[kprobe bug] pre-handler changes on globals lost

[Ananth N Mavinakayanahalli]

Rusty Lynch wrote:
> While working with some kprobe unit test, I stumbled across some wacky
> behavior that the below module exposes.
> 
> If you build and install this module, you can run tc1 (the working case)
> by doing a "cat /proc/run_tc1".  You can then run the broken case by 
> doing a "cat /proc/run_tc2".
> 
> When the module is installed a kprobe with a pre_handler that increments
> a global counter is registered for a target function.
> 
> When the tescase is run it stores the current value of the global counter, 
> calls a simple function (the target function that has been armed with a 
> kprobe) that just returns the value of the global counter, and then the 
> testcase verifies that the returned value has incremented (thanks to the 
> kprobe.)
> 
> By adding a simple printk("peep\n"); to the target function, the effects
> of the pre_handler on the global counter are lost.
> 
> Can someone else verify they see this behavior?  I see the bug on both i386
> and x86_64 FC3 systems, but not on my ia64 REL4 system.

I don't see the bug on a RHEL4 ppc64 system.

Ananth