for everyone to check in their regression test.
Bugs –
#2453 – Ananth fixed the bug.
#2465 – Await Guang Lei to test Frank's suggestion.
#2452 – Assigned to Bibo, it’s fixed and should be closed.
#2185 – Anil will update the bz, assign it to himself and post the patch 
to the x86_64 mailing list.
#2183 – Assigned to Bibo, it’s fixed and should be closed.

OLS Paper –
The first draft should be done by Monday. Satoshi will provide some 
benchmarks and tests to demonstrate the performance improve by going 
from spin_lock to RCU implementation of kprobes.

Fault Handling –
Patches are posted on LKML. Prasanna will take Richard’s suggestion and 
try to improve the fault handling code.

Watchpoint probes –
Not implemented. It will be next after userspace probes.

LKML watch –
Intel is on next week (Josh)

Elfutils – No new news.

stap gcc warning issue –
Martin needs to fix it in the runtime.

Binary tracing-
Tom is closed to have a proposal, he is working on s prototype and some 
concrete examples. Discussion regarding binary tracing will continue on 
bz#2307

Performance monitor tapset –
Will got some comments from Frank and Josh and going to update the 
paper. The main issue is perfmon itself is no in the main line.
Benchmark framework –
Martin posted the benchmark framework yesterday. We can use this 
framework to find out where systemtap spends its time.
Some concerns on the compilation time, there are some suggestion to 
improve that
- Pre built library
- Pre built module (this is on Frank’s todo list)

Kprobes on Xen –
Vara indicated that IBM don’t have the resources to work on this.
Production ready?
Frank indicated that we need to beef up the test suite with test cases 
that puts probes all over the kernel such as function (“*”), then if the 
test suite passed, systemtap is ready for production. Black list any 
functions/places that causes kernel panic is OK.
Will volunteered to work on the test suite. Jim suggested put probes in 
kmalloc, switch_to, or interrupt service routines. Look at Josh 
stress-test.