02-09-2006 meeting minutes

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

Attendees
IBM - Jim, Hien, Vara, Richard, Prasanna, Maneesh, Ananth
Redhat - Will, Martin, Frank
Intel - Anil, Josh
Hitachi - Satoshi

LKML watch -
Redhat is next to watch LKML. Hien will post summary.

elfutils -
No new news. Elena is on PTO.

User space probes -
Jim suggested to stay with Prasanna's approach since there is too much 
objection on the mailing list. Prassana had posted take2 of the user 
probes patches.
Richard indicated that we should not allow interrupt during single step 
out of line or inline (for stack base architecture). Try to disable 
interrupt just for a small window (as small as possible) when single step.

kprobes fault handler -
Prasanna will work on the patch. Anil would like to include this patch 
in the RHEL4 U4 back port set.
We need to delay the kprobe notification registering until some one 
register a probes (other than the inital trampoline probe). We may need 
to consider a way to unregister kprobes notification ....

The discussion moved to disable kprobe completely (not even compiled in) 
when CONFIG_MODULEs= disable

kprobe vs debugger -
The general consensus in the group is to not allow inserting probe if 
other apps already put a breakpoint at that same location. Propose a 
change to ptrace to do the same (not allow setting breakpoint if someone 
else already did). Unpredictable results will happen sometime, but it is 
not terribly bad (document what would happen and why)
Prasanna will work on this and roll them into the future user probe patches.

kprobe test-
Test cases for all bug fixes should be checked in to CVS tests/kernel by 
the end of Feb.

4g/4g -
Satoshi indicated that Redhat supports 4g/4g  kernel, what about 
kprobes/ user probes. Kprobe/user space probe does not support 4g/4g is 
an option, but there are customers currently using it.
We need to understand what are the issues. Perhaps the 4g/4g people 
should look at our current kprobe/user probes implementation and give us 
their opinions.
Ananth said 4g/4g is not in the mainline,  therefore it will be harder 
to user probes support 4g/4g to go to the mainline.

kprobe-booster -
Satoshi announced that kprobe-booster is now in the mm tree. according 
to Satoshi, there will a port for x86_64 and ia64. What's about ppc64?