kprobes problem

[Baruch Even]

Roland McGrath wrote:
> It sounds like there may be a generic bug with "ret" instructions.  These
> are handled specially.  It would be best if you could construct a test
> (e.g. write a test module that inserts kprobes on test code you provide in
> that same module) where you can figure out exactly what happens when the
> probe should be executing the return.  That is, where you know exactly
> where you should have gone after that (perhaps you can do this by having
> another probe at the proper return location), and ideally figure out where
> you actually went instead (or if the problem is different from that
> e.g. the stack contents being wrong).

I fail to see why you'd need a special test case for this, the simple 
code that I use attaches a kprobe to a known kernel function (tcp_ack 
originally), you know exactly where you need to return to. Or you can 
find another function that has only one caller.

I'm sorry but I can't invest the time to debug this myself, kprobes is 
not my core work, just a tool I'm trying to use.

I hope that my bug report is sufficient to help reproduce and fix this bug.

Baruch