Re: [Qemu-devel] Want to be part of Memory Hotplug

[li guang]

在 2013-02-07四的 18:54 +0100，Vasilis Liaskovitis写道：
> Hi Senthil,
> 
> On Mon, Feb 04, 2013 at 04:17:30PM +0530, kumaran wrote:
> > Hi,
> > 
> > I am Senthil, doing post graduation in IIT Bombay,India.
> > 
> > I am looking for some problems related to KVM as part of my research
> > work. I read about memory hotplug and other open issues in KVM.
> > 
> > Can i be part of your work regarding memory hotplug.
> > 
> > My Experience in KVM:
> > 
> > - I have implemented Record and Replay feature in KVM
> > - I have good hands on experience with KVM and QEMU's code base
> > - Thorough knowledge about QEMU's architecture and PCI emulation
> > - I have implemented Record and replay for Intel Ee100Pro network
> > interface and for IDE disks (including DMA)
> > 
> > If I get chance, I can spend around 5 dedicated months on this work.
> > Please let me know your interest.
> 
> thanks for your interest in the memory hotplug effort. You are welcome to
> contribute. Memory hotplug is still not in mainline qemu. It is a qemu-wide
> project rather than a kvm-specific one.
> Some possible directions are:
> 
> - reviewing / improvement suggestions on v4 here:
> http://lists.nongnu.org/archive/html/qemu-devel/2012-12/msg02693.html
> 
> - define the qom interface for Dimms. The current patches use a memory bus
> abstraction (DimmBus) where Dimms can plugin. That's definitely one choice. 
> Some
> people have suggested using <links> from the i440fx/mch (in general the memory
> controller device of the emulated system) to the Dimm devices. That's similar
> to how CPUs are re-modelled in qom-cpu patchsets I think. The final dimm
> interface has not been agreed upon yet.
> 
> - Add acpi native memory hotplug support for q35/ich9. The current code 
> creates
> paravirtual Dimmbus + Dimms, without emulating real hotplug-capable memory
> controller hardware.
> 
> - Make sure ejection is safe i.e. all users of a hotplugged MemoryRegion 
> (not only guest/CPUs but also qemu block layer) have stopped using the memory,
> before actually freeing it.

when can be considered 'have stopped using the memory' for
'MemoryRegion'?
what I can see is only the devices used this 'MemoryRegion'
decided to destroy memory region.

> 
> make sure to include the qemu list (cc'ed) in order to let everyone know what 
> you
> are working on and to get feedback.
> 
> thanks,
> 
> - Vasilis
>