RE: 03-23-2006 Meeting minutes

RE: 03-23-2006 Meeting minutes

[Stone, Joshua I]

Hien Nguyen wrote:
> Xen support -
> Elena indicated that Xen is now available on FC5, we should test
> systemtap on Xen (both DOM0 and DOMu).

I tried this on x86_64 - and immediately ran into a problem with the
debuginfo.  Observe:

jistone@jistone-xen64 /usr/lib/debug/lib/modules
$ ll */vmlinux
-rwxr-xr-x 1 root root  42M Mar 14 13:49 2.6.15-1.2054_FC5kdump/vmlinux
-rwxr-xr-x 1 root root  44M Mar 14 13:00 2.6.15-1.2054_FC5/vmlinux
-rwxr-xr-x 1 root root 7.2M Mar 14 13:15 2.6.15-1.2054_FC5xen0/vmlinux
-rwxr-xr-x 1 root root 6.3M Mar 14 13:22 2.6.15-1.2054_FC5xenU/vmlinux


The sizes for the xen* kernels are way too small to contain all of the
debug information.  And indeed, when I run stap, I'm told "cannot find
module kernel debuginfo: No DWARF information found".


Josh

RE: 03-23-2006 Meeting minutes

[Roland McGrath]

> I tried this on x86_64 - and immediately ran into a problem with the
> debuginfo.

This was already noticed and is some problem in the kernel rpm specs.
There may be an fc5 bug report for it already on bugzilla.redhat.com,
I'm not sure.

Re: 03-23-2006 Meeting minutes

[William Cohen]

Roland McGrath wrote:
>>I tried this on x86_64 - and immediately ran into a problem with the
>>debuginfo.
> 
> 
> This was already noticed and is some problem in the kernel rpm specs.
> There may be an fc5 bug report for it already on bugzilla.redhat.com,
> I'm not sure.

Xen x86_64 kernels in kernel debuginfo package do not contain dwarf symbols

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=185662

-Will

03-23-2006 Meeting minutes

[Hien Nguyen]

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

Userspace probes -
Prasanna posted user space probes patch on LKML. Andrew wanted a details 
description of the design, and advantage/disadvantage of this approach.
Someone suggested to do everything from the user space via a system call 
(similar to ptrace).
Prasanna raised some questions that we need to have the answers for
- Considering for global probes vs local probes
- Should the handler run in the interrupt context?
- Can and should the handler run in the userspace?
(Prasanna please correct my notes )
Even with the system call interface we still be able to use lot of 
current code.
Sometime ago, Roland mentionned that he will improve ptrace interface, 
we probably should talk with Roland on kprobes ptrace marriage (did I 
get it right?), Will volunteer to check with Roland on that. However, 
Richard thinks it is a bad idea.
Prasanna will write up the requirement and the design and post on LKML 
again next week, let's discuss all issues on LKML.
Systemtap core team should be aware of the userspace probes discussion 
on LKML.

LKLM watch -
Hitachi will watch LKML next week.

elfutils - No new news

Systemtap GUI -
Hien will lead a team of three intern students to work on the systemtap 
gui. It is an open source project based on Eclipse. The project 
discussion will go on systemtap mailing list. IBM will introduce the 
SpeedTeam (interns) to the systemtap team when they arrive in April.

Face to face meeting -
Richard would like to meet everyone in his up comming trip to USA. Elena 
is willing to host in the following location
WestFord, MA
Toronto Canada
Mountain view, CA
somewhere? North Carolina
Intel and IBM could host the meeting too (Beaverton for IBM, Mountain 
view for Intel). For India folks to attend the meeting in May, Toronto 
is the best place (due to Visa issue)
Elena will finalize the plan with folks offline.

Redhat support for systems running systemtap -
Currently unverified kernel modules are not supported by Redhat. Elena 
will find out who can make that decision to change that for systemtap 
generated modules.

Xen support -
Elena indicated that Xen is now available on FC5, we should test 
systemtap on Xen (both DOM0 and DOMu).

RHEL4 U4 Back port -
Anil said there is some concerns at Redhat to get the back port patch 
into the RHEL4 U4 kernel. Elena will take care of that.

RHEL5 -
Userspace probes may not make RHEL5. Elena indicated that if the 
userspace patch could get into the mm tree that would be good.

Performance monitor tapset -
Will will rewrite the design base on the feed back he got and post again.
There were much discussion about starting and stopping counters. 
Discussion will continue on the mailing list.

Binary tracing -
Tom posted a proposed implimentation and a prototype on the bz#2046. 
Frank said the discussion should move to the systemtap mailing list.

Systemtap to allow raw access to relayfs channels -
There is a systemtap's customer wants to use systemtap to collect the 
data via relayfs without stpd draining the data. Jim will post a write 
up on this.

Exception Handling -
Anil indicated that currently if people do the usercopy in the kprobe 
handler (in the case of plain kprobe module) they will get the warning. 
Systemtap does not have that issue since it uses its own copyuser in the 
runtime.

Anil announced that the re-entry patch for x86_64 has been accepted.
Ananth - Kprobes Booster patch is on its way to Linus.