[K42-discussion] question about K42 dynamic update feature

[Andrew Baumann]

Hi,

On Wednesday 07 February 2007 13:57, yanmiaobest wrote:
>      I have a new implementation of some object to replace the old one. The
> old one is useless to me, incluing its data and functions. After hot-swap,
> the old version will never be used, so how do I clean up the obsolete
> functions in order to save memory. It seems none of your paper mentions
> that.

You're right, we don't support that at the moment. The data associated with an 
object is always reclaimed, but the space used for code is not and leaks with 
successive updates.

There are two problems here. First, if the object was part of the original 
kernel image that was loaded, it is not easy to identify what part of the 
kernel's text segment it belongs to, as that is all linked together at build 
time. Second, if the object that has been replaced was already a dynamic 
update, it should be possible to reclaim its memory just by unloading the 
relevant module, but our module loader does not support this.

The second problem is solvable with a little bit more implementation work in 
the module loader. The first problem is trickier, but I think could be 
handled by stripping the number of objects in the base kernel down to the 
bare minimum needed to boot, and then inserting the rest as loadable modules 
(from a ram disk or similar) that could be removed again later as needed.

We haven't done any of this, because I think it isn't as interesting 
research-wise, and other systems have already shown that you can build 
smaller kernels and load much of the system as modules. Obviously, if you 
were implementing a real system, you would need to address the memory leak, 
but we haven't found it to be a problem in our experiments so far.

Good luck with your work, and feel free to ask me if you have more questions.

Regards,
Andrew