Re: [PATCH] GSoC 2010 - Memory hotplug support for Xen guests - third fully working version

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Wed, 25 Aug 2010 14:33:06 -0700
Jeremy Fitzhardinge <jeremy@goop.org> wrote:

> >> 2 requires a deeper understanding of the existing hotplug code.  It
> >> needs to be refactored so that you can use the core hotplug machinery
> >> without enabling the sysfs page-onlining mechanism, while still leaving
> >> it available for physical hotplug.  In the short term, having a boolean
> >> to disable the onlining mechanism is probably the pragmatic solution, so
> >> the balloon code can simply disable it.
> > I think that sysfs should stay intact because it contains some
> > useful information for admins. We should reconsider avaibilty
> > of /sys/devices/system/memory/probe. In physical systems it
> > is available however usage without real hotplug support
> > lead to big crash. I am not sure we should disable probe in Xen.
> > Maybe it is better to stay in sync with standard behavior.
> > Second solution is to prepare an interface (kernel option
> > or only some enable/disable functions) which give possibilty
> > to enable/disable probe interface when it is required.
> 
> My understanding is that on systems with real physical hotplug memory,
> the process is:
> 
>    1. you insert/enable a DIMM or whatever to make the memory
>       electrically active
>    2. the kernel notices this and generates a udev event
>    3. a usermode script sees this and, according to whatever policy it
>       wants to implement, choose to online the memory at some point
> 
> I'm concerned that if we partially implement this but leave "online" as
> a timebomb then existing installs with hotplug scripts in place may poke
> at it - thinking they're dealing with physical hotplug - and cause problems.
> 

IIUC, IBM guys, using LPAR?, does memory hotplug on VM.

The operation is.
	1. tell the region of memory to be added to a userland daemon.
	2. The daemon write 0xXXXXXX > /sys/devices/system/memory/probe
	   (This notifies that memory is added physically.)
	   Here, memory is created.
	3. Then, online memory.

I think VM guys can use similar method rather than simulating phyiscal hotplug.
Then, you don't have to worry about udev etc...
No ?

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>