Re: [perfmon2] perfmon2 in 2.6.28-rc6 is broken on Power6

[Corey A. <cja...@li...>]

Hi Stephane and Paul,

After comparing sources, I didn't find any remarkable differences 
between the two (bare 2.6.28-rc6 and Stephane's perfmon2-enhanced 
2.6.28-rc6)

It didn't immediately occur to me that I could hit SysRq when the system 
hung, so I decided to give that a try, and I did get a stack trace, showing:
pc = ._mcount + 0x1c
lr = .pfm_handle_work + 0x30

The stack trace showed that rc.sysinit called into the kernel for some 
service (I'm not sure what service at this point), and on its way out 
again, it made a call to pfm_handle_work, which then calls _mcount. 
Looking at the source, pfm_handle_work doesn't call _mcount explicitly.

It appears that this _mcount function is used for kernel function 
tracing, and by default (via arch/powerpc/pseriesdefconfig) the 
CONFIG_FUNCTION_TRACER feature is turned on.  It also appears that this 
is an area of code that's changed between fairly recent SLES kernels 
(beta), and 2.6.28-rc6, because the CONFIG variables are named differently.

As an experiment, I disabled CONFIG_FUNCTION_TRACE, re-built the kernel, 
and it booted successfully.

So there's something in pfm_handle_work that is confusing the function 
trace code, at least on Power.  Paul, do you know what might be causing 
this problem?

In any case, if you need a short term work-around, this appears to be 
the way to do it - turn off CONFIG_FUNCTION_TRACER.


Regards,

- Corey

Corey Ashford
Software Engineer
IBM Linux Technology Center, Linux Toolchain
Beaverton, OR
503-578-3507
cja...@us...


Corey Ashford wrote:
> I want to report that perfmon2 in 2.6.28-rc6 is broken somehow on 
> Power6, and possibly all Power-based machines.
> 
> Today, I tried booting Stephane's 2.6.28-rc6 perfmon2 kernel with 
> perfmon2 enabled, and it hangs during the boot-up process right after 
> outputting the line "INIT: version 2.86 booting"
> 
> Without perfmon2 enabled, it boots up fine.
> 
> So something has regressed, and I'm not sure what.  I will see if I can 
> find the culprit and submit a patch.
> 
> Regards,
> 
> - Corey
> 
> Corey Ashford
> Software Engineer
> IBM Linux Technology Center, Linux Toolchain
> Beaverton, OR
> 503-578-3507
> cja...@us...
> 
> 
> 
> 
> ------------------------------------------------------------------------------
> SF.Net email is Sponsored by MIX09, March 18-20, 2009 in Las Vegas, Nevada.
> The future of the web can't happen without you.  Join us at MIX09 to help
> pave the way to the Next Web now. Learn more and register at
> http://ad.doubleclick.net/clk;208669438;13503038;i?http://2009.visitmix.com/
> _______________________________________________
> perfmon2-devel mailing list
> per...@li...
> https://lists.sourceforge.net/lists/listinfo/perfmon2-devel