[GSoC] Making vkernels checkpointable - final report

[Justin Sherrill]

This is open source, so you can pretty much work on whatever schedule you
want.

Something that might be useful for the future, if you wanted to work on it:
a specific test that can be run to make sure vkernel checkpoint works.
 Given that checkpointing is complicated, and vkernels are going through
some changes, it is useful to have a tool that could regularly check to
make sure this hasn't been broken by other work.  This would be especially
useful near release times.


On Tue, Sep 24, 2013 at 9:17 PM, Pawel Dziepak <pdziepak at quarnos.org> wrote:

> Hello,
>
> During the last week of this year Google Summer of Code I continued
> testing the code, including more tests of 32 bit versions of both
> vkernel and the "host" kernel (during the GSoC I used mostly 64 bit
> versions). Fortunately, I didn't came across any serious problems.
> Moreover, I updated the sys_checkpoint(2) manpage removing information
> about the limitations that are no longer present and adding
> implementation notes that explain the format of the checkpoint files.
>
> The goal of my project is completed. It is possible now to chekpoint a
> running vkernel together with its virtual processes and then restore
> it. I tried to make it possible by enhancing checkpointing
> implementation rather than doing everything inside vkernel checkpoint
> signal handler so that not only vkernels can benefit from these
> modifications. Apart from some minor fixes, making checkpoint file
> format more flexible and future-proof, now all MAP_VPAGETABLE mappings
> are properly supported as well as kqueues and threaded programs.
>
> I've checked and there are no merge conflicts with the master. There
> also doesn't seem to be any problems with the kernels (and vkernels)
> built after merge, but undoubtedly vkernels taking advantage from
> visualizing extensions cannot be currently checkpointed. There is also
> matter of dirfs which was recently added. I haven't really looked into
> it yet but it seems like it will also require some work to get it
> working with checkpointed vkernels.
>
> I am willing to continue working on the checkpointing code and the
> vkernels and make it work with these recent additions hardware
> virtualization and dirfs. Due to previous arrangements in October I
> would have much less time for it that I'd wish to have but still I
> expect to make some progress.
>
> Paweł
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.dragonflybsd.org/pipermail/kernel/attachments/20130925/c07bf5c0/attachment-0003.htm>