[Ksummit-2013-discuss] [ATTEND] CPU hotplug rework

Srivatsa S. Bhat srivatsa.bhat at linux.vnet.ibm.com
Thu Jul 18 11:44:03 UTC 2013


Hi,

I have been contributing to the CPU hotplug infrastructure in the Linux kernel
and to a certain extent I've even helped maintain it, spanning both the core
and the peripheral parts (where it cuts across multiple subsystems).

I have been involved in the redesign of CPU hotplug and had proposed and
posted patches that implement several enhancements such as stop-machine()
removal, consolidation of cpu hotplug code from various architectures into
the core code and so on. There have been several great enhancements from
Thomas Gleixner as well, such as the kthread park/unpark infrastructure,
consolidation of cpu idle/hotplug code, and the rework of the notifiers used
in CPU hotplug (multiple stages of online/offline).

While there have been plenty of active discussions around these designs and
many of them have gone upstream, some of the more complex ones are yet to gain
a fuller consensus, and hence will benefit from discussions at the summit.
The topics for discussions would range from core locking schemes and algorithms
(flexible per-cpu rwlocks, RCU and other schemes), to aspects such as
designing the solutions in such a way that they can be easily adopted to
-rt environments as well.

I would like to discuss some of these topics around CPU hotplug rework at the
summit, to see how best to take it forward and also to discuss other possible
improvements in the future.

The people involved are: Thomas Gleixner, Peter Zijlstra, Ingo Molnar,
Paul McKenney, Tejun Heo, Oleg Nesterov, Steven Rostedt, Rusty Russell,
Michel Lespinasse, Lai Jiangshan and others.

On a related note, I'm also interested in taking part in the discussion around
Parallel CPU boot that Peter P Waskiewicz Jr had proposed in the other thread
earlier, as it also involves CPU hotplug related considerations, IIUC.


Thank you very much!


Links to some relevant patchsets/discussions:

1. Stop_machine()-free CPU hotplug:
----------------------------------

  * Performance difference between CPU Hotplug with and without stop_machine():
    http://article.gmane.org/gmane.linux.kernel/1435249

  * Links to patchsets/discussions around alternative synchronization schemes
    to replace stop_machine() in the CPU Hotplug code:

    v6: http://lwn.net/Articles/538819/

  * Patchset intended to upstream stop-machine removal in parts:

     v3: http://lwn.net/Articles/556727/


Regards,
Srivatsa S. Bhat



More information about the Ksummit-2013-discuss mailing list