[Bug runtime/12947] New: Use of hardware watchpoint causing oops

[Bug runtime/12947] New: Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

           Summary: Use of hardware watchpoint causing oops
           Product: systemtap
           Version: unspecified
            Status: NEW
          Severity: normal
          Priority: P2
         Component: runtime
        AssignedTo: systemtap@sourceware.org
        ReportedBy: wcohen@redhat.com


Created attachment 5825
  --> http://sourceware.org/bugzilla/attachment.cgi?id=5825
Serial console of the run with the oops

I was experimenting to the hardware watchpoints.

I had one script that would give me an address of the place going to probe and
then had another another script to initialize the hardware watchpoint and
record backtraces when the memory location was accessed.

Find the address with:

[wcohen@montague tmp]$ stap -g find_free_ram_addr.stp
&global_page_state(NR_FREE_PAGES) = 0xffffffffc0bb0eb0

Then set monitoring with:

[wcohen@montague tmp]$ stap -g track_free_ram.stp 0xc0bb0eb0

Attached is log of the run.

This was using

kernel-2.6.38.8-32.fc15.i686
systemtap-1.5-1.fc15.i686

It seems like that the hw watch points should be better behaved.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12947] Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

--- Comment #1 from William Cohen <wcohen at redhat dot com> 2011-06-28 21:04:00 UTC ---
Created attachment 5826
  --> http://sourceware.org/bugzilla/attachment.cgi?id=5826
Script to find address of element to watch

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12947] Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

--- Comment #2 from William Cohen <wcohen at redhat dot com> 2011-06-28 21:04:44 UTC ---
Created attachment 5827
  --> http://sourceware.org/bugzilla/attachment.cgi?id=5827
Script to watch changes to the amount of free memory

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12947] Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

--- Comment #3 from William Cohen <wcohen at redhat dot com> 2011-06-29 15:56:54 UTC ---
Created attachment 5830
  --> http://sourceware.org/bugzilla/attachment.cgi?id=5830
Ensure that the rc code is set when attempting to register a hw_breakpoint

The setup code for the hw_breakpoint doesn't check the return code. The problem
occurs later when there is an attempt to unregister the invalid hw_breakpoint.
This makes sure that rc gets propogated to minize the possibility of later
problems.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12947] Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

--- Comment #4 from William Cohen <wcohen at redhat dot com> 2011-07-05 15:46:31 UTC ---
The 2011-06-29 15:56 UTC patch allows systemtap to exit more gracefully when
unable to register the hw_breakpoint probe. Need to be root when using the
hw_breakpoint.

However, there still issues when systemtap attempts to use the hw_breakpoints.
Will attach log of back trace in a moment.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12947] Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

--- Comment #5 from William Cohen <wcohen at redhat dot com> 2011-07-05 16:00:26 UTC ---
Created attachment 5834
  --> http://sourceware.org/bugzilla/attachment.cgi?id=5834
oops when using hw_breakpoint

The output was generated with

/home/wcohen/systemtap_write/install/bin/stap -v hwkbrpt_sample.stp pid_max
Pass 1: parsed user script and 76 library script(s) using
25000virt/13840res/2568shr kb, in 410usr/30sys/927real ms.
Pass 2: analyzed script: 1 probe(s), 1 function(s), 2 embed(s), 0 global(s)
using 25276virt/14428res/2836shr kb, in 20usr/0sys/85real ms.
Pass 3: using cached
/root/.systemtap/cache/4a/stap_4a5b6a0564c3a5f41b9730f6cd92e255_1130.c
Pass 4: using cached
/root/.systemtap/cache/4a/stap_4a5b6a0564c3a5f41b9730f6cd92e255_1130.ko
Pass 5: starting run.
WARNING: Hwbkpt probe kernel.data("pid_max").length(1).rw: registration error
-902930368, addr   (null), name pid_max
ERROR: probe kernel.data("pid_max").length(1).rw registration error (rc
-902930368)
Pass 5: run completed in 10usr/60sys/516real ms.
Pass 5: run failed.  Try again with another '--vp 00001' option.

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.

[Bug runtime/12947] Use of hardware watchpoint causing oops

[wcohen at redhat dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=12947

William Cohen <wcohen at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #6 from William Cohen <wcohen at redhat dot com> 2011-07-27 18:22:47 UTC ---
Fixed in:

43650b10c3f42dedb0d83ce7ecff4957be49ef90
217ef1f405ba7d425c8f5ab3ca730bfa4a395c75

Tests checked in:

feb9187ba645c198e2e177080b4563080c0ee69e
73017154bbe6f1d6da5ff23c5c6c03324799c40d

Note that the tests need to run as root to function

-- 
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.