Re: [GSoC] Make vkernels checkpointable - week 2

[Pawel Dziepak <pdziepak@xxxxxxxxxxx>]

2013/7/7 Samuel J. Greear <sjg@evilcode.net>:
> How did everything go this last week?

I was starting to work on network interface restore. I've created a
signal handler for SIGCKPT that will take care for all additional
actions that vkernel needs to take in case of a freeze/thaw (and for
now also will enable me to test whether everything is working
correctly). I also had to create a temporary support for
MAP_VPAGETABLE mappings since it is necessary to get at least signal
handler continue without any errors after checkpoint restore.
Restoring MAP_VPAGETABLE mappings is currently just a hack with many
hardcoded values, but will do for testing purposes until August when,
as planned, I will implement the proper and full support of this type
of mappings. I will push recent changes tomorrow, sorry for the delay.

> Would it make sense to look at going ahead and committing the existing work
> you have done to master? (Since if it can save/restore a regular threaded
> process, it is independently useful).

I think it will be better to wait with merging saving/restoring of
threaded processes. The number ELF notes in checkpoint file and
ckpt_siginfo structure have changed what breaks compatibility. Since I
will need to make more changes to the format of checkpoint files
during this GSoC I do believe the best solution is to wait until all
its done. I particularly don't like the way number of threads is
computed which despite being correct is not flexible at all. It
probably may be a good idea to put more of the additional data that is
stored in these files into ELF notes. However, since it is a bit out
of scope of my project I will wait with this until the main goal is
completed.

Pawe³